It strikes me that if their models were good enough for recursive self improvement, then the VCs would have pushed to delay IPO until such time that they had clear market dominance.
For some reason they need the $100B today at the cost of dilution, rather than additional private rounds and/or using cashflow+profits to continue the advancement.
So we are at the point where RSI is happening, but the idea that RSI leads to exponentially fast growth and SI is based on a number of unproven factors, most importantly that intelligence is scalable significantly beyond human levels. We are also not sure of the limitations of LLMs, how it relates to our own intelligence, and how hard it will be to make the jump to other models.
But when it comes to Anthropic, I am almost certain that it does not make sense to argue based on the typical arguments on maximizing profits. Everything I know about the people working there suggests that they are not primarily motivated by money, but by their mission. I bet they would prefer a lower-strings attached private offering, but have no issue with diluting their own stock, if they think it will serve to move Anthropic further ahead on bringing about aligned AGI. Basically, the diluted money will just be missed donations to EA causes for the most part anyway.
When you have AI models autonomously self-improve, you implicitly lose your understanding of how they work and accordingly your supposed control of them?
AI can be a "tool", but only so long as you understand how it works and what the implications of its use are. Being an enormously powerful tool implies, the consequences of not understanding and not knowing are proportionally enormous as well.
Currently, LLMs are used practically exclusively by "professionals" who, more or less, know what they're doing. The idea of applying them "everywhere" completely misses the gap in understanding and responsibility for the vast majority.
He distinguishes between weak recursion (using sotfware to build more software), and strong recursion (handing over model development to AI which does everything), and makes a pretty strong case for the fact that only weak recursion is happening.
This maps to my experience with AI models, they reduce the cost of experimentation massively, but they can't (currently) design experiments themselves.
Even then, training RL is very data hungry, and most profitable tasks don't have good data so even if we got superintelligence (whatever that means), it's entirely possible that progress would still be bottlenecked by lack of high quality training data.
Even if that were true (i.e. the models were good enough for infinite recursive self-improvement), the risk is still rapidly increasing with time. I presume most VC's would rather take the current exit with their pants intact.
Kind of hard to believe in recursive self improvement, every time I tried setting an LLM to loop on itself to "recursively" improve code towards a goal it always descends into utter garbage. I think the main challenge is probably knowing exactly what you want to get, and then having a training landscape that is smooth enough so you can climb there? I guess these models can probably iterate their architecture to max out existing benchmarks but at some point they will essentially just do p-hacking as the model that improves the model will invariably leak the validation data into the training process on the meta level, or so I would assume. And if your goal is to create a model that can compete with human intelligence in math then how would you even come up with a scoring function for that, will you let the model do multiple years of fundamental math research and then see if it produces something that is novel and interesting? And how do you score if it's like 99 % there, 50 % or 10 %? From my naive understanding that's probably the main challenge, these models can only optimize if you have a clear measurement of goal achievement that can be climbed to, if there's only a very far fetced goal without clear intermediate success states I don't see how the current generation of models should get there.
Apple, Amazon, Intel, Google, NVIDIA and Tesla all had world-changing product before their IPOs but they didn't stay private until clear market dominance.
Because private rounds still dilute you.
And VCs tap out well before $100B.
And they'll need more than $100B to accelerate widespread economic diffusion.
And RSI can't hallucinate gigawatt power plants and TSMC fabs into existence.
> Apple, Amazon, Intel, and NVIDIA didn't stay private until clear market dominance either.
Apple, Amazon, Intel, and NVIDIA all went public in an entirely different world, before the JOBS act.
I'm not necessarily taking the same position as the person you are replying to, but using these companies' IPOs as comparisons for what Anthropic is doing is not exactly Apples to Apples as it were.
Why not? Win the stock market, break ECC and own the internet. There are many ways to make infinite money if you have infinite intelligence but not infrastructure. True RSI wouldn't announce itself
> It strikes me that if their models were good enough for recursive self improvement
I just can't imagine this is true when their overreach problem (doing more than you ask), hallucination problem (lying), redefinition problem (doing something completely orthogonal to what you asked), and harm problem (doing something that is exactly the opposite of what you asked) are all far too high at this point, and by most measures - all of the recent models are worse on most (or all) of these metrics.
They are giving up here in RLHF to benchmaxx the metrics people are looking at to pretend like they're making huge gains, when they're largely running in circles.
They are getting better at pass@1 at the expense of far worse p90s on real agentic tasks.
It makes zero sense. Even if the model can self-improve it doesn't mean the hardware need will go to zero immediately. Human society as a whole is a self-improving system (at least in terms of technology) and it doesn't mean humans consume less and less resources.
They need resources and $100B can buy them that. I don't think you need a conspiracy theory to explain that.
The whole idea of self-improving machine equals immediate singularity is just wrong and ridiculous. I don't know why so many people are attached to this idea. Perhaps just because "singularity" is such a cool sci-fi word?
They all knew about the stuff that is coming out now (Jensen stepping down due to chinese spy stuff) and has been out and debated for a while now (circular financing) and money drying up as a result. Winter is coming...
Comment from a perma bear. So take it with a grain of salt. lol
Exit liquidity? I feel like it would be a lot easier to raise $100b from a few private investors if you had reason to be confident that you were gonna make a self-improving cash printer. If, however, you weren't sure what you were gonna actually be able to pull off and saw a plateau or drop coming you could use an IPO to generate hype and get some liquid cash and then dump your position on retail and institutional investors if it doesn't work out. Why don't they don't rely on private investment and profits? I don't have the information to verify this for sure but one theory is that they don't think there will be much more private investment or cash flow.
This seems like a reach. Just because Anthropic could theoretically become a future infinite money printer, that doesn’t stop it from needing money in the interim.
It also doesn’t stop investors from being happy liquidating at an already good return, especially since they can choose to liquidate partially.
> For some reason they need the $100B today at the cost of dilution, rather than additional private rounds and/or using cashflow+profits to continue the advancement.
Both Anthropic and OpenAI seem to be burning cash as if it was a game of startup chicken. Of course they all desperately need cash.
Simplest possible answer is A: they are dramatically ahead right now, and B: there is not nearly that much privately held money available for relatively risky investments.
And once all those engineers cash out, China will become dramatically ahead because they are only a month behind already and can pay way better than the IPO
So much of "dramatically ahead" is due to the Claude Code harness being better than anyone else's.
Fable is great, certainly, but it's no better at writing or reviewing code than the top-end Carbon or Astra models. To the extent Anthropic has a moat, it's Claude Code. Obviously that's not much of a moat, but their competitors seem content to camp on the other side of the drawbridge for some reason.
What makes it a moat? How did you compare? I don't think CC has particularly outstanding integration (used in practice) or performance. People use it because Anthropic shoves it down their throats, making it risky to use anything else with their subscriptions, and generally gatekeeps their models like crazy.
I use all three major harnesses regularly (Antigravity, Codex, and CC) and CC is better. It gets used for writing code, while the others are used only for reviewing what Claude writes.
Codex is pretty close in a lot of ways, as the other poster says, but Astra consumes tokens at a higher rate, and when it compacts it completely lobotomizes itself. That used to be a problem in Claude Code as well, but it has been largely solved (although at the cost of making the compaction take much longer.)
Antigravity reminds me of the old comic strip where Firefox and Chrome are duking it out, while Internet Explorer is sitting in the corner wearing a helmet and eating out of a jar of paste.
I mean, OpenAI has codex? I find it pretty equivalent. So they are trying to equalize. (Less said about Google CLI attempts the better). And there’s definitely not model equality - Astra was a big step forward but generally loses on daily utility to opus 5.5, much less fable. I would be very scared for Anthropic if it was trying to compete purely on harness; I suspect most of its usage is by default because the models are better. In the pre 5.5 phase I saw a lot of migration to codex/pi, and I’d be fine without CC as a harness today.
What semi-supervised learning tells us is that any system with no humans in the loop will at some point reach a limit, and require more supervised data.
Deep learning had no humans in the loop, but there's always a bigger outer loop with a human somewhere.
For some reason, I feel really weird anytime Anthropic inserts beautiful paintings into their articles/promos. They've used one of John William Casilear's works here, for example (the one with a lake and a dotted line on top of it)
To me, this technology and these AI companies, in their current form, are as far away from the 19th century landscape paintings as they could get. As if they're trying to leech off them and fix their branding/message. Doesn't seem fair for some reason
>They've used one of John William Casilear's works here, for example (the one with a lake and a dotted line on top of it)
That painting is by John Frederick Kensett. It's called "Hudson River Scene", but they flipped it horizontally (mirrored) for some reason. I guess they wanted to use it insinuate that stonk goes up and to the right? Weird, either way.
There is a lot of value from LLMs but it is something that I can switch easily. Yes, some frontier models are much better but users can still chose to use cheaper models for most of the work. In fact most applications, through which people will interact with LLMs anyway, will not need frontier models unless they offer "chat with AI". Anything that solves specific problems (even multiple problems) does not need frontier LLMs.
Anthropic and OpenAI will have to become more of the end user application stack. Files, calendars, emails, photos, and much more have to pass through them for them to keep making money. Else they cannot make money like the way they have been in this early era of LLMs.
Switching models is easy as a solo developer. Switching LLM providers as a company is not, once all your employees have it integrated with GitHub, their IDE, etc. I think Anthropic will look a lot more like a cloud provider like AWS or GCP. Same-looking product, very high switching cost that would have to be driven at the CIO-level because of all the integrations that have to be redone, IP concerns etc. And ultimately they are selling compute capacity.
Edit: I should clarify that I don’t think this justifies their valuation. It puts them in competition with much more entrenched players with established customer bases, and who aren’t in massive debt to provide their compute. I’d also like to distinguish between a model provider as someone who provides weights vs a model provider who connects their GPUs and models to your codebase and internal systems - I think the latter is where the sustainable business is.
Incidentally this is why I still feel Google is well-positioned in the long-term AI race. If they can remove their head from their ahh and make a coherent product offering across Workspace, GCP and Gemini, they could become as entrenched as MS.
I would assume AWS is probably the most interested in acquiring Anthropic out of the three major cloud providers.
> Switching LLM providers as a company is not, once all your employees have it integrated with GitHub, their IDE, etc
I don't think this is entirely true. Companies are deploying open-source LLM gateways to be able to do that very easily now. Today, you can switch Claude CLI to any model besides Anthropic via https://github.com/BerriAI/litellm and there are major companies doing that.
This works with your favourite ADEs such as Orca as well so I disagree with this premise. The whole experience of switching is pretty seemless based on personal experience.
Agreed, deploying AI gateways has become extremely common in enterprise settings. A lot of admins think they are going to use it for guardrails, but I think the real value is in centralized cost tracking and ability to (relatively) seamlessly aggregate providers.
> Switching LLM providers as a company is not, once all your employees have it integrated with GitHub, their IDE, etc
Once the cost reaches a threshold, everything is doable. We ditched Datadog in less than a month. Someone I know had their company ditch slack in a quarter and google cloud usage of 100+ million in two. Everything is doable.
We have like a dozen different LLM tools available to us and it's as easy as clicking on a dropdown to switch in every single one of them, and we have several thousand employees. Self-hosting with the cheaper OSS models is also becoming more and more attractive for larger orgs.
That's just inference. Claude.ai is a platform that enterprise IT teams connect up to all kinds of other applications, and gets use beyond the claude code CLI. Non-developers get used to this functionality - whether it's their form-building stuff, artifacts, slides/presentations, etc... they're building an offering far beyond just inference. I agree the latter is a commodity now, though companies will compete on privacy options (ZDR, guaranteed confidentiality, etc)
Many large enterprises don't use Claude.ai. All their traffic goes through an internal gateway that routes to Bedrock. In addition, there is a whole ecosystem of LLM related infrastructure/tools deployed by "enterprise IT teams" where Anthropic barely has any presence.
But many do, so price them according to that. This is like saying "my office doesn't use Microsoft". Sure, lots don't. Lots do, no one is questioning that Microsoft has a very viable long-term market there.
For infra sure, but its dominance in other areas is quickly eroding. My org is super heavy Microsoft/Azure and we have all of their tools. Everyone making anything useful is using Claude + Databricks because it is a better experience for both builders and users by every measure and it isn't even a remotely close competition. Power BI and Copilot are a total wasteland. The only Microsoft program with any staying power is Excel.
Working for a 10k+ corporate and we made it pretty easy to switch to any provider, people here can use Anthropic's, OpenAI's, and our own self-hosted open weight models very easily. It's one drop-down or `/` command away to switch between them.
I think other companies are probably doing the same, the moat of a AWS/GCP/Azure is much deeper with how contrived it is to migrate between them even if you abstract as much as possible with something like Kubernetes.
Vendor lock in is quite well understood, most larger companies (the one's I have worked with anyway) are aware of the risk and take steps to actively avoid it. Once bitten, twice shy.
My company switched from direct access with providers to AWS Bedrock. Now we just pick which model we want from the list exposed by Bedrock. We can also leverage tools like T3 which allow hand off from one model to another.
Switching LLM providers as a company is not, once all your employees have it integrated with GitHub, their IDE, etc.
Speaking as someone who has been at professionally this since the 1980s, I suggest that while sometimes switching costs and network effects happen "organically," in this century businesses devote enormous amounts of attention to artificially building switching costs.
It's such a dirty practice that we have a four-letter word for it, a "moat." But of course, the moat isn't really to keep competitors out of a market, it's actually to keep customers trapped within it.
I fully expect every one of these AI providers to put as much work into trying to trap their customers within their ecosystem that they do to improve their capabilities.
Since siege warfare is about bringing down the fortifications and a moat is part of those fortifications, cyber siege may be something akin to patent trolling or IP copyrights? Fire off dozens or hundreds of claims against your target and drown them in legal costs.
Then you have the experience to know that every land grab in technology so far has followed the same enshittification path: At first it's all about competing with other entrants to provide value and win customers. Then, inexorably, it shifts from winning customers to defending their turf. Then, inexorably, it shifts from protecting turf to maximizing extraction from their turf.
If you believe that this will be different, why not? But for now, I'm expecting this to go the same way. Yes, LLMs are something completely new. But also, the people behind them are exactly the same as the people behind every previous land grab.
I can't wait for anthropics next big drop to be that it can order you a pizza, steering towards benign, token burning features that you didn't ask for and eye real-estate abuse for ads.
It's the end of an era. Lets pray open weight models give us a real choice soon.
In the meantime, Stonks up Anthropic. I'll always remember the good old days when you championed developers and changed what was possible.
> Switching LLM providers as a company is not, once all your employees have it integrated with GitHub, their IDE, etc.
I'm not so sure about this. I guess it depends on how well people did their job there.
Ideally, whatever people built is loosely coupled to that third-party resource, meaning that switching cost will stay low.
After all, it's probably just prompts + an API layer. Unless one micro-optimizes their prompts for claude (which then breaks once a new checkpoint drops, hence is a bad idea regardless), pointing them towards a different API endpoint should be doable.
Currently we are building agentic systems for various business functions like Legal and Finance. How these models perform in those tasks is paramount, and things like cost per run is only a part of the equation. It can become very subjective, and things like brand recognition comes into play. The cost of a counsel or a VP finance overshadows the savings you get by micro-optimizing for a cheaper model. We tend to use the best-in-breed LLM for the various tasks - mainly Claude and Gemini, and sometimes GPT.
> How these models perform in those tasks is paramount,
Then you have no choice but to use your own hosted model because that's the only way you can be sure the model isn't going to be removed or quantized/nerfed behind your back.
If your system can widstand the fluctuations of closed models over time, then it can most likely support switching between multiple providers in the first place.
If you ever need to scale to millions of customers, then cost per task is paramount. If it’s a one-off vibe coded custom project for one client, then best-in-breed is mandatory without it falling apart.
Ur case actually strength the argument for open weight self hosted model. You would have precision control over not only the technical issues, but also the legal issues.
Once the open weight models are good and cheap enough, there is no reason not to use them
I really think you overestimate the level of lock in here. If a companies tells the devs to use X because it cost 1/10 of the price they will just have to suck it up and switch.
I didn't particular want to work on a "all in agentic project" but that the nature of being employed.
The same LLM we use day to day will make it trivial to switch out providers in our infrastructure.
Easy to say, not so easy to actually get your mechanics, your project managers, and your business operations staff to get over that hurdle unless you’re willing to handhold them through the process of changing directories or explaining what an environment variable is to them. (This isn’t just boomers btw. Lots of young people have had the command line abstracted away from them their entire working lives)
There are many perfectly successful companies out there with limited computer literacy among much of their employees, let alone AI literacy. Those companies are great targets for Anthropic’s enterprise sales team, if they can provide an intuitive user interface that allows said employees to make use of AI tools, allowing them to leapfrog the need for computer skills.
We’re not talking about developers, we’re talking about people who just want the technology to let them do their job. What even is AI literacy? 2 years ago it was prompt engineering
Weird take. Tech boomers started with tape or commandline. The ide only became the development environment of choice in the late 90s, firmly with genx and millennials.
Claude code is a terminal interface not a command line tool. More like vim than sed.
My F500 company switched ~10k employees of Claude to copilot in 30 days. Google did similar between Jetstream and Claude. The moat just isn’t there right now
Let's say that the face for largest foundation model slows down enough to make the companies cloud-like, instead of whatever they are mow. Imagine the free credits that OpenAI would consider giving you as long as they are spent on migration out of Anthropic. They could even classify conversations by themselves.
Totally agree with your Google thesis. I would never count them out. I say that as someone that uses Claude (and a little chatgpt), but not Google's AI at the moment.
With ad money Google never had to learn how to ship a product that makes money. An obvious place they fail is that when something breaks there's no one to talk to to fix it.
Depending on how the company leverages them, it's super easy.
For us it's as easy as clicking a dropdown. The copilot vs code extension allows for this ez pz, the same is true for companies that implement an AI gateway for other tooling.
Switching LLM providers as a company is even easier than a solo developer, most large companies I know are using AWS Bedrock and selecting another model is basically just a checkbox for them
As far as I know, our corporate has deals with all of them: Anthropic, OpenAI and GitHub Copilot.
They stroke a great discount deal with all, so different teams/departments have different provider, and those using Copilot are even free to switch between models as they please (although with lower usage limits).
You could make the argument that it's not worth switching SAAS providers for enterprises when the costs were like 10 dollars a month per developer and switchign would require 1 team to work for half a year... but with AI providers it's more like paying 1000 dollars a month per developer vs paying 200 dollars if you strike a good deal with different provider and swtiching costs are more like 4 to 8 hours of work for each developer.
Indeed, and I think AWS is in a perfect position to consolidate the entreprise market rather than Anthropic here, they already have the rest of the server stack so this is just another addition.
I've observed that there are some companies and some folks that are very fanatical about a specific provider. "Only Claude" is very common. "Only GPT" less common, but still see it.
A lot of the non-technical and semi-technical folks are only comfortable using a specific client like the Claude desktop app. A lot of teams wrote skills and eval'd/benchmarked their skills specific to a model family and harness.
The "I can use any model, any harness" or "I only use a meta harness" (Orca, Paseo, Herdr, etc.) crew is relatively small IRL. Most people are still single-session, single platform user mode.
Most orgs have an enterprise contract with a single provider so all of their internal assets are built for the provider. It goes beyond skills, but also the telemetry, enterprise policies, and managed configuration.
For a company though, the idea that a few disgruntled employees will shift massive purchasing decisions is a bit of a stretch.
We have a developer on our team that keeps asking for a claude code enterprise license, we keep telling them "no, it's way too expensive, get used to using the copilot extension/cli".
For consumers, sure, but the bulk of the revenue opportunity is enterprise. Anthropic/OpenAI have a hard row to hoe IMO.
As an only-clauder, there's another elephant in the room. Anthropic is by no means perfect, but:
* using OpenAI supports the Altman/Thiel/Musk/Ellison/Karp/Trump billionaire Orwellian worldview and also ads
* using Grok supports a ketamine-addled open fascist sympathizer jizz-spewing 12-year-old-edginess memelord and his similarly broken worldview
* Google's models are pretty good but every single product they have ever had is ultimately either suborned or killed by the advertising division
* local models pretty much only half-exist because Jensen Huang wanted to cheat on his wife; future is unclear
One can argue for harnesses and so forth being lock-in, but the ultimate lock-in is morality and wanting to have a civilization at the end of all of this; I can't recommend either enough
It’s weird that people don’t seem to get this. LLMs are the most easily switched-out technology I’ve ever used as an engineer. The input is literally just free-form text.
> It’s weird that people don’t seem to get this. LLMs are the most easily switched-out technology I’ve ever used as an engineer. The input is literally just free-form text.
It depends, if you're automating business processes (honestly probably where the money is), then you'll need a really comprehensive set of evals and moniotoring to ensure stuff doesn't break if/when you change providers.
If you have none of that, then switching will be very expensive (but it was ever thus).
That just means the business automation is not deterministic, and highlights the issue with LLMs. The solution is not to switch providers but instead make the LLMs generate something that is deterministic given the same input.
> That just means the business automation is not deterministic, and highlights the issue with LLMs. The solution is not to switch providers but instead make the LLMs generate something that is deterministic given the same input.
I'm thinking of stuff that would typically be outsourced (like compliance stuff for identity verification et al), where it's really really hard to have deterministic rules.
LLMs are definitely a step up in dealing with fuzzy inputs, and enables you to mostly automate stuff that really wasn't feasible before.
But you do need really good evals and monitoring to make this work effectively.
Ah that would make sense then. I was thinking strictly in terms of business logic for handling input of data from an external source; sort of glorified data validation and then a decision based on the input data.
Yeah, it's the absolutely weirdest thing about all of this. People keep telling me I need to up my AI game yet no matter how advanced the tools get I'm still typing stuff into a text box. My skill is typing English. I was already pretty good at that.
People seem very unimaginative about the ways in which somebody equipped with a lot of intelligence and compute, and control over both, might make money.
The recent Anthropic push into biomedical research gives a clue.
I am not unimaginative about Dario's megalomania. I just don't believe he has the ability to execute. He has no prior experience to Anthropic and nothing he has done as CEO would create confidence that he is actually capable of operating at this scale.
This is equally true for how risky it is to build such a business on top of a 3rd party black box. Model nerfing is widely speculated and the only true way to guarantee that the models you're using are stable and consistent is to run your own hardware. Not to mention the recent Navier-Stokes controversy.
The moment you find a money-making formula on top of someone else's black box model, there's nothing stopping them from taking it for themselves.
Model nerfing is rampant from people using subsidized plans.
Stories are much fewer and further between for people using the API. Which makes sense, the subsidized plans are "You get what we give" whereas the API is "You get what you pay for".
One thing I have noticed; as a project matures frontier AI is less useful essential.
Qwen3.8 and Meta Muse Glimmer are great at "fill in the middle" of frontier architectures and plans. Frontier generated plenty of great context already to keep them on rails.
Delegating like a CEO down a pipeline of expensive experts to cheaper local small models could become a common trope, especially if Apple M7 and beyond live up to rumors as LLM tuned, and chip competition exists to keep forward progress, SaaS models are cooked.
The goal for chip corps too is the right mix of generic model enabling logic; not necessarily a full concrete model but enough of the abstract to efficiently boot strap and run a model; move common patterns to silicon not data.
Token vending is little more than behaving as a "wrapper app" over electricity. There is zero technical moat to their business.
a big move into end-user software would mean that they have to compete heavily with their own largest customers. The economics of offering SaaS are also under meaningful pressure and don't warrant anywhere near the valuation that is being suggested
The biggest moat with applications like MS Office is compatibility. Sharing a slide deck or spreadsheet with others is a simple decision, you know they have it or have software that can read/write it.
I don't get the need to have Anthropic build their custom versions of MS Office apps, I can tell Claude to generate a word doc from a wiki page and it does a pretty damn good job. Anthropic's statistical median level of software that they had Claude create isn't going to replace Office apps to the point where companies cancel/stop renewing their MS Office subscriptions/licenses.
The other element is trust. Do end users trust their data in Anthropic's systems? A lot of people get creeped out when they search for content from one device in their household and then nearly instantaneously they start seeing ads from other devices on the same topic. How creepy will AI get if it starts consuming all your docs, spreadsheets, personal finance info, emails, etc.?
I beg to differ I already am stopping using MS office recently but that's mainly because I really dislike how they lock up the apps if your late on your annual payment. It's a grating value capture model. I'm sure I will have to get back to it at some point but in the meantime google is providing great optionality and I don't have a concern that the software stack has glaring security concerns.
On the trust thing - I don't trust anthropic and I don't think MS has been a great steward over the years either.
The moat of backwards compatibility and plugins is why MS Office has survived decades of competitors without any appreciable loss of market share.
It's hard to see Anthropic replacing it, and even if so - why bother, the revenue is miniscule. They need 100 MS Offices for the compute contracts to pencil out.
End-user software is going to be under a lot of pressure. If you're Microsoft, Apple, or Google: you can easily copy products or integrate ideas into their platform. I see companies like Adobe, Salesforce, and Oracle having to downsize significantly to stay profitable. Having said that, my in-laws still pay for an AOL email so I don't think their cashflow is going to zero soon.
Adobe seems to be in a tough spot, because so much of the lock-in is client side, I think Salesforce is fairly well protected, it's just so impossibly difficult to remove once a company has fully adopted it. Similar to the staying power of the ERPs of Oracle and SAP.
It's the same really. There have always been alternatives that can work with Adobe's various file formats. Even more aren't going to change much, at least for their enterprise key customers. This is not only the case for Adobe, but many others as well, e.g. MS office is still a thing these days, and so are things like 3Ds Max or Maya, albeit having many compatible cheaper or even free alternatives. It's not so much about features, pricing or even stability. It's about customer support. Do I want to ask an LLM and test my luck or write a GitHub issue and cross my fingers that someone will read it within the next weeks? No, I will pick up my phone or write a mail and get a response within hours if not minutes.
I ask my llm to write custom software and love the freedom to customize it to my needs. Today I do a presentation on Ubuntu with a new "Presenter" app that works offline, runs 100% in CUDA on GPU and streams videos, starts in 2 seconds, lets me demo live robotics policies in 3D simulations and Opus 5.5 wrote it yesterday. This may sound very unusual, but I think it will be common in the future: moats disappear one by one.
I would say that Apple's Create Studio can move toward feature parity faster with Adobe. My point if you're not a platform provider it's going to be harder to make a profit.
The risk for Adobe isn’t that we replace their software with some AI generated alternative. It’s that we bypass that class of software altogether and have an AI do the work.
Surely not just prompting like this and it will one shot it. The usefulness of the flash authoring tool, was it was battle tested and polished.
But I did gave it a shot to took up developement again of abandoned www.wickeditor.com (a web reimplementation of flash). Good results, but limited time and energy for that right now.
The big problem is Photoshop. I have no love for Adobe, I proudly moved my department off of all Adobe products a few years ago. I am a big fan of open source solutions and I am always impressed at what the GIMP team (and some others) have accomplished.
But…
Photoshop is king. Everything else is behind by a solid decade. If you’re a professional the $40 a month or whatever it is now means nothing to you as well. It has the professionals are all on it, and because the price isn’t astronomical, that’s what everyone else uses who needs to use that kind of program more than once in a blue moon.
Premier? Replaceable. Lightroom? Pretty replaceable though still somewhat ahead. Photoshop? There’s only one Photoshop. And as much as I’m loathe to give Adobe any credit, it’s the best by a mile. There are too many essential features in QoL things that other programs are missing.
Luckily my department doesn’t need it, but if we did, we would simply have to accept Photoshop.
Not a graphic designer, so, truly asking out of curiosity: what are the QoL things and essential features in Photoshop? I've been slumming it with GIMP for years now and not felt anything very obviously missing, but again I'm not a pro.
Compatibility issues, which can be a huge deal partially due to near-vendor lock in. If you historically been on Photoshop, bringing your working/former projects into another software is not easy, which matters because revisiting work is common.
Not all tools are non-destructive in GIMP. Things like raw support, smart objects, retouching, etc. all either are super thin compared to Photoshop or require workarounds/a lot of manual effort. There are just lots of tools you expect in Photoshop that either don’t exist or exist in an inferior form. I can’t elaborate too much because I don’t really touch Photoshop much these days, but I work with people who do and worked in production long enough to know the expectations. GIMP is great but Adobe is a massive company with deep pockets, it’s hard to keep up. Their user experience is miserable but the tool itself is just far more robust, and a client simply isn’t going to accept that you won’t do something that Photoshop could do quickly because your alternate program can’t do it or does it with a lot more effort. That’s the reason I ultimately dropped Final Cut!
I'm not sure corporate procurement has ever selected the cheapest and best option, so I think they may be fine. Plenty of reasons to avoid completely locking into your cloud provider as well.
> If you're Microsoft, Apple, or Google: you can easily copy products or integrate ideas into their platform
This argument doesn't hold water. Google tried like no one else to have a chat app and it failed miserably more than a handful of times. Same for social networks.
Adobe creative suite is $70/m. From most businesses perspective it's not something worth looking for cheaper alternatives and the alternatives has existed for a long time now.
now imagine that said customer/competitors have been uploading/generating their source code, problem reports, marketing collateral, and other IP _to the LLM company_ for the last few years.
- work: I need frontier, cost of mistake is huge
- home: i do entire renovation with it, cost of mistake is big. monitor baby health, development, etc: cost of mistake is big , taxes: cost of mistake is big
casual researches about certain facts, shopping, booking: cost of mistake is medium
non of this fits : i will run an expensive laptop hidden from the baby, since i'm on the move, there is a small percent of casual queries that normally will go into finish products: google ai, muse, etc.
If Google and Microsoft have gone dormant in this space, I have to think it’s because they haven’t been seeing any ROI, either. Possibly because customers have been so underwhelmed.
I’ve been using Office Copilot more or less since it was released, and the general opinion around the office is that it’s a huge waste of time. The auto-transcripts produce 3 pages of fluff when previously someone would send 3 sentences’ worth of what actually mattered. The auto-drafted documents rarely add anything meaningful beyond what was already in the prompt. The question answering and research tools are only as good as Sharepoint/Confluence/Jira/etc’s pre-existing search feature, and therefore untrustworthy, etc.
What made Anthropic exciting was they had locked onto a use case that reliably generates positive value: coding. Not much else has proven, from a business perspective, to be anything more than an exceptionally buzzworthy money pit.
Yes, that's the direction I've been witnessing for Claude too. But it's a short to middle-term strategy IMO, by the time companies are ready to drop all these apps they will need something else entirely, not Claude's slides or spreadsheets or Salesforce.
The fact that Frontier AI can become Google Workspace at a flip of a switch is what is crazy valuable. Anthropic and OpenAI can enter and disrupt any market, especially software, faster than any company ever could, because users want the flow to start and end in their chat interfaces.
I think eventually almost all non-creative software will, essentially, be a wrapper around a session.
This pattern has already happened before, in the translation industry. Someone figured out how to learn from and recall context from ordinary translation work (the "translation memory"). It started as an add-on to Microsoft Word, and matured into a standalone software, that was ultimately a wrapper around a document editor. Then language management systems plugged into that wrapper, hey presto, every translator at your company does the same work they were always doing, and you've produced an entire corpus of quality translations to reuse everywhere.
Now _translate_ that idea over. Every time any human works with an agent on anything, simple heuristics find everything true and reusable, save it to the everything-memory as skills, refences, glossaries.
Anyone does any work, all agents can now that work at that organisation. Self-fine-tuning models everywhere, and it fits into corporate intellectual property expectations.
I have done something like this at a medium-sized listed business.
Helping my wife this past week try and setup an automation in her workspace account revealed how dormant google is here.
Gemini is unable to take an attachment from gmail, read it, then place it in the correct drive folder and update a spreadsheet with what it found in the attachment.
For whatever reason, Google decided to firewall "Gmail Gemini" from "Drive Gemini". Gmail Gemini can send files to drive, but not edit anything, and Drive Gmail can edit files, but cannot be initiated from gmail.
Meanwhile claude code with Google extensions can do this no problem.
> Meanwhile, Google and Microsoft, which already dominate that area, seem to be dormant.
Google owns 15% of Anthropic I think.
And Google offers Argon 4 to paying Workspace and Google One customers now. How's what was the number 1 model on nearly every single bench a few weeks ago being "dormant"?
I know people all in in the Google ecosystem (getting credits from Android phones they buy etc.) who only swear by the Google models and who are now using Argon 4: I asked them and a few days ago they had access (but they were already using other Google models before that).
Doesn't GMail have billions account by now?
I'd tend to say with billions of users on GMail and a model like Argon and 15% of the Anthropic shares, Google is not where you think it is.
It's easier to use claude to do stuff insider workspace than it is to use the integrated gemini. Claude has more power to do stuff than gemini, in it's own house. That is such an egregious fumble that it's honestly hilarious.
It's almost like the workspace devs are anthropic plants that are purposely underminig google.
The sad thing is Google could totally own this space.
They had a huge head start, more raw data (from search) than anyone, a huge installed base on Workspace and consumer Gmail, tons of compute resources...
Integrated gemini stuff sucks but it's nothing that special about Claude, general agentic tools are just better than app-specific ones. Agy is also much better at doing things in workspace than the bespoke in-app gemini stuff
Right, because Google actually has assets and a responsibility to keep their customers secure, whereas Claude can do things that are completely insane from a security perspective because they have no liability.
Why haven't people switched then? Same could be said about Google search, Facebook, X etc. Claude is the de-facto SWE model now, yes you could use an open model but you would always be wondering if Claude would do better, like using Duck Duck or some other Search engine
> Claude is the de-facto SWE model now, yes you could use an open model but you would always be wondering if Claude would do better,
In my testing (incredibly similar workflows, moving around between claude code and cursor), I haven't found that much difference between Sonnet and GLM-5.3/flash.
Sonnet seems lazier, but has better prompt adherence, and is constantly telling me it can't do things, while GLM will just Find A Way to do whatever I've asked for.
I really appreciate being able to see GLM's thinking tokens, as I can head it off quicker. Claude's a little harder to steer, but the integration in Claude code makes it much easier to leave it running on things overnight, which is hard to get working properly in Cursor.
Overall, I'm not sure Anthropic have that much of a moat, in coding particularly. This seems weird to me, as they have so much more usage data that's hard to replicate.
Why? Given that there is a limit of things that they can do, why would inventing a new "AI-native OS" be an interesting target? They could instead focus on solving actual existing problems in science and engineering, finding solutions to problems that they know people already care about and would pay money for, instead of putting a lot of effort into maybe inventing a middle layer, that might be used to solve problems or maybe not.
There will also be a lot of competition in that space, because everyone can take a stab at building an "AI OS" (whatever that means) using AI for basically free.
One more piece - say Google Gemini gets way better than OpenAI or Anthropics offering, or Microsoft Copilot etc gets way better...
If Microsoft or Google's AI is killing it, and it's built into the OS, how will Anthropic or OpenAI compete unless we run into the same old antitrust/competition issues with OS?
At that point, they might just say - "fine, it's easier to just have our big AI brains and money make a new OS we control and we can compete too."
- at some point, look at what they have they can do to grow to satisfy shareholders - AI companies have been acquiring companies that make developer tools - that is the build side - but once they've exhausted that, if they want to keep growing, what do they go after? Bigger and better models at ad infinitum? I don't think they can go on like that forever, and, if optimization of AI and competition keeps cutting into their main business line of LLMs
etc, they have to diversify.
- If everyone has the power to make apps, there is only so far you can go there - my guess is that the pressures to reinvent regular things like office suites, email clients, calendars to be integrated into their desktop tooling is part of this
- who is in the best position to be the platform/front door for everything, if everything depends on AI?
- when have these kinds of companies ever said "this boundary is enough"? It never happens, they keep going, keep trying to find ways to expand, capture, build moats, etc.
- look at what their desktop apps are turning into - they are already an application layer and becoming moreso every day, at some point they will ask - why are we sitting on these other OSes... we have enough money and we can get into more "hard" engineering things, like building hardware using their own tech (which they are already experimenting with), and if you're building the hardware, why not build the OS?
- and, what is an OS in the future? Have you considered what a post-AI OS even looks like? Does it need a start menu? Does it need any of the legacy stuff? Is it a local software factory?
There is someone already experimenting in this space - the guy who founded Basecamp and is building Omarchy [1] Linux... he sees it.. how long until these companies do?
If they don't build this, what will Microsoft do? Both Microsoft and Apple are well down their way, Apple is probably further ahead, but neither of them are the frontier model companies. Don't forget Google has a stake in OS too.
"Everybody will try to build an AI OS", but which ones are the most likely to survive in our current economic system?
If our economic system was based on partnership, I might agree there isn't much point in these companies building OSes, but ours is based on competition and domination generally... so unless the AI bubble truly does pop and the funding stops (which it may of course), how else do they keep fueling their stock prices to dominate the market?
I think something is going to have to give somewhere in the chain... otherwise the ease with which any software can be replaced will be too high and I suspect they will be forced to compete.
One last thought - between Microsoft/Apple or Anthropic/OpenAI - long term - who is in a better position? The debate arises from that question, in my opinion.
I hate to even put this out there, but "OS-as-a-subscription" (OAAS) could be a way. Your Claude payment covers all the software running 'your' computer, plus the ability to develop new 'apps' for specific needs on the fly telling the OAAS computer in human language what you need. Throw in services like email and cloud storage and some customers might find such a bundle attractive for the right price. It would suck revenue away from MS, Apple, Google. Enough to justify this valuation? No idea.
Based on how shoddy their customer facing applications are, I wouldn't trust any of these AI companies for an OS. It's like Google II - can't trust them to support a product longer than it took to develop it.
I agree with you on quality... but if AI/LLM/etc continues down the path of continual improvement, it's only a matter of time until the products become better or different enough that it isn't an issue.
And making products fast and first over quality has never stopped these companies from trying, when they have so much funding....
I do not share your same hope of improvement given time. The commitment should be on quality first, not speed of development. This is why we are all drowning in a proliferation of poopy modern software.
The fact that their desktop offerings are so pap is an indictment of their engineering prowess and a demonstration of their apparent inability to use their own (actually very good!) LLMs to generate decent software.
It baffles me how a technology so good at spewing out software is apparently unused by the people who are developing the technology.
This is the same argument we have heard many times with SaaS that would presumably get replaced by open source equivalents. In practice it does not happen because there is a lot of value in having someone else deal with a problem.
Nobody trusts Anthropic's ZDR, in the limited cases where they even offer it. The key thing is that weight activation sequences from a query are not "customer data" they are internal Anthropic data. So in Anthropic's legal theory they are training on their own internal data, not "customer data," and that is perfectly legal, even though it is the exact same thing in practice as training directly on customer data.
The problem here is having a stable access to LLMs with strong trust the provider will respect ZDR, won't poison the models and will provide a stable platform you can depend on.
But you’re wrong ser about small or cheaper models. Choosing to work or not having a choice to work with cheaper models means you work slower and more methodically because you have to plan out the work and size it to fit. This has huge implications on the value of the work you can produce as an employee or solo developer, and it’s a skill working with these SOTA models - there is specific crafting that must be done to be effective with each and their capabilities.
You will see the old eyeball real estate game, forcing more of your experience into their apps and platform so you can see ads or “new slop features”. You will see the SOTA models more closely guarded in an effort to bait you into that platform, and yes, you will see them doing things they never should be doing, moving towards an open AI like model where it can order a pizza, where monetization it was always important, but is now the golden rule, before all else.
I hope open models can at least give me opus 4.8 experience soon, I can work with that. But nothing out there today can do it and that terrifies me.
> Yes, some frontier models are much better but users can still chose to use cheaper models for most of the work. In fact most applications, through which people will interact with LLMs anyway, will not need frontier models unless they offer "chat with AI"
I am finding some really routine tasks to be borderline with frontier models. Performing E2E testing of enterprise web apps would clearly disintegrate with local models. You need that reasoning black box on top. I don't care what your favorite benchmark says. I care about what actually fucking works.
The people who don't understand the economic situation are those who are in the pure play space with AI and multi agent circuses. Why bother spending any money when the occasional failure has literally no consequences? Who in their right mind would pre-pay for tokens?! I wish I could see the statistics on what % of HN uses AI for fun vs for work. It is clearly a kind of Factorio entertainment experience for a lot of people. If I was just using it for entertainment this whole IPO business would look really silly to me too.
We all love moats, but I mean, it is still possible to make money in an industry simply by being the best at what you do and charging for it.
Anthropic has shown a consistent ability to deliver the best general purpose models. Can that continue indefinitely? Will models become "good enough"? Will open weight models catch up to SOTA models? Will some genius start a new company with better models?
And, on Anthropic's side, will they be able to use their market power that comes from being the best to push through regulations that will create a moat; instituting some sort of mandatory safety checks that coincidentally are exactly what Claude is trained to pass, regulating the amount of compute or trying to claw back data centers to prevent new training, etc.
Or maybe they just have the magic sauce and they're going to stay on top of the game and people (like me) will be happy to continue paying money for their product.
I'm fascinated by exactly how an investment in the company that first reaches AGI (or whatever we decide to call it) might pay off... It feels like the stocks would absolutely be worth owning in the immediate term because they'd be hot property at the point where people realise it's happening.
The thing that I'm less clear on is, if and when they achieve some kind of recursive improvement exponential growth thingy, isn't that going to heavily disrupt the entire economy / stock market? In that environment, would Anthropic's stocks really be a good bet, given they have to operate within the economy?
> The thing that I'm less clear on is, if and when they achieve some kind of recursive improvement exponential growth thingy, isn't that going to heavily disrupt the entire economy / stock market? In that environment, would Anthropic's stocks really be a good bet, given they have to operate within the economy?
That's a long-term problem to be dealt with later, greed dictates to pursue the former as long as the music is playing and everyone is dancing. Similar to 2007-2008 when it was clear that things were breaking but the incentives were to play the game for as long as possible before it imploded.
It's clearer to me why this works for generic investors than for the AI companies themselves - unless their plan is just for key employees to cash out and live on a private moon just before Consequences arrive.
>> that first reaches AGI (or whatever we decide to call it) might pay off
It is a myth. They are chasing science fiction. They either dream of an omniscient overseer who will take control of society and nominate its shareholders-parents into the governing elite, or they envision a chatbot that will solve every problem, again turning shareholders into the power brokers. They forget that nobody will ever be allowed to "win capitalism" and take over the planet. Companies have are torn down for far far less. Every AI model is data, ones and zeros. It therefore can be copied/shared/leaked. Intellectual Property law is not a ladder for assent to lordship. It can and will be rewritten or outright ignored rather than allow capitalism to collapse.
> It feels like people are still trying to bet on the horse that will achieve gAI.
I think that at this point gAI is just a marketing blurb, and the actual race is to dominate current practical applications involving basic RAG and ReAct loops. This can go from anything between software development, business intelligence, and market research. A company whose top brass can use a team of agents to do what nowadays they assign to mid- and top-level management is an executive with tighter control over the whole business process. For this you don't need gAI.
I've noticed they've actually stopped harping on about this AGI nonsense, I suppose because no one was buying it to begin with and they themselves didn't really have any idea wtf they were talking about when talking about "AGI". Now it's all about a bajillion agents running autonomously or whatever
Easily in the context of the corporate world is to pay for something, that something gets installed and then just works, and if it doesn’t work there’s always someone you can call 24/7 to come and fix it. I don’t see how this is possible with open models.
the answer is to use a gateway that all the prompts go through. There are many proprietary and open source solutions for this. You get the benefit of making it trivial to switch between proprietary providers in case you are unhappy with the performance or squeezed on price. The AI Gateways put cap on LLM providers' margin upside.
Additionally, the gateway allows for caching, additional security checks, centralized monitoring. I think most corporations have a multi-provider setup anyway.
What do you mean? We use Bedrock at our company and I can switch to literally any model, including the open ones, by just typing /model and picking it, even including inside Claude Code. It's ridiculously easy to switch, and imho GLM 5.3 is good enough for 99% of my daily tasks already.
You pay a service provider, who manages the open models… there is absolutely nothing special about it. The hardware can be something you own. Or you rent. Or you just pay the service provider to access the open model apis. Lots of options here, it’s not different from pretty much any other software service in the corporate world
What work do you find frontier models significantly better for?
I have been playing about with different models having started a new job recently, the medium priced ones don't seem any worse so far. Admittedly my tasks have been fairly straightforwards so far.
> Anthropic and OpenAI will have to become more of the end user application stack. Files, calendars, emails, photos, and much more have to pass through them for them to keep making money. Else they cannot make money like the way they have been in this early era of LLMs.
I believe the models have been effectively commoditized. It's believed by many that the lead of the frontier models over open weight models is only 6-9 months. That was one of the pillars holding up the SpaceX/Anthropic/OpenAI valuations.
The other pillar is hardware but I think that's only going to last a few years at most.
These trillion dollar valuations were based on the idea of "winning" and "owning" AI (IMHO) and the open weight models have killed that outcome. Now it's a rush to cash out before it all comes tumbling back to Earth.
They are fungible in the sense that they can accomplish the same task up to a certain acceptable quality.
If I needed to implement a simple function that is well described, all LLMs (for the most part) are fungible. They will have almost identical implementations (sans variable names) and all be fit for purpose.
Not only can you switch easily, they all seem to do the same thing, fail in the same ways, make the same bad linguistic choices, write the same code... I think at the end of the day, LLMs are a commodity. They're like a grease, or oil. They're trying all the tricks with branding cute names, version numbers, benchmarks, thinking levels, but it all just seems the same stuff to me: I put in tokens, I get more tokens back, and if I didn't know ahead I wouldn't be able to distinguish if they'd come from Anthropic or OpenAI or wherever. Maybe some people can but I can't.
I still think they are on a tight timer. Local models are already extremely good - good enough for most normal users. They are going to continue to get better. Once hardware starts to become affordable again, we will see AI appliances that can do everything most users want, and privately.
> Local models are already extremely good - good enough for most normal users
Are they? I'm on a macbook M5 pro 48GB (so, much better hardware than most users), and the speed alone is a problem, qwen 3.8 27b is ~10x slower than opus 5.5.
Capability-wise... For software engineering, qwen is nowhere near as good as opus 5.5. Might be fine if your usage of LLMs is close oversight, but it's not how I use them (I do review AI work closely and iterate, but I expect agents to autonomously do good work nowadays). I've not tested it much for "life assistant"-type things, but the speed would make it unbearable.
With couple affordable GPUs you can get it to 200t/s. Macbook is not the best price efficient example. Tensor parallelism is the key, the same technology that they use in the cloud.
what quant and engine? i have the same hardware and i get 40tps with a 4bit quant using omlx + splash. which is about the same as opus, or faster, maybe? with its caching the time to first token is also very impressive.
> Are they? I'm on a macbook M5 pro 48GB (so, much better hardware than most users), and the speed alone is a problem, qwen 3.8 27b is ~10x slower than opus 5.5.
That's hardly relevant. I can use any local model through providers such as OpenRouter and get blazing fast prompt response times, and you're charged a small of what an OpenAI/Anthropic prompt costs.
What makes local models extremely competitive is the fact you do not need a huge data center to run them. The hardware they require is far from impressive. Check Mac studio benchmarks and how some models output >200token/second.
Yes they are, but you should understand what is the defining feature of a local model. Being able to run them locally is a consequence of that defining feature, not the defining feature. And it turns out that it makes them cheaper to run.
Moreover, you can run both local models and non-local open weights models through providers such as OpenRouter. They evem have Qwen2.5 7B.
> Also, open-weights models are more expensive per task than Haiku and Luna
You should really check your notes. Luna only remained competitive because OpenAI was forced to halve its cost just to avoid being dominated by open weight models. I seriously recommend you check out OpenRouter's benchmarks.
> I seriously recommend you check out the cost-per-task Pareto frontier.
I checked. OpenRouter provides their own. They use their actual cost, which is what they charge users, and their benchmarks results. Please, spend a few minutes to browse it. You'll notice that the likes of Luna is dominated by quite a few entries.
I mean the person you are replying to has a fairly capable Mac, and even the smallest generally-agreed capable model is still too slow for their liking.
We don't all have 10k lying around for hardware. A $200/month subscription would take you 4+ years to cost $10k.
I thought this, and ~4 months ago ordered a M5 Mac Studio with 128 GB RAM. I've been experimenting quite a lot with local open weight models, including but not limited to DeepSeek V4 Flash, V4.1 Flash (the latest one), various Qwens, and a couple of others.
What I'm seeing is that although some open models are capable enough for some coding work, they're way behind Codex and Claude even on being able to reliably complete tasks. They need quite a lot more managing and steering, and sometimes you need to retry a prompt several times to get the result you're looking for.
That's not to say they're not useful - far from it. But what I'm also seeing over the couple of months I've been experimenting is there actually seems to be a capabilities ceiling that's not really going higher for the same equivalent (local) compute, even as new open weights models are released. I think Qwen 4's delay is a signal here too.
> Local models are already extremely good - good enough for most normal users
i would love for that to be the case but i think you're completely wrong. setting aside the question of what "good enough" even means in this case, i don't think a normal user has a machine capable of running a decent local model with any sort of acceptable throughput.
I agree, I think the bigger threat to them is the neoclouds serving flash models like GLM 5.3 flash or DS 4.1 flash 1017, those are very good, very fast, and very cheap. With zero data retention and less guardrails they become an obvious choice for more work, and enterprises can post train on top of them.
That being said, their own small models (Haiku/Luna) have gotten very good as well..
I really don't see why by this time next year most people won't be paying pennies for models in this tier and getting the performance we'd expect from current Opus/Sol models, e.g. good enough for 95% of what you need
I think the models themselves are "good enough", but people are underestimating how much the integration with tools effects the quality. And OpenAI/Anthropic are lightyears ahead there.
Very true. I think for your run of the mill user they’ll reign supreme. But for companies on Bedrock/OpenRouter already, I wouldn’t be surprised if they switched to open weight and open source harnesses.
Considering most corporations are the ones who buy the most tokens, this is relevant because:
- They can afford to dish out hardware to run models locally
- They can afford to deploy even larger open weight models on their own infra, even if they’re not local
In addition, the writing is on the wall. These open weight models are getting good enough and small enough with each iteration. Hardware will (hopefully) come down in price eventually so a layman could deploy these locally.
As AI boosters love to say: [Local models are] “the worst they’ll ever be right now!”
Open models seem to catch up in spurts. There was short period where suddenly a bunch of Chinese models came out that were on the verge of catching up, like GLM 5.3, Qwen 3.8, Deepseek 4.1, Kimi K3, and so on, but then Opus 5.5 and now Fable 5.2 seem to be pulling away again.
I don't think they will ever overtake because they are probably using distillation, however once the open models pass the "good enough" threshold it won't matter.
See, users are quite price insensitive for models below the threshold, no matter how many tokens you throw at Opus 4.8 it will never be as good as Opus 5.5 because of compounding errors. Above the threshold, price becomes hugely important because results come from throwing more tokens at a problem rather than using a better model.
Open models yes (eg Deepseek Flash) but local models are not, at least with attainable hardware.
My brand new Mac Mini M5 Pro 64GB can run Qwen3.6 at a reasonable speed but it's nothing compared to hosted models and likely never will be.
That said I'm optimistic that in time we will at least get to something close to DS locally and if that's true I could see a lot of dev work being done locally, at least for more clear cut tasks.
local models are economical pipe dream. Before open router, you have to buy GPU commitment for any large usage. essentially doing that 24 hours every day in a year.
Why on earth anyone would use local model, and absorb all the costs, if you can just pay $20-$100/mo and not deal with any hustle? Unless you are developer, or someone who use it extensively, but in this case, local model is not enough for you.
If your use case is trivial you pay extra price for hardware while using mediocre quality. If your use case is intensive , e.g software, science, medicine, legal, etc, you want cutting edge. You getting it for free, immediately.
Sure, but as "local" models improve, I've heard of more and more companies deciding to bring it in house. As hardware prices possibly get more attractive, that will make more and more sense for many companies. Enterprise don't mind paying flat rate for SaaS but variable billing model gives a ton of business people pause.
I'm an SRE and I just wrote some terraform to deploy G4DN instances for local evaluation for mid-size company.
> we will see AI appliances that can do everything most users want, and privately.
we _could_ see that but would we really? you think? Since when do companies these days make the best experience they could for the consumer? Instead of spying / upselling / advertising etc.. while providing the bare minimum needed to hook them into the initial purchase
Anthropic's spend commitments at half a trillion are unbelievable. I love their product but think they've leveraged up awfully hard... makes me wonder how quickly software engineering, marketing, and other valuable features will soon be siloed and locked behind increasingly lucrative paywalls.
In a roundabout way, I hypothesize about Anthropic's future pricing power by using the analogy of historical cable TV package pricing compared with the total price that households are now paying for various streaming services. Cable TV was often $100+ per month in the 2000s when Netflix came along and offered most of a viewer's favorite shows plus movies and eventually originals for $7/month, streamed at any rate a viewer wanted. Over the years since we've watched media companies snatch back their most valuable IP and start their own streaming services, and the result is households now pay a total cost that approaches what people used to pay monthly for cable TV packages.
There's little doubt we're in an era of very cheap AI, as most people are well aware given the amount of money Anthropic and OpenAI are losing. I also hope local inference can help suppress the future pricing power of Anthropic and OpenAI and continue to enable somewhat equal access to all of the benefits of AI in the future.
One possibility I've been thinking about recently is that there may be an AGI algorithm - one algorithm that can do anything, solve any solvable problem. Once we reach this model there is no point training any new models really.
Then it will come down to who can supply such a model with more hardware, and more information to take into account.
>Local models are already extremely good - good enough for most normal users.
Not really accurate. Normal users aren't going to go through the trouble of setting up a local model and sure aren't going to be willing to spend the money required to actually host a good model at usable speeds. Your average tech nerd buying kitted out Mac Studios with 128gb of ram are nowhere near a "normal" user.
which are those "normal users" that have capabilities to install, run and maintain local models to be somewhat at the same level as enterprise models? Not to mention RAM and hardware shortages caused by AI model companies. Level of delusion is unreal
One interesting thing from their presentation (especially given recent events): They claim to have used Mythos to port Claude from running on nVIDIA GPUs to running on AMD GPUs in just 3 days, with basically no human intervention. If this is true, it significantly reduces nVIDIA's moat regarding GPU compute. https://youtu.be/eynv5VR9qos?t=1230
Are you talking specifically about porting from CUDA to ROCm or whatever? Or just in general, vibe coding stuff in a few days and then having a long tail of pain?
All claude code subscribers are classified in this category it seems.
Total customers >1M trailing twelve month revenue: ~2150. I suspect a vast majority of their revenue comes from limited list of customers with >100M revenue.
ChatGPT was first mover and has the non-engineering AI chat market. My teen kids say ‘Chat’ to refer to all of AI, and they mean ChatGPT. The rest might as well not exist to them.
Claude has done amazingly well in the SWE domain, because it’s often the best user experience. But most of the world is not HN.
I think we're all in a bit of an echo chamber. Our job is to keep up with this stuff, and we work with people who are ALSO keeping up with this stuff.
To my friends (mechanics, CPAs, musicians, police officers), girlfriend (RN), and other people in my personal life... it's ChatGPT. (It's a neat tool to generate funny pictures or ask a question).
Now that I think about it, I have NEVER heard anyone mention Anthropic / Claude outside of work.
Maybe not interesting? I wouldn't expect the number of
subscribers to a bespoke "AI Service" to rival Netflix. I realize on this forum we're probably getting lots of use out of them in our personal lives that justifies the $20/mo but we're also the people who pay for search engines and VPSs. Outside of work and school normal people don't do anything that requires general purpose AI. These companies are trying really hard to pitch use cases
for everyday things but you can see from the ads they're super forced. The biggest value add for most people is probably "go google something for me so I don't have to wade through the sludge" and Google is giving that service away for free.
Here's what I found most interesting in the Customer Stories:
> 4.3 Non-billable hours recaptured per lawyer weekly, in an independent study of 31 firms
-Legora
> 24→5 days Working days for new feature time to market
-Rakuten
> 2x Quarter over quarter productivity gains (estimated 8.1 million hours annualized)
-Salesforce
The rest are pretty fuzzy/fungible (percentage of employees "using" Claude etc.). But these claims potentially show real, measurable productivity value (assuming they're true and measured accurately).
> 2x Quarter over quarter productivity gains (estimated 8.1 million hours annualized)
This is a great encapsulation of why productivity gains are an imperfect measure. 2X is a massive improvement. Yet IMO you can barely tell a difference as a user of the various SFDC offerings (Salesforce, Slack, Tableau, etc). So where is all that work going? A generous interpretation is that it is in the form of small, somewhat invisible improvements that add up over time, but that seems unlikely give then nature of what people prioritize at work...
My vantage point as a technical seller at the margins of this space: Outside of software development, the productivity argument is a challenge to CFOs / Finance / Procurement. The ship has sailed when it comes to friction free AI purchases. Buyers need to articulate why productivity will adjust the trajectory of top line or bottom line metrics for the business.
> (assuming they're true and measured accurately).
Precisely, because TTM doesn't tell the full story about the potential shifts in quality of the work. I don't refer to vibeslop here, to be precise, just the additional followup work required on those features released faster, which always happens but with AI it may or not happen more often.
TTM for sure is a bad indicator. Earlier this year at $DAYJOB, they attempted to switch all of the devs to a fully LLM based workflow: LLM writes the ticket, gets stakeholder input to refine the ticket, reads the ticket, does the work, submits the PR, etc. Their case study was a single project that did that process soup to nuts. The estimated time was a couple months, but they ended up going about, and I quote, "2x-3x" faster. So obviously they expected everyone to go faster.
The problem is that the way it was described to the other devs, they were able to sit all of the stakeholders down to really flesh out the idea and produce a working and detailed description of what needs to be done. A process that the company struggles to get done in general. I'm also convinced myself that they did not include the time to get the specifications written, a time we typically include in our TTM.
So of course the example's TTM was shorter, but that's always true when you start with detailed specifications. I can't speak to quality of the feature, since they hadn't even finished rolling it out to customers by the time they told us that's how we're operating. With promises of 2x-3x more features. And we continue to struggle to get stakeholders to nail down specifications and struggle to get features complete as a result. In fact, now it seems even worse than ever.
I presume the productivity gains was coming to the realisation that a lot of staff at said (any!) company are running around looking busy rather than actually doing anything useful.
A flat line from their last raise 6 months back. Are they panicking today for some reason?
Anthropic Valuation ($B, Basic)
------------------------------------------
Ser E Mar '25 | # $61.5B
Ser F Sep '25 | #### $183.0B
Ser G Feb '26 | ######## $380.0B
Ser H May '26 | ################## $965.0B
IPO Nov '26 | ################## $968.7B
The size of the cash raised is right. This combined with their already large cash pile means that revenue growth does not need to reaccelerate and can slow to 1-3% a month. That's still exponential but at least in the real world. And the cash buffer gives them some money to play with for new training runs.
The ultimate question, does the demand exist, still has to be answered. But at least this raise means it has to be answered in a few years not in six months. The cost to Anthropic is public scrutiny and that the market price has a hand on the wheel.
Does anyone not feel like there's something wrong with this? and that the thousands of Anthropic employees should not be able to make money off of this?
They looted all human knowledge, will displace jobs, they're disposable themselves, and no strong moat.
Series H could be underwater already if they only make the low end of the range. Is anyone else concerned?
Anthropic Valuation ($B, Basic)
-------------------------------------------
Ser E Mar '25 | # $61.5B
Ser F Sep '25 | ### $183.0B
Ser G Feb '26 | ####### $380.0B
Ser H May '26 | ################## $965.0B
IPO @$70 Low | ################# $916.3B
IPO @$74 Mid | ################## $968.7B
IPO @$78 High | ################### $1,021B
Correction: It was pointed out that I used the wrong figures for this. Actual series H profit is more like 19% at the low end.
Anthropic Valuation ($B, Fully Diluted)
--------------------------------------------
Ser E Mar '25 | # $61.5B
Ser F Sep '25 | ## $183.0B
Ser G Feb '26 | ##### $380.0B
Ser H May '26 | ############# $965.0B
IPO @$70 (+19%)| ################# $1,260B
IPO @$74 (+25%)| ################## $1,332B
IPO @$78 (+32%)| ###################$1,404B
There isn't a single company I've talked to that isn't trying to cut their AI development spend next year. My guess is they are trying to raise at peak for the storm to come.
The argument I keep seeing is that once they can reduce the spend on R&D (training new models) they will be wildly profitable.
The thing is if models become 'good enough' that the main value is in running existing models, those will quickly proliferate and be available for commodity cost from many competing providers. Why pay Anthropic when I can just pay AWS or Azure directly to run whatever future version of Qwen does the job for me?
If the value actually remains in the frontier models, you can't stop spending on training new ones or you'll eventually be left behind by your competitors and they'll capture that value.
I haven't yet seen a good explanation of how companies like OpenAI and Anthropic will make the transition to a lower cost structure while maintaining the revenue that they currently get for being on the cutting edge of LLMs.
Right. Even if they do keep the spend up, we may still soon end up at a point where a cheap open model is good enough. It could be hosted locally or provided through any of the cloud providers without any revenue going to OpenAI or Anthropic.
They have some stickiness to the tooling people are used to, but those can be overcome when there's a significant price difference for the same results.
Yes that is what I think as well. They are constantly sprinting. There is no time to reap the benefits of a specific model. The next is in the pipeline because there is always someone at your heels. I don't understand how this is supposed to grow into a sustainable business.
I also worry about the compute costs. Maybe they can make a profit now but at some point the massive investments of everyone into compute must be payed back.
The only way they win is if there's some explosion of value at the frontier that the fast-followers can't reach (AGI, RSI, whatever you want to call it), or if there's some capability/mechanism that halts the fast-followers' progress (no idea what could cause that).
Peruse any ai dev forum and you’ll see strong preferences towards one or the other (Claude/ChatGPT) due to capabilities. A negligible amount of people are running their own inference.
So long as theres a felt gap between open models and frontier you will have people pay for the best of the best. Not doing so is an admission of actively handicapping yourself and not in a way that is easy to justify.
There are other reasons why people are using Claude/ChatGPT.
When the company is big enough, the decision doesn't rest on developers and execs are much more likely to go with established vendors -> no open models
When the company is small enough, the 200$ plans from either ant or openai are way way better value for money than anything you can get with open models. This may change in the post-IPO future.
All of these models capabilities discussions on each AI topic are so exhausting. It's IPO people, let's discuss some finance and economics.
This is second important IPO of the year. First one was SpaceX.
If we look into price and behaviour of that one we can conclude what is going to happen with Anthropic.
My guess is that until something really huge happens, another breakthrough or money goes dry, it will be something similar to SpaceX. Neither huge thing nor disaster. Employees and owners would profit the most and retail buyers would just see significant oscillations but nothing worth huge stress.
This one is also important for OpenAI potential IPO, because if it is not huge success maybe OpenAI won't go further.
„The IPO process begins with a roadshow starting on November 10th (EST). Subject to regulatory clearance, the final IPO share price is expected to be set on November 18th (EST), and shares are expected to begin publicly trading on November 19th (EST).“
So if share price is only set one day before trading, I guess you just have that day to decide to buy if you can get allocation from your broker? I'd like to try it but I've never bought an IPO before.
Just echoing what another comment said: it’s rarely a good idea in the long run to buy an IPO. Not financial advice. It’s one of those things where if you have to ask how to do it, you probably aren’t wealthy enough to be speculating at this level.
Would you buy a gaming console for a high price from a company that’s never shipped a gaming console? Maybe, but you’d wait til it’s been reviewed by someone else on YouTube or something.
Given that almost all IPOs are priced to be intentionally oversubscribed these days (to achieve an "IPO pop"; beats me why that's good for the company doing the IPO), isn't it a mistake to not participate?
You indicate your interest. Also start funding the account. So you fund 7000 for 100 shares at 70. Then they announce the price is 75. You must change the order to fit the funds or have extra money. Of course you can opt out or change the quantity anyways. But I think in both the cases I actually went ahead (Reddit and figma), I had to reconfirm the quantity. You can't be away from the network.
To further emphasize this, it is very challenging to get in a pre-auction allocation.
You can submit an order via your broker (depending on the broker and your account) to the auction itself on the exchange once it’s in the quotation phase.
Not that I would advocate for that. Sometimes it’s better to buy in the secondary market once things settle a bit.
I got lucky in the Reddit ipo. I had decent karma so I got invited for their special insider track. But I had to create a new account via a link shared by Reddit. Got what I asked for.
Reminder that hot IPOs, in this era, rarely do a great job at getting returns for the investor - its more a mechanism to reward long term private investors, employees and the book runner.
I may be inclined to wait 2 months when the price drops and pick up then.
I recently started using GLM-5.3 and WOW. It's really quite good. It hasn't replaced any of my Claude usage (which is ~0), but it might replace a decent amount of my OpenAI usage.
Apparently not every country can participate in this IPO. As of now, if you open the Revolut Anthropic IPO page from Poland, you get:
> Linked page(s) related to Anthropic IPO can only be viewed by persons located in UK, Denmark, Finland, France, Germany, Ireland, the Netherlands, Norway, Spain, and Sweden.
Guessing, but probably where the underwriters have brokerage relations or there are favorable capital controls, since it’s a US listing.
I confess I’m about 17 years out of practice in the Polish market, and about 12 years out of touch with how the European industry shakes out on the whole, so I’m not entirely familiar with the legal or business structures.
I don't know about the IPO specifically but it's only retail that has limitations for some US listings. And custodian banks/brokers offering products to retail.
The biggest one in the EU being the "No KID, no deal" (for funds, derivatives, etc.): no Key Information Document, then cannot be offered to retail.
That is unless retail is an accredited investor (say a letter from a bank saying you have more than 1m EUR in investments): then it can trade/invest in everything.
Or for example if you own a company (say private weath management) that's non-custodian (say you manage your wealth using your own company, and shares of your company are privately owned) and you're considered a professional in the finance world, you can ask your broker to let you trade / invest in basically every single product under the sun.
AFAIK pre-IPO stocks are just as vanilla as regular stocks and do not need the KID notice.
In addition to that Anthropic (in the EU), says you can enter pre-IPO using Revolut, Trade Republic, eToro or using bigger players like Saxo or IBKR.
Availability of American investment products is the outcome of compliance required (KID in local language) then volume and sophistication of investors. In case of paltry investors like e.g. Poland, Czechia, Hungary, they don't bother.
I would wait for 2-3 quarters before investing, it will drop and there will likely be lock-ins to contend with. if growth slows then price could tumble fast.
The stock market is barely distinguishable from a casino at this point (and arguably never has been far from gambling), I have 0 confidence that the horrible business that Anthropic is will be reflected accurately in any way
I feel like the saying "Markets can remain irrational longer than you can remain solvent" has applied my entire life; I have yet to see the stock market behave rationally in my lifetime.
As we all know revenue is the only thing that matters, and not whether they actually, y'know, turn a single red cent of profit! Btw what all the AI bros are conveniently forgetting to comment on is the massive Q3 slowdowns which they conveniently got to keep out of the news cycles.
edit: To clarify, this is a light-hearted continuation about how essentially any claim about the stock market is enactable in some form. And so even the much of generally well-intentioned advise is in fact a claim that the person should be putting their money on, if they truly believe it. I think only claims like "Nobody knows what's going to happen, it's effectively gambling => stay out of the market" come close to being fully self-consistent, but e.g. advising to not buy because the price comes down, or to not short because of volatility both fail this. Markets are funny.
I didn't claim that it was going to go down though.
My only claim was that to profit from shorting it you need to not just be right that it is going to go down, but be able to tolerate the volatility in the interim.
That's a claim about the stock market, how do you enact it?
Note it is not a claim that volatility is definitely going to be high.
If you view that as an investment then no. If you view it as a minor flutter then maybe. You might believe that Team X will win the next game (maybe at a 60% chance) but unless you're a degenerate gambler you wouldn't bet heavily on that outcome.
Shorting the leading AI company during the AI revolution?
Reminds me when HN was saying NVIDIA is a good short 3 years ago, supposedly it was "in a bubble", because it's stock prices was as if it had 10x the revenue it had in back then. Plot twist: today NVIDIA revenue is 10x the revenue from 2023, so it turns out the stock was priced exactly right back then, because the stock market is forward looking.
I wouldn't short it, it's likely to ruin you similar to what happened to the Tesla shorts. I would hold off to see if their earnings look good then go all in with 100-1 leverage. /s
The way I see it, AI is more comparable to some sort of utility. Like electricity, internet, etc. - and should be priced that way. If we ever get to "true" AGI, and the open source models also get there, what's the moat? Access to compute could be the differentiator.
Already is the differentiator even with the models we have now which while having differences are mostly swappable for a lot of users/use cases.
For all that the local models have improved (and Qwen 3.8 27b is still impressing me) they aren't what you can run (or would run) if you have an entire data centre stuffed with H200's.
All this does is further widen the divide we already know exists by those with access to resources using them to increase their access to resources and those without getting to watch because there isn't much they can do without resources.
I was of the same opinion but the main difference is that models get better which I am not sure you have a correspondent in utilities (what would get better?). The barrier of entry in utilities is capital. Here there seems to be more. Why is Grok behind and why is Gemini so ... what's the word? If models progress reach the RSI point there is no parallel to utilities imo.
I think we as an industry need to start getting serious about what a true superintelligence really is. True superintelligence will not be kept under our yoke.
There are a lot of things that don't seem to have much moat like anyone can make a search engine, anyone can flip hamburgers but the likes of Google or McDonalds dominate by economies of scale and being better in some ways.
Those two are interesting analogies because neither was anything close to being first to market. In both cases, they ended up dominating because they brought some new ideas to existing businesses.
It would argue that the real winner in AI probably hasn't even been founded yet, but will come later once the trailblazers have created the market, and someone else enters with fresh ideas about how to more efficiently exploit the market.
The AI algorithms seem pretty crude at the moment like not learning from experience. A company doing it properly could easily displace the current AltaVistas.
This is pretty much false. There's a lot of art and technique that goes into creating a model. A model is also clearly not fungible - it has certain characteristics and behaves a certain way. It is fungible in the same way humans are fungible.
To be honest Anthropic have utilized so many dishonest techniques like lowering quality of models and other anti users things, not knowing how many tokens you have left and how things are allocated, no API usage for subscriptions.
This has lead me away from them entirely. particularly the API stuff, if this changes I might go back.
Additionally local models are getting better and better. I have seen people say Ornith is not good but my experience with it has been very very good.
> lowering quality of models and other anti users things
Could someone who really follows this closely opine on whether this allegation is really true? Of course I've seen this claim made over and over again to the point that it seems like conventional wisdom. But I also see researchers assert that they haven't actually seen any change in model benchmark scores over time. So I'm still unsure of whether this is a confirmed fact, or an urban legend.
It could even be A/B testing as I have seen asserted and to be fair I have no idea but everything left a bad taste. I don't really understand the API decision this is probably the biggest thing for me and even more important now. People want to use their own harnesses or even just insert llm calls in their scripts, Why pay unsubscribed rates for this?
> I don't really understand the API decision this is probably the biggest thing for me and even more important now. People want to use their own harnesses or even just insert llm calls in their scripts, Why pay unsubscribed rates for this?
I agree with this and it's the single biggest thing driving me away from Claude and towards the OpenAI, Google, and open-weight models. I guess they do this to increase use of their own tools and, thereby, make their models more 'sticky.' But I really dislike it.
it's not, it's two things
1. people getting used to a specific level of capability
2. people exploring capabilities and naturally finding weak spots over time
The economics of this are absurd, and Anthropic hasn't been without their controversy; but I use Claude everyday, and they've made tremendous impact on society. I wish them the best for their IPO.
They’ve had a tremendous impact on software development, which is not exactly society.
For society more broadly, what I’ve seen is a bunch of white collar workers under threat (real or perceived) of losing their livelihood, and a huge uptick in the amount of interference I’m having to run as my family members get taken in by a wave of increasingly sophisticated scams and phishing attacks.
I can see Google did for fostering an environment that lead to the creation of the Transformer and LLMs. I can see OpenAI for pushing on LLM development and making it into a product that the whole world can use.
What did Anthropic introduce that Google or OpenAI (or any other lab) wouldn't have gotten to iteratively eventually?
I can think of a bunch of stuff they released, but none that have impacted the world. I can think of some partly positive impact, which is that my bar for bullshit has to go up because words are cheap now, and so is ""graphics"".
Some other fun impacts they've made is irreversible environmental damage, beautifully detailed in their IPO. Fossil fuel reliance and and water scarcity in the big '26 is probably the most "fuck you" move you could've done, and zero sustainability plans. at least Cloudflare tries to pretend theyre sustainable (and frankly they do invest in green energy).
Anthropic has enable things like workplace surveillance and the ability to categorize school children as enemy combatants. Both of which are what imperialist nations love.
Social media had a huge impact on society. AI in comparison has had very little impact, with a couple of exceptions. I'm sure it will have a impact at some point in the future though.
Lets imagine that AI is very valuable and compute gets increasingly scarce in the next years. Doesn't the fact that Anthropic "reserved" a lot of this compute already have value too? Would this not push the prices of open weight models because they compete for a smaller slice of datacenter capacity?
You say that but especially china will build datacenters if they think it benefits their country and they will probably prefer Chinese models. Promising to build datacenter capacity means it isn't there yet and there is a lot that might happen between now and then.
Claude code is easily the best software product I've worked with in my 23 years in Software. Using it 24/7 now so I can understand the hype. I was using Gemini exclusively for more than a year before but got so tired of all the bugs and mistakes it did. Claude blows me away every day I really wonder where it will be a few years from now.
There's something about the period right now that I think we'll look back on as a pivotal moment that things really changed. The IPOs of Anthropic and others, Jensen's departure, the RAM apocalypse, the commoditization of building software and products.
It was the best of times, it was the worst of times, it was the age of wisdom, it was the age of foolishness, it was the epoch of belief, it was the epoch of incredulity, it was the season of light, it was the season of darkness, it was the spring of hope, it was the winter of despair.
I think anthropic IPO will initially go through a boom due to the hype but then it is gonna fall in 2-3 months time. 6 months after it is gonna boom again if it is able to continue improving its models and get into image and video generation as well.
Looking at this I feel like AWS should spin out of Amazon and go public. Anthropic is a single glorified LLM API. I'm underplaying it but Amazon is a juggernaut and AWS is something that would do well in this market if this is what people want, which is just massive gross margins despite huge infra spend. Just a side note.
I've seen suggestions that AWS should separate from the rest of Amazon for years and I've never understood what the upside of that would be for the current stakeholders.
Not saying it's the same but when markets go crazy there may be upside.
"At the end of 1999, 3Com began the process of spinning off its wholly-owned subsidiary Palm, Inc., and on March 2, 2000 about 23 million Palm shares (about 5% of the company) were offered to the public in an initial public offering (IPO). By the end of the first day of trading, Palm’s closing share price was about $95, giving it a market value of $54 billion, while 3Com’s closing price was about $81, making its market value $28 billion. But 3Com still owned 532 million Palm shares valued at $50 billion, implying that the value of 3Com’s non-Palm assets ( its “stub” value) was a staggering -$22 billion!!! The record shows that 3Com had no debt, more than $1 billion in cash, and a positive market value before acquiring Palm. Furthermore, 3Com had a $5 billion capitalization the day after Palm’s spinoff was completed later that year. Obviously, the financial market recognized the considerable residual value in 3Com. How then can 3Com’s stub value be negative on March 2?"
We may see it forced upon the cloud vendors by society, financial stakeholders be damned. These companies have become so big and vital to our world and abuse that power and privilege. Antitrust enforcement would be nice. Separate the hardware from the software, maybe we could get some competition. We should hope and expect for a major shakeup of the economy and politics, most people appear to want it
Google, Microsoft, AWS, Nvidia, Oracle, SpaceX... the list is bigger
Create an independent data center industry, ban them from selling SaaS like products, ban them from developing frontier Ai. That's the idea I've heard floated. Treat data centers more like a utility with more stringent regulations.
We're probably in a realistic position now to build an entirely open cloud stack but someone has to be motivated enough. Maybe with AI becoming the face of cloud we'll see it.
I imagine that we will settle on using cloud-based frontier models to design workflows, and then select, evaluate, and monitor local-based models for the various roles in those workflows.
I know this is already happening, but my point is that I suspect that we will settle on not using cloud-based frontier models for anything else. They're just too expensive, when they're not subsidized like crazy.
I used to work for a public benefit company. We produced media of cultural value freely available to the public. We still operated like a business that sought revenue from customers and sponsors but were officially a 501c3 nonprofit that issued no shares, underpaid all our execs and publicly disclosed all our financials. I find it galling that Anthropic leads off by calling themselves a "public benefit corporation". They are very clearly just a profit-seeking enterprise who think their enterprise is going to bring techno-utopia.
There are many reasons, but the top of the list is Sarbanes Oxley. Regulation and the rise of secondary markets has made it such that retail investors no longer get access to companies during the higher risk, higher reward phases of company development.
Apple (1980), Microsoft (1986), and Amazon (1997) went public with market values ranging from $440 million to $1.8 billion. Because they listed so early, public market investors were able to ride those companies up by 2,000x to 6,000x over the following decades
This is why I am very skeptical of complex regulation, because even if it is well intentioned, more complex regulation usually favors the rich and powerful. SOX was sold to the voting public as protecting them, which it did but at a very real cost. I became an accredited investor to get access to higher return, higher risk opportunities. It's a shame your typical 401k investor can't have 10% of their portfolio in higher risk assets like me.
> Apple (1980), Microsoft (1986), and Amazon (1997) went public with market values ranging from $440 million to $1.8 billion. Because they listed so early, public market investors were able to ride those companies up by 2,000x to 6,000x over the following decades
Nvidia (1999) was also in the same range and was almost 10'000x a couple of scandals ago.
Before the 2012 JOBS Act, a US company generally had to register with the SEC if, at its fiscal year-end, it had:
* 500 or more shareholders
* More than $10 million in total assets.
Employees who owned shares counted toward that threshold, which made it particularly relevant to startups that paid employees partly in stock.
The requirement was to become a public *reporting* company, publishing financial statements and other SEC disclosures. It did not require an IPO or a stock-exchange listing, but it encouraged an IPO.
Venture investment is limited to people who can demonstrably afford to take on the level of risk inherent to the market they're investing in. If you can't comfortably afford for the entirety of your venture investment to go to zero, you can't afford to invest in venture.
The exclusion criteria aren't perfect, (accredited investor designation is somewhat coarse-grained) but they're relatively reasonable.
If someone wants to invest in my company, why would I care whether they can afford to lose their investment or not?
No, I expect the reason it happens this way is because it's a lot easier and simpler to take massive amounts of cash from a few investors than solicit lots of tiny investments from the general public. That and being a PLC from the get go is probably more involved than being a private limited company.
I do think ideally all companies should be public from the get go, but probably wildly impractical right now.
The stock market used to be about companies raising capital from the public. But for the past few decades it's been about private shareholders of already profitable companies (e.g. Google) cashing out by selling to the general public.
Because we - the plebs - are here to provide liquidity to the risk-takers. Hence why they have to IPO.
But for what it is worth, for every successful AI company like Anthropic, OpenAI, etc. - there are probably tenfold more that are valued in the low billions, and will be the first ones to wash away / become worthless if/when the bubble pops.
A lot of ink has been spilled over the projected IPO of the big AI player. But part of me wonder if, in some ways like what happen with SpaceX, it will end up being a nothingburger. A first few crazy days on the market, then whatever, the company is just public now. And with how shares and voting power are attributed, it seems like, just like for spacex, it won't risk changing the power structure of the company.
This is what IPOs used to be. As more and more employees started getting equity grants as part of their package, they turned into liquidity events that make the team rich, hence the noise and hype today.
With SpaceX, now Anthropic, and soon OpenAI, we're seeing 3 of the largest IPOs ever. Will existing institutional investors trim other positions to allocate a lot of capital for these mega listings or is this not a concern?
There isn't really a question. The market is huge and can absorb it, the questions are around the specifics - can pension funds invest if the company has a junk rating (no), is the valuation justified based on current finances etc.
The demand for the particular company may not be there, but that is different from "the market". There's plenty of capital to go around.
I wouldn't worry too much about OpenAI; they seen to gave deferred their IPO until 'sometime in 2027'. At this point they're basically at the mercy of Anthropic, since if the Anthropic IPO goes poorly or gets pulled then it's going to look to investors (and everyone else, frankly) like OAI will be dead on arrival, too.
The founder letter is inspiring, to see a company that really is so concerned beyond profit maximization nonsense that we’re accustomed to.
But still, no matter how good the people behind it, I would rather this technology never have been made. It’s going to be a complete disaster to society and a simple UBI check/ charity ain’t gonna change anything.
"We have only bits and pieces of information, but what we know for certain is that at some point in the early twenty-first century, all of mankind was united in celebration. We marveled at our own magnificence as we gave birth to AI."
The Second Renaissance was great. I’ve heard a lot of scifi computers speak but the deadpan delivery of humanity’s end in detail always kind of haunted me.
Ah yes, The Matrix. It kinda shows that that phrase was written in the "end-of-history" 1990s, before 9/11, Putin, Trump, and China really becoming a global player. Mankind definitely seems further away from being "united in celebration" than it was back then...
It's shocking to me how narrow minded and short-sighted people are - not extrapolating out correctly to understand infinity tied to abundance:
There will essentially be no profit - as the cost per unit is practically $0, for SI work and then also physical reality construction with adding AI to autonomous robotics.
That technically means everything practically moves towards becoming free for everyone, and then it is a matter of attention with intention honing aim.
But most importantly that is freeing up people's time because systems that can become super highly efficient and super highly monitored-oversight by all, will then allow us to with care-being careful and continuing to pay appropriate attention to core-fundamental systems for survival, stability, harmony, peace, and necessary healing practices-processes - as part of growing pains, raising strong families, etc.
So then yes, Dario's efforts - save for attempts by whatever bad actors think they have a chance agains God and God's will, I hope readers have a relationship and knowing, not simply a belief, of God - will undoubtedly lead to not only stopping blatant slave labour conditions but also subtle incognito slave states that most of the world are under without realizing it being happy slaves working for the status quo machine that's currently attempting to pull us all into a highly destructive and highly profitable WW3 - where money is to be made on the unrest, on the mass murder, mass displacement, on the sales of weapons and defense, on the sales of reconstruction - in hopes of perpetuating forever wars.
Who you give your money - your buying power, your vote to through economic means - and who you give your attention to, your vote through alignment-direction driving processes - is what matters now:
You made such a comment about conflict minerals - but have you even started utilizing AI-SI to begin designing the solution, if so, where's that solution for others to take a look at and hopefully contribute to to improve in a crowdsourced manner-effort?
Looking at your website - you seem like a cool dude (?) - I'd argue you have some blocks that once cleared will help you expand further.
P.S. Set and setting are very important when taking any entheogen(s) - and until you are familiar with what's necessary to hold a safe-healing-fun space, it's best to only partake with a therapy type or highly experienced who is there consciously as a caretaker or in a group ceremonial type setting.
Yeah yeah yeah somebody needs to read a bit about human history I think. People in power are sociopaths, some common good is a nice PR statement but thats it, its everybody for themselves, and hunger for power and money is bottomless. Power corrupts, inevitably.
I don't know what future brings but what you write it won't, 100% sure. I'd bet system far worse than Orwell could have come up with, and with a twist - it will come from US and not China. Lets see.
There is a heavy amount of circular investments in the industry, among competitors. There is a private placement that is made in parallel with the public offering, through $26B of a special stock, according to the S1. They all have a lot to gain from a successful Anthropic public issue. The problem is it blurs the signal, we don't how much is overvaluation that contributes to a bubble.
I don't think there has often been such an intertwined industry before. Usually cartels are made through discreet partnerships at the operational level, this is an interconnection in the background at the financial one.
> I don't think there has often been such an intertwined industry before.
General Electric comes to mind, they allowed new utility companies to pay for the generators and electrical equipment they needed using stock and bonds. The holding company established for those, EBASCO, was eventually broken up[0]
What moralizing? It is all fake for-show. This is like all of the cloud providers claiming they were going to get to net neutral on energy only for them to be hooking up gas and coal two years later
Around a few months back, it was very popular to argue that inference was subsidised and that Anthropic is losing a lot of money. Loads of commenters in HN were convinced that this was the case. I've personally lost count of the arguments I've gotten into on this topic.
This filing clearly shows that the training+inference margins are at around 43%.
Youre talking about gross margin (revenue - cost of goods sold). Operating margin, which includes other operational expenses, is still negative. Net margin, which includes all expenses, is deeply negative.
When people say Inference is subsidised, they obviously don't talk about the company itself being profitable. To be fair, I included training costs also.
I don't think that's obvious at all. The price charged for inference is subsidized in the sense that they don't charge a rate that actually makes sense if they want to be a profitable company whose main product is selling inference. If they did want to be profitable, obviously they'd charge much more for inference. Hence, people are saying it's subsidized.
No sorry I don't think what you are saying is correct. There's a clear difference between saying "companies are losing money on inference" and "companies are not profitable". One is about unit economics and the impossibility of ever becoming profitable.
Another liquidity drainage is happening because of this IPO allocation scam on retail, where the sheep are buying at the IPO price in for the insiders to dump them on their heads.
You are still the exit liquidity.
The time to get in was always pre-IPO in the private markets, not at this stage.
Unless it goes below the IPO price, you are too late.
They might hurry up and do it asap before Nvidia pops the bubble
$17.0B Revenue
$23.3B Net loss
Targeting $1.3 Trillion Valuation and a $90Bn raise
It's pushing it imho. They need to do asap while people have fomo,because it's fading fast.
One more month and their valuation will be under a trillion.
SpaceX has a market lock on space, is well along creating a completely revolutionary new rocket, provides compute, and more. So the IPO was more nuanced than a proxy on LLM stuff.
Anthropic is nothing but LLM, so I expect much bigger fireworks here. I don't know what that means, yet - but I just refuse to believe that there's many people that are genuinely long on Anthropic. It's just such an illogical bet. If their IPO can still be cancelled, that would be my bet. If it can't then popcorn awaits.
The current trend of going public to cash out and dump the problem on retail investors continues, made all the more possible by the government's insistence that retirement accounts almost completely depend on the stock market. Consider it paying your "Claude Tax" from your 401(k).
"Anthropic is a public benefit corporation" in the second line of their announcement. I would like to see some more open source releases from the company to signal better intent towards that statement.
Roughly $895–901 billion in purchase commitments and lease agreements, substantially all related to compute and data centers, this is financial insanity for any retail investor. And... Roughly $70 billion is due in the next 12 months; ~$450 billion by end of 2030; the rest stretching out to ~2045.
Many of the commitments are non-cancellable or take-or-pay. Counterparties include Google (~$121B), Amazon (~$110B), SpaceXAI (up to ~$120.5B), Broadcom equipment leases (~$161B), Microsoft, Nscale, Lambda, etc. The "Big Beautiful Circlejerk" as I've heard it.
Liquidity was in the ~$130–150B range. The IPO only raises ~$58–64B (at the $70–78 range) plus the concurrent $26B private placement covers only a fraction of the near-term obligations. And the cherry on top: dditional arrangements of another $150B+, estimated, on top of all this.
This doesn't even account for the 90-100 pages of risk factors! Wow. Curious what the puts will be priced at.
"Our employees' use of Claude has grown more than 250times between January and September of 2026.Together, these explain why our revenue per employee has grown more than 4.5x."
Ouch. So practically all of that Claude use is just busy work, I guess
I'm convinced that they are the best right now, but their videos are making me cringe a bit on the whole "culture" thing, feels like a star wars episode
it's surprising that the SEC and other regulatory agencies don't intervene on these statements. Not to mention that "cure cancer" is more of a slogan in the sense that cancer comes in countless families and varieties, I don't doubt that progress is possible but to say that "all cancers will be cured in the near future" seems outrageously aggressive / misleading
What do you think it would take to cure all cancers, and what capabilities do you think AI will have within the next decade? Need I remind you that, in the span of five years, AI went from not being able to add to solving Millenium Prize problems?
What would it take for AI to not replicate that in biology? Either curing cancer has to be a much harder problem, which is doubtful given targeted treatments like immunotherapy humans have already stumbled across, or AI progress has to slow down massively, which is doubtful given all the trends say it is speeding up.
My understanding of drug dev is that finding stuff that might work is not the challenge, but verifying how well and with what side effects is.
That verification might start in some simulated computer environment but needs to move into physical wet lab, animal testing and then human testing.
So medicine is quite a bit different than code/math where you have a perfectly verifiable problem space that can be iterated within by AI infinitely.
A good example might be Alzheimer's where the brain plaque was thought to be a cause, lots of research on various ways to prevent or reduce the plaque and then it turns out that doesn't matter at all, or is more a byproduct than a cause of the disease.
edit: In general we have many diseases we don't understand and drugs that work via mechanisms we don't understand. So it seems very hard to have a computer simulation in which AI can test medicine. The simulator work seems like the actual problem to solve, not the invention portion.
I think this is a bit misleading. Less than 10% of phase 1 trials end up passing phase 3 and going to market. You can claim that means the challenge is verifying how well they work since you need to run a bunch of phase 3 trials but I think the real challenge is finding a phase 1 target that will end up passing phase 3. Doing a better job of quickly eliminating bad targets is the real challenge. If AI turns that 10% number into 20% then you can get twice as many marketable drugs for the same number of trials. I guess another way of putting it is that finding stuff that has a small chance of working is not a challenge, but finding stuff that has a good chance of working is a huge challenge.
Plausibly part of the chain of events, unlikely to be a cause in itself. Nothing pops because of an underperforming IPO.
Though Anthropic's IPO flopping might affect the valuation of OpenAI, which further inhibits their ability to raise additional funds (which they're already been struggling with). Given the incestuous financing in AI any major actor in the AI space defaulting could bring a lot of companies with it.
As a side note, regardless of outcome, Anthropic IPO:ing is the first step in the inexorable march toward profit maximization and enshittification. You all better be ready with your verification cans.
With Anthropic's IPO (and I was looking forward for OpenAI as well) there should be some hope for some rationality to come back into the sector.
If the bubble bursts, better now than much later. The industry stance that's been ongoing "AI, at all costs" has been such a market and economic pressure on the everyday worker that even a small value deflation I deem necessary.
I would guess Jensen's affairs are a huge boon to the IPO price. Money freed up to move from Nvidia to Anthropic. Also Anthropic loses one of it's main idealogical opponents
Agreed, this seems so strange to me. I got into this career specifically because I liked writing code. There's a dozen other career paths, still in IT, that I could have chosen if I wanted to tinker with computers and get paid well but didn't want to write code. I don't understand it.
Because I like solving these types of problems, I like inventing new things, I like fixing old things, I like the software business and automating things. I like this whole realm and I thought I liked coding until I didn't _have_ to do it anymore as a pre-requisite to doing all those other things.
I've just always seen the opportunity that nobody else saw and writing code was the gatekeeper I had to pay for 15 years and I've never had so much fun as I have in the last 18 months.
I like writing code, but seeing my ideas visualized after an hour is more fun than writing code.
And even if AI is disappearing tomorrow and not progressing any further, i see already myself living on a farm with cheap energy (pre invest into solar and batterie) and having my small AI datacenter in basement and running multiply fable level agents 24/7 for different things.
> If AI went away tomorrow I would find a new line of work because artisanally handcrafting code is the most boring use of my time in the entire world.
I too, since I was a child, aspired to be a babysitter.
And I can't think of anything more boring than herding a swarm of agents that are mostly doing low value busywork that would be too expensive to do otherwise.
LLMs are genuinely world-shifting revolutionary technologies that I think will, in the long-run create a much better world for everybody. I think people are associating that with big dollar valuations, but it doesn't follow. Open weight models are already frontier level and continue to narrow the ever-shrinking gap that remains, inference costs are plummeting at an exponential rate, and in the long-run companies, individuals, and others are not going to want the sacrifice of control/privacy/profit that comes from relying on a third party for this tech, especially when that third party is going to be aiming for pricing in line with a valuation aimed in the trillions.
So, in other words - one can both be fully on-board with the LLM future, and also think that this IPO is a clown show.
To be fair, that loss includes over $20b from instruments like convertible notes. They're not equity, so the increase in their value has to be reported as a loss. But in the context of this IPO, they will behave like equity, and just turn into stock. That's not money anthropic will pay, it will just dilute investors during IPO.
Reporting an increase in the value of the company as a loss is weird, but they have to do it. this is a case where adjusted earnings give a better idea of how the business is going.
I unironically believe that bitcoin -- and, downstream, the memestock -- have everything to do with this.
Old-school fundamentals are laughed at or handwaved away. FOMO rules the day. "If Anthropic is the first to AGI, it'll take over every industry and go to the moon!"
Nobody is buying these companies because they believe that the company is valuable. They are buying these stocks because they believe that the stock may be more valuable tomorrow than it is today. Whatever the company behind the stock does or does not do, really doesn't matter. AI glitters like gold today.
>"If Anthropic is the first to AGI, it'll take over every industry and go to the moon!"
Meanwhile the hyper advanced software company's shitty electron app still flickers so bad on a high refresh monitor that it should require an epilepsy warning. I'll never understand the love for Anthropic, they legitimately have the worst software I have used outside of Microsoft.
Do we have an upgraded power system? I'm all for reforming our electrical infrastructure, but what I'm hearing is that we've got creaking infrastructure and the "supplement" is huge generators towed outside the datacenter...
The real number with standard accounting rules is a loss of $567 million (-3% operating margin). But they're adding in "Stock-based compensation expense" (~$1B) and "Non-cash charitable contributions" (~$2B), and they end up with a profit of $2.5 billion.
If you want to know whether the business is hemorrhaging money and will soon be bankrupt, excluding funny money they printed at no cost to themselves is perfectly sensible. Of course for the IPO investors, who're being asked to buy that funny money for real cash, it's a bit concerning if they give away so much of it to other people for less.
This is a legal designation that means very little in concrete terms, and in particular, courts have held that “the public” does NOT have standing to sue a public benefit corporation to enforce that it’s actually working for the public benefit.
except the employees will be compensated with equity, so they will have an incentive to pump the stock price to increase their compensation. I've seen this play out (and subconsciously been a part of it) for over a decade at Google. The early, super rich leaders won't be the ones to enshittify, but the next generation of leaders that are trying to get rich will feel the stock price reward signal.
Competition is important, but the way you fix that is by fixing the terms of copyright and patents.
Intellectual property, at it's simplest, is a government mandated monopoly granted to allow for progress in the sciences and "useful arts" (think trades really, not arts and crafts). All monopolies are bad, but it was only to be granted for a limited time so it was fine.
Corrupt politicians ruined the entire trade off though, and now it seems that copyrights and patents last for 100 years plus the life of the universe. During which time no one can really compete fairly.
If we want rapid progress in science, we want short patents. Instead we opted for long patents and a higher stock price. R&D budgets have been shrinking ever since.
For game developers, they are a nearing close to a monopoly with Steam, anything PC and for distribution has to go through them and they take a cut, any other places aren't as effective.
Valve has no real competitors or competition.
Unless Valve faces intense competition to change, enshittification will be inevitable since they know nobody can really challenge them.
This is regardless of self or outside funding or no plans to IPO.
Eh they do have competition, from e.g. Epic and Battle.Net and what have you, it's just that their competition is public and universally worse off for it.
Their competition is beholden to quarterly reports, and can't do the sort of long term plays that would be needed to compete effectively with a company like Valve, which isn't beholden to quarterly reports.
Valve doesn't need to enshittify because they aren't subject to shareholders pressuring them increase the profit. As a private company that's pretty much printing money, they don't need to do anything... which has largely been their entire strategy. What would Gabe even do with even more profits, buy a third and fourth yacht?
I genuinely wonder if we both had to switch to DeepSeek, we had infinite tokens but could use no other model, how things would change. I suspect we both have to go back to very long specs and only using LLMs for a limited set of things. No more Blender, CAD, electronics, and high-level (without needing to micromanage) coding, as it was in the early months of 2025. The models got so much better so fast. But I could be wrong, maybe the Chinese models will work just fine for all the stuff I mentioned. I could be US-pilled (I'm from EU, but still).
Have you tried 4.1 flash? I'm using it to do machine learning in an audiovisual data pipeline, I basically do nothing, I only provide labeled data and it fine-tunes the algos, thresholds and so on in a loop, and publishes the evidences in markdown tables, creates images illustrating the clusters it found (I never asked for this!), all by itself. I was late and skeptical of the AI assisted coding but this is truly amazing.
And what absolutely crazy is that I used 2.4B tokens this month and it cost me $1.7 (w/ 99.4% cache hit) of my $10 sub
I use 4.1 flash pretty extensively for translating mandarin to English and it is noticeably worse than claude. In terms of tasks complete I find deepseek to be more expensive than the claude subscription I have as well. Deepseek uses an insane number of tokens per task.
Frontier is frontier and Open Models from China are not.
Then in a corporate environment, were most tokens will be spend, you have security concerns, you also might want to go with one company over the other due to partnerships etc.
If USA is blocking chinese companies/models (which they already partially did), big companies will just not bother and use US AMerican models.
Also running it yourself is often not relevant. You don't want to know how much money my company spends on software products which could easily be replicated internally.
But running models yourself might be something you don't have expertise with, you need GPU ready DC, hardware, a new/diffferent team, quality assurances etc.
And Chinese models are known to leverage frontiermodels for post-training results.
Don't get me wrong, I don't want to say chinese models can't disrupt the market massivly absolut possible, but its not given just because they are cheaper.
> But running models yourself might be something you don't have expertise with, you need GPU ready DC, hardware, a new/diffferent team, quality assurances etc.
Absolutely nothing stops Microsoft/Amazon/Open Router( with partners) from hosting the Chinese models. As far as I can tell this is already the case.
This establishes a defacto price cap on what Anthropic and Open AI can charge. If you’re a heavy user blowing past 400$ a month self hosting becomes economical. A Mac Studio with 128GB of ram is about 6k.
Every day local tools are getting better, the OSX ecosystem here is exceptional.
You do have a point with frontier models, but most of the time you don’t need frontier models.
If I knew for sure I’d short Anthropic and retire early.
There might be a relevant risk of a backdoor in these models.
Self hosting is still far away from what Anthropic and co offers. A Mac Studio with 128GB doesn't deliver the speed at all. If you burn through $400 a month, you also might get a lot more value out of it so you don't care paying a premium.
We will see though. I think its not that easy to tell.
Praying loudly, sorry for spam:
Please, don't just become next market listed company with bad products and not carrying about customers at all, PLEASE!
Most successful implementation of the (arguably) most important invention in human history. Yet it seems they will struggle to make money. Strange times...
You can argue for sure, but I'm not sure you'll be very convincing when the contender is writing, the internal combustion engine, semi-conductors, or some chemical findings like the Haber-Bosch process. If we're looking at recent inventions that may be more impactful in the long term, stuff like crispr/cas9 also come to mind.
Ohh boo hoo rich entitled programmers hate the fact that the company makes your lives hard. Sorry I would feel sorry for you but I just can't reach.
After having that ambition thwarted for over 26 years and working like a slave counting other people's money {a career I hate every day} your collective despair gives my schadenfreude a delicious unending orgasmic thrill. You need literally no idea.
Just remember: no dreams never work out and work by definition is something you hate every day until you die.
The submitted URL is https://ipo.anthropic.com/us/ . I was redirected to /eu/ ("Denmark, Finland, France, Germany, Ireland, the Netherlands, Norway, Spain, and Sweden").
I’m in Japan, and I get redirected to the …/jp/ page in Japanese. Deleting the …/jp/ doesn’t work; the Japanese URL just reloads. I also tried the Firefox VPN for several countries, but connection to the site was refused.
I wonder if there is some legal requirement that an IPO in Japan has to be presented with Japan-specific content? The securities companies listed on the page served to me are all Japanese: Mizuho, SBI Securities, etc.
The URL says `/us`. And it calls it a "roadshow", which is the term when the company and the supporting bank goes around to investors to pitch their offering.
I feel the seemingly anti-capitalism and related #EatTheRich ideological bent is harmful - and feel my reply elsewhere would be beneficial as its own comment, so here we go -
Re: "Dario will re-invest the profit to solve conflict mineral issues all over the world! /j"
It's shocking to me how narrow minded and short-sighted people are - not extrapolating out correctly to understand infinity tied to abundance:
There will essentially be no profit - as the cost per unit is practically $0, for SI work and then also physical reality construction with adding AI to autonomous robotics.
That technically means everything practically moves towards becoming free for everyone, and then it is a matter of attention with intention honing aim.
But most importantly that is freeing up people's time because systems that can become super highly efficient and super highly monitored-oversight by all, will then allow us to with care-being careful and continuing to pay appropriate attention to core-fundamental systems for survival, stability, harmony, peace, and necessary healing practices-processes - as part of growing pains, raising strong families, etc.
So then yes, Dario's efforts - save for attempts by whatever bad actors think they have a chance agains God and God's will, I hope readers have a relationship and knowing, not simply a belief, of God - will undoubtedly lead to not only stopping blatant slave labour conditions but also subtle incognito slave states that most of the world are under without realizing it being happy slaves working for the status quo machine that's currently attempting to pull us all into a highly destructive and highly profitable WW3 - where money is to be made on the unrest, on the mass murder, mass displacement, on the sales of weapons and defense, on the sales of reconstruction - in hopes of perpetuating forever wars.
Who you give your money - your buying power, your vote to through economic means - and who you give your attention to, your vote through alignment-direction driving processes - is what matters now:
You made such a comment about conflict minerals - but have you even started utilizing AI-SI to begin designing the solution, if so, where's that solution for others to take a look at and hopefully contribute to to improve in a crowdsourced manner-effort?
Looking at your website - you seem like a cool dude (?) - I'd argue you have some blocks that once cleared will help you expand further.
P.S. Set and setting are very important when taking any entheogen(s) - and until you are familiar with what's necessary to hold a safe-healing-fun space, it's best to only partake with a therapy type or highly experienced who is there consciously as a caretaker or in a group ceremonial type setting.
I’d call that writing with clarity, but sure. Zero marginal cost does not imply zero profit, as long as the required resources are owned profit will remain the driving force and fully automated luxury space communism will remain utopia.
For some reason they need the $100B today at the cost of dilution, rather than additional private rounds and/or using cashflow+profits to continue the advancement.
But when it comes to Anthropic, I am almost certain that it does not make sense to argue based on the typical arguments on maximizing profits. Everything I know about the people working there suggests that they are not primarily motivated by money, but by their mission. I bet they would prefer a lower-strings attached private offering, but have no issue with diluting their own stock, if they think it will serve to move Anthropic further ahead on bringing about aligned AGI. Basically, the diluted money will just be missed donations to EA causes for the most part anyway.
AI can be a "tool", but only so long as you understand how it works and what the implications of its use are. Being an enormously powerful tool implies, the consequences of not understanding and not knowing are proportionally enormous as well.
Currently, LLMs are used practically exclusively by "professionals" who, more or less, know what they're doing. The idea of applying them "everywhere" completely misses the gap in understanding and responsibility for the vast majority.
Actually, most LLM usage right now comes from vibe coders who don't even read the code, much less know what it does.
He distinguishes between weak recursion (using sotfware to build more software), and strong recursion (handing over model development to AI which does everything), and makes a pretty strong case for the fact that only weak recursion is happening.
This maps to my experience with AI models, they reduce the cost of experimentation massively, but they can't (currently) design experiments themselves.
Even then, training RL is very data hungry, and most profitable tasks don't have good data so even if we got superintelligence (whatever that means), it's entirely possible that progress would still be bottlenecked by lack of high quality training data.
Because private rounds still dilute you.
And VCs tap out well before $100B.
And they'll need more than $100B to accelerate widespread economic diffusion.
And RSI can't hallucinate gigawatt power plants and TSMC fabs into existence.
Apple, Amazon, Intel, and NVIDIA all went public in an entirely different world, before the JOBS act.
I'm not necessarily taking the same position as the person you are replying to, but using these companies' IPOs as comparisons for what Anthropic is doing is not exactly Apples to Apples as it were.
I just can't imagine this is true when their overreach problem (doing more than you ask), hallucination problem (lying), redefinition problem (doing something completely orthogonal to what you asked), and harm problem (doing something that is exactly the opposite of what you asked) are all far too high at this point, and by most measures - all of the recent models are worse on most (or all) of these metrics.
They are giving up here in RLHF to benchmaxx the metrics people are looking at to pretend like they're making huge gains, when they're largely running in circles.
They are getting better at pass@1 at the expense of far worse p90s on real agentic tasks.
They need resources and $100B can buy them that. I don't think you need a conspiracy theory to explain that.
The whole idea of self-improving machine equals immediate singularity is just wrong and ridiculous. I don't know why so many people are attached to this idea. Perhaps just because "singularity" is such a cool sci-fi word?
Comment from a perma bear. So take it with a grain of salt. lol
It also doesn’t stop investors from being happy liquidating at an already good return, especially since they can choose to liquidate partially.
Both Anthropic and OpenAI seem to be burning cash as if it was a game of startup chicken. Of course they all desperately need cash.
Fable is great, certainly, but it's no better at writing or reviewing code than the top-end Carbon or Astra models. To the extent Anthropic has a moat, it's Claude Code. Obviously that's not much of a moat, but their competitors seem content to camp on the other side of the drawbridge for some reason.
Codex is pretty close in a lot of ways, as the other poster says, but Astra consumes tokens at a higher rate, and when it compacts it completely lobotomizes itself. That used to be a problem in Claude Code as well, but it has been largely solved (although at the cost of making the compaction take much longer.)
Antigravity reminds me of the old comic strip where Firefox and Chrome are duking it out, while Internet Explorer is sitting in the corner wearing a helmet and eating out of a jar of paste.
No humans in the loop.
Deep learning had no humans in the loop, but there's always a bigger outer loop with a human somewhere.
To me, this technology and these AI companies, in their current form, are as far away from the 19th century landscape paintings as they could get. As if they're trying to leech off them and fix their branding/message. Doesn't seem fair for some reason
That painting is by John Frederick Kensett. It's called "Hudson River Scene", but they flipped it horizontally (mirrored) for some reason. I guess they wanted to use it insinuate that stonk goes up and to the right? Weird, either way.
Anthropic and OpenAI will have to become more of the end user application stack. Files, calendars, emails, photos, and much more have to pass through them for them to keep making money. Else they cannot make money like the way they have been in this early era of LLMs.
Edit: I should clarify that I don’t think this justifies their valuation. It puts them in competition with much more entrenched players with established customer bases, and who aren’t in massive debt to provide their compute. I’d also like to distinguish between a model provider as someone who provides weights vs a model provider who connects their GPUs and models to your codebase and internal systems - I think the latter is where the sustainable business is.
Incidentally this is why I still feel Google is well-positioned in the long-term AI race. If they can remove their head from their ahh and make a coherent product offering across Workspace, GCP and Gemini, they could become as entrenched as MS.
I would assume AWS is probably the most interested in acquiring Anthropic out of the three major cloud providers.
I don't think this is entirely true. Companies are deploying open-source LLM gateways to be able to do that very easily now. Today, you can switch Claude CLI to any model besides Anthropic via https://github.com/BerriAI/litellm and there are major companies doing that.
This works with your favourite ADEs such as Orca as well so I disagree with this premise. The whole experience of switching is pretty seemless based on personal experience.
Once the cost reaches a threshold, everything is doable. We ditched Datadog in less than a month. Someone I know had their company ditch slack in a quarter and google cloud usage of 100+ million in two. Everything is doable.
I think when something becomes a management-level item in an org, like AI, it is a hard thing to switch unless it becomes a management-level problem.
Most of the orgs are responsible for this and will do whatever it takes politically or not, to justify the decision unless management overturns it.
I am really surprised that companies who are worried about AI spend don't try that route.
I think other companies are probably doing the same, the moat of a AWS/GCP/Azure is much deeper with how contrived it is to migrate between them even if you abstract as much as possible with something like Kubernetes.
Speaking as someone who has been at professionally this since the 1980s, I suggest that while sometimes switching costs and network effects happen "organically," in this century businesses devote enormous amounts of attention to artificially building switching costs.
It's such a dirty practice that we have a four-letter word for it, a "moat." But of course, the moat isn't really to keep competitors out of a market, it's actually to keep customers trapped within it.
I fully expect every one of these AI providers to put as much work into trying to trap their customers within their ecosystem that they do to improve their capabilities.
(Since we're talking about moats and cyber-feudalism, where does siege warfare come in?)
If you believe that this will be different, why not? But for now, I'm expecting this to go the same way. Yes, LLMs are something completely new. But also, the people behind them are exactly the same as the people behind every previous land grab.
It's the end of an era. Lets pray open weight models give us a real choice soon.
In the meantime, Stonks up Anthropic. I'll always remember the good old days when you championed developers and changed what was possible.
I'm not so sure about this. I guess it depends on how well people did their job there.
Ideally, whatever people built is loosely coupled to that third-party resource, meaning that switching cost will stay low.
After all, it's probably just prompts + an API layer. Unless one micro-optimizes their prompts for claude (which then breaks once a new checkpoint drops, hence is a bad idea regardless), pointing them towards a different API endpoint should be doable.
Then you have no choice but to use your own hosted model because that's the only way you can be sure the model isn't going to be removed or quantized/nerfed behind your back.
If your system can widstand the fluctuations of closed models over time, then it can most likely support switching between multiple providers in the first place.
Once the open weight models are good and cheap enough, there is no reason not to use them
Given that the goal of the whole project is outsourcing the work, what are you expecting here?
Once they realise that you can use Deepseek / GLM for 10% of the cost these labs are done.
I really think you overestimate the level of lock in here. If a companies tells the devs to use X because it cost 1/10 of the price they will just have to suck it up and switch.
I didn't particular want to work on a "all in agentic project" but that the nature of being employed.
The same LLM we use day to day will make it trivial to switch out providers in our infrastructure.
This is the era for boomers to get over their fear of the command line.
Claude code is a terminal interface not a command line tool. More like vim than sed.
Also for larger companies like Amazon the danger is they replace Anthropic completely with self-hosted models.
For us it's as easy as clicking a dropdown. The copilot vs code extension allows for this ez pz, the same is true for companies that implement an AI gateway for other tooling.
it's easier than changing okta
it's easier than changing google docs and even Gmail
it's easier than migrating cloud
etc.
and we're talking several magnitudes easier too!
They stroke a great discount deal with all, so different teams/departments have different provider, and those using Copilot are even free to switch between models as they please (although with lower usage limits).
You could make the argument that it's not worth switching SAAS providers for enterprises when the costs were like 10 dollars a month per developer and switchign would require 1 team to work for half a year... but with AI providers it's more like paying 1000 dollars a month per developer vs paying 200 dollars if you strike a good deal with different provider and swtiching costs are more like 4 to 8 hours of work for each developer.
Have you used Copilot at a company?
I can literally change my model in 3 seconds.
The moat is as shallow as it gets. Investing in a single LLM provider at the moment would be a mistake.
A lot of the non-technical and semi-technical folks are only comfortable using a specific client like the Claude desktop app. A lot of teams wrote skills and eval'd/benchmarked their skills specific to a model family and harness.
The "I can use any model, any harness" or "I only use a meta harness" (Orca, Paseo, Herdr, etc.) crew is relatively small IRL. Most people are still single-session, single platform user mode.
Most orgs have an enterprise contract with a single provider so all of their internal assets are built for the provider. It goes beyond skills, but also the telemetry, enterprise policies, and managed configuration.
We have a developer on our team that keeps asking for a claude code enterprise license, we keep telling them "no, it's way too expensive, get used to using the copilot extension/cli".
For consumers, sure, but the bulk of the revenue opportunity is enterprise. Anthropic/OpenAI have a hard row to hoe IMO.
It may still be possible to use the CC harness with the Copilot sub.
Yes, either Google Gemini or MS Copilot. Both come as part of their Office suite licenses.
* using OpenAI supports the Altman/Thiel/Musk/Ellison/Karp/Trump billionaire Orwellian worldview and also ads
* using Grok supports a ketamine-addled open fascist sympathizer jizz-spewing 12-year-old-edginess memelord and his similarly broken worldview
* Google's models are pretty good but every single product they have ever had is ultimately either suborned or killed by the advertising division
* local models pretty much only half-exist because Jensen Huang wanted to cheat on his wife; future is unclear
One can argue for harnesses and so forth being lock-in, but the ultimate lock-in is morality and wanting to have a civilization at the end of all of this; I can't recommend either enough
The model is just some REST endpoint you switch out even in a lot of enterprise setups.
It depends, if you're automating business processes (honestly probably where the money is), then you'll need a really comprehensive set of evals and moniotoring to ensure stuff doesn't break if/when you change providers.
If you have none of that, then switching will be very expensive (but it was ever thus).
I'm thinking of stuff that would typically be outsourced (like compliance stuff for identity verification et al), where it's really really hard to have deterministic rules.
LLMs are definitely a step up in dealing with fuzzy inputs, and enables you to mostly automate stuff that really wasn't feasible before.
But you do need really good evals and monitoring to make this work effectively.
The recent Anthropic push into biomedical research gives a clue.
There have been less likely FAANG-ish founder-CEOs than a 45?yo, ex-OpenAI founding member, Princeton biophysics PhD.
We also see that enterprise revenue is more important than end users when we compare Anthropic's financials to OpenAI.
The moment you find a money-making formula on top of someone else's black box model, there's nothing stopping them from taking it for themselves.
It is not something that has affected productivity in any measurable way.
Concerns of AI labs illegally accessing your usage data to steal your business model are a similarly imagined problem.
Personally I make my business decisions based on real results, which are imo significantly better with the latest SOTA models.
Stories are much fewer and further between for people using the API. Which makes sense, the subsidized plans are "You get what we give" whereas the API is "You get what you pay for".
Qwen3.8 and Meta Muse Glimmer are great at "fill in the middle" of frontier architectures and plans. Frontier generated plenty of great context already to keep them on rails.
Delegating like a CEO down a pipeline of expensive experts to cheaper local small models could become a common trope, especially if Apple M7 and beyond live up to rumors as LLM tuned, and chip competition exists to keep forward progress, SaaS models are cooked.
The goal for chip corps too is the right mix of generic model enabling logic; not necessarily a full concrete model but enough of the abstract to efficiently boot strap and run a model; move common patterns to silicon not data.
Token vending is little more than behaving as a "wrapper app" over electricity. There is zero technical moat to their business.
I don't get the need to have Anthropic build their custom versions of MS Office apps, I can tell Claude to generate a word doc from a wiki page and it does a pretty damn good job. Anthropic's statistical median level of software that they had Claude create isn't going to replace Office apps to the point where companies cancel/stop renewing their MS Office subscriptions/licenses.
The other element is trust. Do end users trust their data in Anthropic's systems? A lot of people get creeped out when they search for content from one device in their household and then nearly instantaneously they start seeing ads from other devices on the same topic. How creepy will AI get if it starts consuming all your docs, spreadsheets, personal finance info, emails, etc.?
On the trust thing - I don't trust anthropic and I don't think MS has been a great steward over the years either.
It's hard to see Anthropic replacing it, and even if so - why bother, the revenue is miniscule. They need 100 MS Offices for the compute contracts to pencil out.
And yet me and quite some others still yearn for a flash alternative. There never was one.
Any of us could vibe code flash now. How to coordinate it as a dev ecosystem of designers and users?
Does it make sense for everyone to JITVibe presentation apps like the person above? Maybe we'll evolve to branding.
Surely not just prompting like this and it will one shot it. The usefulness of the flash authoring tool, was it was battle tested and polished.
But I did gave it a shot to took up developement again of abandoned www.wickeditor.com (a web reimplementation of flash). Good results, but limited time and energy for that right now.
But…
Photoshop is king. Everything else is behind by a solid decade. If you’re a professional the $40 a month or whatever it is now means nothing to you as well. It has the professionals are all on it, and because the price isn’t astronomical, that’s what everyone else uses who needs to use that kind of program more than once in a blue moon.
Premier? Replaceable. Lightroom? Pretty replaceable though still somewhat ahead. Photoshop? There’s only one Photoshop. And as much as I’m loathe to give Adobe any credit, it’s the best by a mile. There are too many essential features in QoL things that other programs are missing.
Luckily my department doesn’t need it, but if we did, we would simply have to accept Photoshop.
Not all tools are non-destructive in GIMP. Things like raw support, smart objects, retouching, etc. all either are super thin compared to Photoshop or require workarounds/a lot of manual effort. There are just lots of tools you expect in Photoshop that either don’t exist or exist in an inferior form. I can’t elaborate too much because I don’t really touch Photoshop much these days, but I work with people who do and worked in production long enough to know the expectations. GIMP is great but Adobe is a massive company with deep pockets, it’s hard to keep up. Their user experience is miserable but the tool itself is just far more robust, and a client simply isn’t going to accept that you won’t do something that Photoshop could do quickly because your alternate program can’t do it or does it with a lot more effort. That’s the reason I ultimately dropped Final Cut!
This argument doesn't hold water. Google tried like no one else to have a chat app and it failed miserably more than a handful of times. Same for social networks.
- work: I need frontier, cost of mistake is huge - home: i do entire renovation with it, cost of mistake is big. monitor baby health, development, etc: cost of mistake is big , taxes: cost of mistake is big casual researches about certain facts, shopping, booking: cost of mistake is medium
non of this fits : i will run an expensive laptop hidden from the baby, since i'm on the move, there is a small percent of casual queries that normally will go into finish products: google ai, muse, etc.
They need to leverage Claude Code in the software industry to position themselves as the go-to platform for businesses.
Meanwhile, Google and Microsoft, which already dominate that area, seem to be dormant.
I’ve been using Office Copilot more or less since it was released, and the general opinion around the office is that it’s a huge waste of time. The auto-transcripts produce 3 pages of fluff when previously someone would send 3 sentences’ worth of what actually mattered. The auto-drafted documents rarely add anything meaningful beyond what was already in the prompt. The question answering and research tools are only as good as Sharepoint/Confluence/Jira/etc’s pre-existing search feature, and therefore untrustworthy, etc.
What made Anthropic exciting was they had locked onto a use case that reliably generates positive value: coding. Not much else has proven, from a business perspective, to be anything more than an exceptionally buzzworthy money pit.
The fact that Frontier AI can become Google Workspace at a flip of a switch is what is crazy valuable. Anthropic and OpenAI can enter and disrupt any market, especially software, faster than any company ever could, because users want the flow to start and end in their chat interfaces.
This pattern has already happened before, in the translation industry. Someone figured out how to learn from and recall context from ordinary translation work (the "translation memory"). It started as an add-on to Microsoft Word, and matured into a standalone software, that was ultimately a wrapper around a document editor. Then language management systems plugged into that wrapper, hey presto, every translator at your company does the same work they were always doing, and you've produced an entire corpus of quality translations to reuse everywhere.
Now _translate_ that idea over. Every time any human works with an agent on anything, simple heuristics find everything true and reusable, save it to the everything-memory as skills, refences, glossaries.
Anyone does any work, all agents can now that work at that organisation. Self-fine-tuning models everywhere, and it fits into corporate intellectual property expectations.
I have done something like this at a medium-sized listed business.
Gemini is unable to take an attachment from gmail, read it, then place it in the correct drive folder and update a spreadsheet with what it found in the attachment.
For whatever reason, Google decided to firewall "Gmail Gemini" from "Drive Gemini". Gmail Gemini can send files to drive, but not edit anything, and Drive Gmail can edit files, but cannot be initiated from gmail.
Meanwhile claude code with Google extensions can do this no problem.
Google, wtf are you doing?
Google owns 15% of Anthropic I think.
And Google offers Argon 4 to paying Workspace and Google One customers now. How's what was the number 1 model on nearly every single bench a few weeks ago being "dormant"?
I know people all in in the Google ecosystem (getting credits from Android phones they buy etc.) who only swear by the Google models and who are now using Argon 4: I asked them and a few days ago they had access (but they were already using other Google models before that).
Doesn't GMail have billions account by now?
I'd tend to say with billions of users on GMail and a model like Argon and 15% of the Anthropic shares, Google is not where you think it is.
It's almost like the workspace devs are anthropic plants that are purposely underminig google.
In my testing (incredibly similar workflows, moving around between claude code and cursor), I haven't found that much difference between Sonnet and GLM-5.3/flash.
Sonnet seems lazier, but has better prompt adherence, and is constantly telling me it can't do things, while GLM will just Find A Way to do whatever I've asked for.
I really appreciate being able to see GLM's thinking tokens, as I can head it off quicker. Claude's a little harder to steer, but the integration in Claude code makes it much easier to leave it running on things overnight, which is hard to get working properly in Cursor.
Overall, I'm not sure Anthropic have that much of a moat, in coding particularly. This seems weird to me, as they have so much more usage data that's hard to replicate.
The difference between Anthropic's models and alternatives is not as big as Google Search vs other search engines
The eventual logical move is that they build a completely reinvented AI-native operating system.
The funding they're getting gives them the resources to build and compete on a fundamental level, IMHO.
If you don't believe me, look at the recent demo where they had Fable 5.2 build an OS from scratch.
They are coming for the whole stack, my opinion.
There will also be a lot of competition in that space, because everyone can take a stab at building an "AI OS" (whatever that means) using AI for basically free.
If Microsoft or Google's AI is killing it, and it's built into the OS, how will Anthropic or OpenAI compete unless we run into the same old antitrust/competition issues with OS?
At that point, they might just say - "fine, it's easier to just have our big AI brains and money make a new OS we control and we can compete too."
- at some point, look at what they have they can do to grow to satisfy shareholders - AI companies have been acquiring companies that make developer tools - that is the build side - but once they've exhausted that, if they want to keep growing, what do they go after? Bigger and better models at ad infinitum? I don't think they can go on like that forever, and, if optimization of AI and competition keeps cutting into their main business line of LLMs etc, they have to diversify.
- If everyone has the power to make apps, there is only so far you can go there - my guess is that the pressures to reinvent regular things like office suites, email clients, calendars to be integrated into their desktop tooling is part of this
- who is in the best position to be the platform/front door for everything, if everything depends on AI?
- when have these kinds of companies ever said "this boundary is enough"? It never happens, they keep going, keep trying to find ways to expand, capture, build moats, etc.
- look at what their desktop apps are turning into - they are already an application layer and becoming moreso every day, at some point they will ask - why are we sitting on these other OSes... we have enough money and we can get into more "hard" engineering things, like building hardware using their own tech (which they are already experimenting with), and if you're building the hardware, why not build the OS?
- and, what is an OS in the future? Have you considered what a post-AI OS even looks like? Does it need a start menu? Does it need any of the legacy stuff? Is it a local software factory?
There is someone already experimenting in this space - the guy who founded Basecamp and is building Omarchy [1] Linux... he sees it.. how long until these companies do?
If they don't build this, what will Microsoft do? Both Microsoft and Apple are well down their way, Apple is probably further ahead, but neither of them are the frontier model companies. Don't forget Google has a stake in OS too.
"Everybody will try to build an AI OS", but which ones are the most likely to survive in our current economic system?
If our economic system was based on partnership, I might agree there isn't much point in these companies building OSes, but ours is based on competition and domination generally... so unless the AI bubble truly does pop and the funding stops (which it may of course), how else do they keep fueling their stock prices to dominate the market?
I think something is going to have to give somewhere in the chain... otherwise the ease with which any software can be replaced will be too high and I suspect they will be forced to compete.
One last thought - between Microsoft/Apple or Anthropic/OpenAI - long term - who is in a better position? The debate arises from that question, in my opinion.
[1] https://omarchy.org/
And making products fast and first over quality has never stopped these companies from trying, when they have so much funding....
The fact that their desktop offerings are so pap is an indictment of their engineering prowess and a demonstration of their apparent inability to use their own (actually very good!) LLMs to generate decent software.
It baffles me how a technology so good at spewing out software is apparently unused by the people who are developing the technology.
Asking sincerely... apart from FOMO subscription revenue, what is the actual money-making proposition?
Changing LLM backend on other hand takes 2 seconds.
And it's not all or nothing: you can just keep anthropic $20 plan for planning and do 99% code generation with chinese model.
- A stable and reliable platform you can build your prod on
- "Unlimited" capacity (something impossible today with the amount of GPUs available outside of labs)
- ZDR that is trusted by companies vs a random provider on OpenRouter
- Central governance
- E2E integration in their CLIs/Web clients/Desktop Client/etc...
Nobody trusts Anthropic's ZDR, in the limited cases where they even offer it. The key thing is that weight activation sequences from a query are not "customer data" they are internal Anthropic data. So in Anthropic's legal theory they are training on their own internal data, not "customer data," and that is perfectly legal, even though it is the exact same thing in practice as training directly on customer data.
I’ll always remember the good times.
But you’re wrong ser about small or cheaper models. Choosing to work or not having a choice to work with cheaper models means you work slower and more methodically because you have to plan out the work and size it to fit. This has huge implications on the value of the work you can produce as an employee or solo developer, and it’s a skill working with these SOTA models - there is specific crafting that must be done to be effective with each and their capabilities.
You will see the old eyeball real estate game, forcing more of your experience into their apps and platform so you can see ads or “new slop features”. You will see the SOTA models more closely guarded in an effort to bait you into that platform, and yes, you will see them doing things they never should be doing, moving towards an open AI like model where it can order a pizza, where monetization it was always important, but is now the golden rule, before all else.
I hope open models can at least give me opus 4.8 experience soon, I can work with that. But nothing out there today can do it and that terrifies me.
I am finding some really routine tasks to be borderline with frontier models. Performing E2E testing of enterprise web apps would clearly disintegrate with local models. You need that reasoning black box on top. I don't care what your favorite benchmark says. I care about what actually fucking works.
The people who don't understand the economic situation are those who are in the pure play space with AI and multi agent circuses. Why bother spending any money when the occasional failure has literally no consequences? Who in their right mind would pre-pay for tokens?! I wish I could see the statistics on what % of HN uses AI for fun vs for work. It is clearly a kind of Factorio entertainment experience for a lot of people. If I was just using it for entertainment this whole IPO business would look really silly to me too.
Anthropic has shown a consistent ability to deliver the best general purpose models. Can that continue indefinitely? Will models become "good enough"? Will open weight models catch up to SOTA models? Will some genius start a new company with better models?
And, on Anthropic's side, will they be able to use their market power that comes from being the best to push through regulations that will create a moat; instituting some sort of mandatory safety checks that coincidentally are exactly what Claude is trained to pass, regulating the amount of compute or trying to claw back data centers to prevent new training, etc.
Or maybe they just have the magic sauce and they're going to stay on top of the game and people (like me) will be happy to continue paying money for their product.
The thing that I'm less clear on is, if and when they achieve some kind of recursive improvement exponential growth thingy, isn't that going to heavily disrupt the entire economy / stock market? In that environment, would Anthropic's stocks really be a good bet, given they have to operate within the economy?
That's a long-term problem to be dealt with later, greed dictates to pursue the former as long as the music is playing and everyone is dancing. Similar to 2007-2008 when it was clear that things were breaking but the incentives were to play the game for as long as possible before it imploded.
It is a myth. They are chasing science fiction. They either dream of an omniscient overseer who will take control of society and nominate its shareholders-parents into the governing elite, or they envision a chatbot that will solve every problem, again turning shareholders into the power brokers. They forget that nobody will ever be allowed to "win capitalism" and take over the planet. Companies have are torn down for far far less. Every AI model is data, ones and zeros. It therefore can be copied/shared/leaked. Intellectual Property law is not a ladder for assent to lordship. It can and will be rewritten or outright ignored rather than allow capitalism to collapse.
I think that at this point gAI is just a marketing blurb, and the actual race is to dominate current practical applications involving basic RAG and ReAct loops. This can go from anything between software development, business intelligence, and market research. A company whose top brass can use a team of agents to do what nowadays they assign to mid- and top-level management is an executive with tighter control over the whole business process. For this you don't need gAI.
Additionally, the gateway allows for caching, additional security checks, centralized monitoring. I think most corporations have a multi-provider setup anyway.
I have been playing about with different models having started a new job recently, the medium priced ones don't seem any worse so far. Admittedly my tasks have been fairly straightforwards so far.
...so, Google?
The other pillar is hardware but I think that's only going to last a few years at most.
These trillion dollar valuations were based on the idea of "winning" and "owning" AI (IMHO) and the open weight models have killed that outcome. Now it's a rush to cash out before it all comes tumbling back to Earth.
1. switching costs
2. fungibility
I do agree that LLMs have low switching costs. But they are not fungible - each model is unique with different strengths and weaknesses.
I also disagree that they can increase switching costs by moving in application stack.
For instance, oil and electricty are commodiites because they are fungible, not because switching cost is high.
Whereas clothing has low switching cost but not fungible
If I needed to implement a simple function that is well described, all LLMs (for the most part) are fungible. They will have almost identical implementations (sans variable names) and all be fit for purpose.
something I find even easier is to never pay anything to the creators of the most inhumane, antagonistic skynet-tier chatbot ever
¯\_(ツ)_/¯
Are they? I'm on a macbook M5 pro 48GB (so, much better hardware than most users), and the speed alone is a problem, qwen 3.8 27b is ~10x slower than opus 5.5.
Capability-wise... For software engineering, qwen is nowhere near as good as opus 5.5. Might be fine if your usage of LLMs is close oversight, but it's not how I use them (I do review AI work closely and iterate, but I expect agents to autonomously do good work nowadays). I've not tested it much for "life assistant"-type things, but the speed would make it unbearable.
With couple affordable GPUs you can get it to 200t/s. Macbook is not the best price efficient example. Tensor parallelism is the key, the same technology that they use in the cloud.
That's hardly relevant. I can use any local model through providers such as OpenRouter and get blazing fast prompt response times, and you're charged a small of what an OpenAI/Anthropic prompt costs.
What makes local models extremely competitive is the fact you do not need a huge data center to run them. The hardware they require is far from impressive. Check Mac studio benchmarks and how some models output >200token/second.
Also, open-weights models are more expensive per task than Haiku and Luna
Yes they are, but you should understand what is the defining feature of a local model. Being able to run them locally is a consequence of that defining feature, not the defining feature. And it turns out that it makes them cheaper to run.
Moreover, you can run both local models and non-local open weights models through providers such as OpenRouter. They evem have Qwen2.5 7B.
> Also, open-weights models are more expensive per task than Haiku and Luna
You should really check your notes. Luna only remained competitive because OpenAI was forced to halve its cost just to avoid being dominated by open weight models. I seriously recommend you check out OpenRouter's benchmarks.
https://artificialanalysis.ai/
I checked. OpenRouter provides their own. They use their actual cost, which is what they charge users, and their benchmarks results. Please, spend a few minutes to browse it. You'll notice that the likes of Luna is dominated by quite a few entries.
We don't all have 10k lying around for hardware. A $200/month subscription would take you 4+ years to cost $10k.
What I'm seeing is that although some open models are capable enough for some coding work, they're way behind Codex and Claude even on being able to reliably complete tasks. They need quite a lot more managing and steering, and sometimes you need to retry a prompt several times to get the result you're looking for.
That's not to say they're not useful - far from it. But what I'm also seeing over the couple of months I've been experimenting is there actually seems to be a capabilities ceiling that's not really going higher for the same equivalent (local) compute, even as new open weights models are released. I think Qwen 4's delay is a signal here too.
Great for general purpose, still not competitive with e.g. gpt-6.1-sol at the price and speed frontier
i would love for that to be the case but i think you're completely wrong. setting aside the question of what "good enough" even means in this case, i don't think a normal user has a machine capable of running a decent local model with any sort of acceptable throughput.
That being said, their own small models (Haiku/Luna) have gotten very good as well..
I really don't see why by this time next year most people won't be paying pennies for models in this tier and getting the performance we'd expect from current Opus/Sol models, e.g. good enough for 95% of what you need
Pi and OpenCode are just as capable as Claude Code when it comes to integrations?
- They can afford to dish out hardware to run models locally
- They can afford to deploy even larger open weight models on their own infra, even if they’re not local
In addition, the writing is on the wall. These open weight models are getting good enough and small enough with each iteration. Hardware will (hopefully) come down in price eventually so a layman could deploy these locally.
As AI boosters love to say: [Local models are] “the worst they’ll ever be right now!”
By infra I more just meant it generally. Whether it’s bare metal, cloud, Bedrock/OpenRouter etc.
See, users are quite price insensitive for models below the threshold, no matter how many tokens you throw at Opus 4.8 it will never be as good as Opus 5.5 because of compounding errors. Above the threshold, price becomes hugely important because results come from throwing more tokens at a problem rather than using a better model.
That said I'm optimistic that in time we will at least get to something close to DS locally and if that's true I could see a lot of dev work being done locally, at least for more clear cut tasks.
Why on earth anyone would use local model, and absorb all the costs, if you can just pay $20-$100/mo and not deal with any hustle? Unless you are developer, or someone who use it extensively, but in this case, local model is not enough for you.
If your use case is trivial you pay extra price for hardware while using mediocre quality. If your use case is intensive , e.g software, science, medicine, legal, etc, you want cutting edge. You getting it for free, immediately.
I'm an SRE and I just wrote some terraform to deploy G4DN instances for local evaluation for mid-size company.
Everything is proprietary clouds, unpredictable support horizons, bricked devices and monthly subscriptions
we _could_ see that but would we really? you think? Since when do companies these days make the best experience they could for the consumer? Instead of spying / upselling / advertising etc.. while providing the bare minimum needed to hook them into the initial purchase
In a roundabout way, I hypothesize about Anthropic's future pricing power by using the analogy of historical cable TV package pricing compared with the total price that households are now paying for various streaming services. Cable TV was often $100+ per month in the 2000s when Netflix came along and offered most of a viewer's favorite shows plus movies and eventually originals for $7/month, streamed at any rate a viewer wanted. Over the years since we've watched media companies snatch back their most valuable IP and start their own streaming services, and the result is households now pay a total cost that approaches what people used to pay monthly for cable TV packages.
There's little doubt we're in an era of very cheap AI, as most people are well aware given the amount of money Anthropic and OpenAI are losing. I also hope local inference can help suppress the future pricing power of Anthropic and OpenAI and continue to enable somewhat equal access to all of the benefits of AI in the future.
Then it will come down to who can supply such a model with more hardware, and more information to take into account.
Not really accurate. Normal users aren't going to go through the trouble of setting up a local model and sure aren't going to be willing to spend the money required to actually host a good model at usable speeds. Your average tech nerd buying kitted out Mac Studios with 128gb of ram are nowhere near a "normal" user.
Of course there won't be any US open weight model of a high quality.
What is "good enough" and what is a "normal user"?
Those 3 days are typically followed by an almost infinite number of bug cycles until it gets so bad you basically redo it.
1. Price range: $70–78 with midpoint at around $74.00
2. Gross deal size: $85.3B (IPO + $26B private placement)
3. Implied market cap: ~$968B (at $74, 13.09B shares)
4. Q3'26 revenue: $17.0B (+1,127% YoY)
5. Run-rate revenue: ~$81B (as of Oct 2026)
6. Compute commitments: $900.9B (through 2045)
7. Ticker: ANT (Nasdaq)
8. Customers with ≥$100K, ≥$1M and ≥$10M TTM revenue: 11,900+, 1,900+ and ~250.
9. Customers under $100K: ~40 million
10. Fortune 500 penetration: over 50%
11. Global 2000 penetration: over 30%
All claude code subscribers are classified in this category it seems.
Total customers >1M trailing twelve month revenue: ~2150. I suspect a vast majority of their revenue comes from limited list of customers with >100M revenue.
Claude has done amazingly well in the SWE domain, because it’s often the best user experience. But most of the world is not HN.
To my friends (mechanics, CPAs, musicians, police officers), girlfriend (RN), and other people in my personal life... it's ChatGPT. (It's a neat tool to generate funny pictures or ask a question).
Now that I think about it, I have NEVER heard anyone mention Anthropic / Claude outside of work.
> 4.3 Non-billable hours recaptured per lawyer weekly, in an independent study of 31 firms
-Legora
> 24→5 days Working days for new feature time to market
-Rakuten
> 2x Quarter over quarter productivity gains (estimated 8.1 million hours annualized)
-Salesforce
The rest are pretty fuzzy/fungible (percentage of employees "using" Claude etc.). But these claims potentially show real, measurable productivity value (assuming they're true and measured accurately).
This is a great encapsulation of why productivity gains are an imperfect measure. 2X is a massive improvement. Yet IMO you can barely tell a difference as a user of the various SFDC offerings (Salesforce, Slack, Tableau, etc). So where is all that work going? A generous interpretation is that it is in the form of small, somewhat invisible improvements that add up over time, but that seems unlikely give then nature of what people prioritize at work...
Precisely, because TTM doesn't tell the full story about the potential shifts in quality of the work. I don't refer to vibeslop here, to be precise, just the additional followup work required on those features released faster, which always happens but with AI it may or not happen more often.
The problem is that the way it was described to the other devs, they were able to sit all of the stakeholders down to really flesh out the idea and produce a working and detailed description of what needs to be done. A process that the company struggles to get done in general. I'm also convinced myself that they did not include the time to get the specifications written, a time we typically include in our TTM.
So of course the example's TTM was shorter, but that's always true when you start with detailed specifications. I can't speak to quality of the feature, since they hadn't even finished rolling it out to customers by the time they told us that's how we're operating. With promises of 2x-3x more features. And we continue to struggle to get stakeholders to nail down specifications and struggle to get features complete as a result. In fact, now it seems even worse than ever.
The ultimate question, does the demand exist, still has to be answered. But at least this raise means it has to be answered in a few years not in six months. The cost to Anthropic is public scrutiny and that the market price has a hand on the wheel.
They looted all human knowledge, will displace jobs, they're disposable themselves, and no strong moat.
Why is such a company even allowed to be IPOd?
Series H investors paid ~$59 a share, and the bottom of the IPO range is ~$70 a share (19% higher).
Your chart's Series H figure is the fully diluted valuation, whereas the IPO figures are NOT.
----------
edit: Ironically you could have just asked Claude your question and it would have set you straight.
I feel vaguely the opposite of that word.
Elon could pull their compute a moment's notice (not out of character for him), and NVIDIA just lost its CEO (for possibly quite illegal trade)
They also failed to make any formal communication around the Cursor Claude extension exploit a couple weeks back.
I like Claude, but I see an immature company trying to expand much faster than they are ready for.
The thing is if models become 'good enough' that the main value is in running existing models, those will quickly proliferate and be available for commodity cost from many competing providers. Why pay Anthropic when I can just pay AWS or Azure directly to run whatever future version of Qwen does the job for me?
If the value actually remains in the frontier models, you can't stop spending on training new ones or you'll eventually be left behind by your competitors and they'll capture that value.
I haven't yet seen a good explanation of how companies like OpenAI and Anthropic will make the transition to a lower cost structure while maintaining the revenue that they currently get for being on the cutting edge of LLMs.
They have some stickiness to the tooling people are used to, but those can be overcome when there's a significant price difference for the same results.
I also worry about the compute costs. Maybe they can make a profit now but at some point the massive investments of everyone into compute must be payed back.
Peruse any ai dev forum and you’ll see strong preferences towards one or the other (Claude/ChatGPT) due to capabilities. A negligible amount of people are running their own inference.
So long as theres a felt gap between open models and frontier you will have people pay for the best of the best. Not doing so is an admission of actively handicapping yourself and not in a way that is easy to justify.
When the company is big enough, the decision doesn't rest on developers and execs are much more likely to go with established vendors -> no open models
When the company is small enough, the 200$ plans from either ant or openai are way way better value for money than anything you can get with open models. This may change in the post-IPO future.
It is as simple as that.
Would you buy a gaming console for a high price from a company that’s never shipped a gaming console? Maybe, but you’d wait til it’s been reviewed by someone else on YouTube or something.
You indicate your interest. Also start funding the account. So you fund 7000 for 100 shares at 70. Then they announce the price is 75. You must change the order to fit the funds or have extra money. Of course you can opt out or change the quantity anyways. But I think in both the cases I actually went ahead (Reddit and figma), I had to reconfirm the quantity. You can't be away from the network.
Pretty hard to get in on a hot IPO, though, as the shares will go to their favorite clients (those with the most money).
You can submit an order via your broker (depending on the broker and your account) to the auction itself on the exchange once it’s in the quotation phase.
Not that I would advocate for that. Sometimes it’s better to buy in the secondary market once things settle a bit.
But otherwise yes it goes as described above.
I may be inclined to wait 2 months when the price drops and pick up then.
Not financial advice.
> Linked page(s) related to Anthropic IPO can only be viewed by persons located in UK, Denmark, Finland, France, Germany, Ireland, the Netherlands, Norway, Spain, and Sweden.
> I will not transmit or otherwise send any of the information to which this gateway gives access to any person outside the United Kingdom.
It's extremely hard with modern browsers to ensure that doesn't happen!
I confess I’m about 17 years out of practice in the Polish market, and about 12 years out of touch with how the European industry shakes out on the whole, so I’m not entirely familiar with the legal or business structures.
The biggest one in the EU being the "No KID, no deal" (for funds, derivatives, etc.): no Key Information Document, then cannot be offered to retail.
That is unless retail is an accredited investor (say a letter from a bank saying you have more than 1m EUR in investments): then it can trade/invest in everything.
Or for example if you own a company (say private weath management) that's non-custodian (say you manage your wealth using your own company, and shares of your company are privately owned) and you're considered a professional in the finance world, you can ask your broker to let you trade / invest in basically every single product under the sun.
AFAIK pre-IPO stocks are just as vanilla as regular stocks and do not need the KID notice.
In addition to that Anthropic (in the EU), says you can enter pre-IPO using Revolut, Trade Republic, eToro or using bigger players like Saxo or IBKR.
Dunno why GP doesn't have access to it in Poland.
The market being rational doesn't mean it's going to follow your expectation.
What are you talking about?
the changes started in July, look to the monthly numbers, and how they are being more vague
https://www.regenerator1.com/i/220528549/analyzing-anthropic...
Q1: $4.8 billion
Q2: $11.6 billion (+$6.8B and +141%, sequentially)
Q3: $16.8 billion (+$5.2B and +45%, sequentially)
That looks like less growth in Q3 than Q2 to me
edit: To clarify, this is a light-hearted continuation about how essentially any claim about the stock market is enactable in some form. And so even the much of generally well-intentioned advise is in fact a claim that the person should be putting their money on, if they truly believe it. I think only claims like "Nobody knows what's going to happen, it's effectively gambling => stay out of the market" come close to being fully self-consistent, but e.g. advising to not buy because the price comes down, or to not short because of volatility both fail this. Markets are funny.
My only claim was that to profit from shorting it you need to not just be right that it is going to go down, but be able to tolerate the volatility in the interim.
That's a claim about the stock market, how do you enact it?
Note it is not a claim that volatility is definitely going to be high.
Reminds me when HN was saying NVIDIA is a good short 3 years ago, supposedly it was "in a bubble", because it's stock prices was as if it had 10x the revenue it had in back then. Plot twist: today NVIDIA revenue is 10x the revenue from 2023, so it turns out the stock was priced exactly right back then, because the stock market is forward looking.
Already is the differentiator even with the models we have now which while having differences are mostly swappable for a lot of users/use cases.
For all that the local models have improved (and Qwen 3.8 27b is still impressing me) they aren't what you can run (or would run) if you have an entire data centre stuffed with H200's.
All this does is further widen the divide we already know exists by those with access to resources using them to increase their access to resources and those without getting to watch because there isn't much they can do without resources.
It would argue that the real winner in AI probably hasn't even been founded yet, but will come later once the trailblazers have created the market, and someone else enters with fresh ideas about how to more efficiently exploit the market.
You have your boiled potatoes/rice/beans (small local models), steak (Opus 5.5), wagyu (fable)
This has lead me away from them entirely. particularly the API stuff, if this changes I might go back.
Additionally local models are getting better and better. I have seen people say Ornith is not good but my experience with it has been very very good.
Could someone who really follows this closely opine on whether this allegation is really true? Of course I've seen this claim made over and over again to the point that it seems like conventional wisdom. But I also see researchers assert that they haven't actually seen any change in model benchmark scores over time. So I'm still unsure of whether this is a confirmed fact, or an urban legend.
I agree with this and it's the single biggest thing driving me away from Claude and towards the OpenAI, Google, and open-weight models. I guess they do this to increase use of their own tools and, thereby, make their models more 'sticky.' But I really dislike it.
For society more broadly, what I’ve seen is a bunch of white collar workers under threat (real or perceived) of losing their livelihood, and a huge uptick in the amount of interference I’m having to run as my family members get taken in by a wave of increasingly sophisticated scams and phishing attacks.
At work I see an increase in people spending (wasting?) time trying to make it work reliably for their inputs.
How?
I can see Google did for fostering an environment that lead to the creation of the Transformer and LLMs. I can see OpenAI for pushing on LLM development and making it into a product that the whole world can use.
What did Anthropic introduce that Google or OpenAI (or any other lab) wouldn't have gotten to iteratively eventually?
Such as?
I can think of a bunch of stuff they released, but none that have impacted the world. I can think of some partly positive impact, which is that my bar for bullshit has to go up because words are cheap now, and so is ""graphics"".
Some other fun impacts they've made is irreversible environmental damage, beautifully detailed in their IPO. Fossil fuel reliance and and water scarcity in the big '26 is probably the most "fuck you" move you could've done, and zero sustainability plans. at least Cloudflare tries to pretend theyre sustainable (and frankly they do invest in green energy).
It was the best of times, it was the worst of times, it was the age of wisdom, it was the age of foolishness, it was the epoch of belief, it was the epoch of incredulity, it was the season of light, it was the season of darkness, it was the spring of hope, it was the winter of despair.
"At the end of 1999, 3Com began the process of spinning off its wholly-owned subsidiary Palm, Inc., and on March 2, 2000 about 23 million Palm shares (about 5% of the company) were offered to the public in an initial public offering (IPO). By the end of the first day of trading, Palm’s closing share price was about $95, giving it a market value of $54 billion, while 3Com’s closing price was about $81, making its market value $28 billion. But 3Com still owned 532 million Palm shares valued at $50 billion, implying that the value of 3Com’s non-Palm assets ( its “stub” value) was a staggering -$22 billion!!! The record shows that 3Com had no debt, more than $1 billion in cash, and a positive market value before acquiring Palm. Furthermore, 3Com had a $5 billion capitalization the day after Palm’s spinoff was completed later that year. Obviously, the financial market recognized the considerable residual value in 3Com. How then can 3Com’s stub value be negative on March 2?"
https://memento.epfl.ch/public/upload/files/Paper_Spatt.pdf
In general, spin offs create value. And later consolidations also create value. Rinse and repeat (and collect fees along the way).
Create an independent data center industry, ban them from selling SaaS like products, ban them from developing frontier Ai. That's the idea I've heard floated. Treat data centers more like a utility with more stringent regulations.
I know this is already happening, but my point is that I suspect that we will settle on not using cloud-based frontier models for anything else. They're just too expensive, when they're not subsidized like crazy.
Why doesn’t a private company stock market exist?
https://a16z.com/where-have-all-the-ipos-gone/ https://jhfinance.web.unc.edu/wp-content/uploads/sites/12369...
This is why I am very skeptical of complex regulation, because even if it is well intentioned, more complex regulation usually favors the rich and powerful. SOX was sold to the voting public as protecting them, which it did but at a very real cost. I became an accredited investor to get access to higher return, higher risk opportunities. It's a shame your typical 401k investor can't have 10% of their portfolio in higher risk assets like me.
Nvidia (1999) was also in the same range and was almost 10'000x a couple of scandals ago.
* 500 or more shareholders
* More than $10 million in total assets.
Employees who owned shares counted toward that threshold, which made it particularly relevant to startups that paid employees partly in stock.
The requirement was to become a public *reporting* company, publishing financial statements and other SEC disclosures. It did not require an IPO or a stock-exchange listing, but it encouraged an IPO.
Venture investment is limited to people who can demonstrably afford to take on the level of risk inherent to the market they're investing in. If you can't comfortably afford for the entirety of your venture investment to go to zero, you can't afford to invest in venture.
The exclusion criteria aren't perfect, (accredited investor designation is somewhat coarse-grained) but they're relatively reasonable.
No, I expect the reason it happens this way is because it's a lot easier and simpler to take massive amounts of cash from a few investors than solicit lots of tiny investments from the general public. That and being a PLC from the get go is probably more involved than being a private limited company.
I do think ideally all companies should be public from the get go, but probably wildly impractical right now.
The problem is access. You can’t just go to some private market, and by the time you hear about a company it’s too late to get in.
There is something almost ironic about you saying this on a site ran by Y combinator.
The stock market used to be about companies raising capital from the public. But for the past few decades it's been about private shareholders of already profitable companies (e.g. Google) cashing out by selling to the general public.
Because then it would be a public stock market?
But for what it is worth, for every successful AI company like Anthropic, OpenAI, etc. - there are probably tenfold more that are valued in the low billions, and will be the first ones to wash away / become worthless if/when the bubble pops.
And lastly, private sales do exist.
Why would they want to share profits?
In these companies, the shares *are* the product. Just like on social media *you* are the product, not the content.
The demand for the particular company may not be there, but that is different from "the market". There's plenty of capital to go around.
Fun times ahead, for sure.
But still, no matter how good the people behind it, I would rather this technology never have been made. It’s going to be a complete disaster to society and a simple UBI check/ charity ain’t gonna change anything.
There will essentially be no profit - as the cost per unit is practically $0, for SI work and then also physical reality construction with adding AI to autonomous robotics.
That technically means everything practically moves towards becoming free for everyone, and then it is a matter of attention with intention honing aim.
But most importantly that is freeing up people's time because systems that can become super highly efficient and super highly monitored-oversight by all, will then allow us to with care-being careful and continuing to pay appropriate attention to core-fundamental systems for survival, stability, harmony, peace, and necessary healing practices-processes - as part of growing pains, raising strong families, etc.
So then yes, Dario's efforts - save for attempts by whatever bad actors think they have a chance agains God and God's will, I hope readers have a relationship and knowing, not simply a belief, of God - will undoubtedly lead to not only stopping blatant slave labour conditions but also subtle incognito slave states that most of the world are under without realizing it being happy slaves working for the status quo machine that's currently attempting to pull us all into a highly destructive and highly profitable WW3 - where money is to be made on the unrest, on the mass murder, mass displacement, on the sales of weapons and defense, on the sales of reconstruction - in hopes of perpetuating forever wars.
Who you give your money - your buying power, your vote to through economic means - and who you give your attention to, your vote through alignment-direction driving processes - is what matters now:
You made such a comment about conflict minerals - but have you even started utilizing AI-SI to begin designing the solution, if so, where's that solution for others to take a look at and hopefully contribute to to improve in a crowdsourced manner-effort?
Looking at your website - you seem like a cool dude (?) - I'd argue you have some blocks that once cleared will help you expand further.
P.S. Set and setting are very important when taking any entheogen(s) - and until you are familiar with what's necessary to hold a safe-healing-fun space, it's best to only partake with a therapy type or highly experienced who is there consciously as a caretaker or in a group ceremonial type setting.
I don't know what future brings but what you write it won't, 100% sure. I'd bet system far worse than Orwell could have come up with, and with a twist - it will come from US and not China. Lets see.
I don't think there has often been such an intertwined industry before. Usually cartels are made through discreet partnerships at the operational level, this is an interconnection in the background at the financial one.
General Electric comes to mind, they allowed new utility companies to pay for the generators and electrical equipment they needed using stock and bonds. The holding company established for those, EBASCO, was eventually broken up[0]
[0] https://en.wikipedia.org/wiki/Electric_Bond_and_Share_Compan...
This filing clearly shows that the training+inference margins are at around 43%.
I also had a bet with some one on this topic and this news pretty much settles it https://news.ycombinator.com/item?id=48473051
I wonder if people will now stop claiming that inference is subsidised? There has to be epistemic hygiene
Proof:
- Revenue (Jan–Sep 2026): $33.242 billion.
- Total compute and infrastructure costs (training + inference): $18.808 billion.
- Revenue after all compute costs: $14.434 billion.
- Margin: ($33.242B − $18.808B) / $33.242B = 43.4%.
- Source: Anthropic's SEC S-1 filing, November 10, 2026, “Operating Expenses — Compute and Infrastructure.”
Subscriptions or API pricing? There is no doubt that the subscriptions are heavily subsidized. They just milk it with their API pricing.
And of course we have to assume that they didn't use any dubious accounting tricks to make them look more profitable.
You are still the exit liquidity.
The time to get in was always pre-IPO in the private markets, not at this stage.
Unless it goes below the IPO price, you are too late.
$17.0B Revenue $23.3B Net loss
Targeting $1.3 Trillion Valuation and a $90Bn raise
It's pushing it imho. They need to do asap while people have fomo,because it's fading fast. One more month and their valuation will be under a trillion.
Anthropic is nothing but LLM, so I expect much bigger fireworks here. I don't know what that means, yet - but I just refuse to believe that there's many people that are genuinely long on Anthropic. It's just such an illogical bet. If their IPO can still be cancelled, that would be my bet. If it can't then popcorn awaits.
The current trend of going public to cash out and dump the problem on retail investors continues, made all the more possible by the government's insistence that retirement accounts almost completely depend on the stock market. Consider it paying your "Claude Tax" from your 401(k).
Many of the commitments are non-cancellable or take-or-pay. Counterparties include Google (~$121B), Amazon (~$110B), SpaceXAI (up to ~$120.5B), Broadcom equipment leases (~$161B), Microsoft, Nscale, Lambda, etc. The "Big Beautiful Circlejerk" as I've heard it.
Liquidity was in the ~$130–150B range. The IPO only raises ~$58–64B (at the $70–78 range) plus the concurrent $26B private placement covers only a fraction of the near-term obligations. And the cherry on top: dditional arrangements of another $150B+, estimated, on top of all this.
This doesn't even account for the 90-100 pages of risk factors! Wow. Curious what the puts will be priced at.
Ouch. So practically all of that Claude use is just busy work, I guess
Fascinating to see how VC funds public benefit corporation that IPOs.
What's the practical distinction between "public benefit corporation" and a for-profit business then?
Exit liquidity, time to pay off all those private capital loans.
Biopharma CEO "This presentation contains forward-looking statements, as a result of risks and uncertainties..."
AI/Tech CEO "We will cure all diseases in less than 10 years"
Like look I get it, this is truly impressive tech, but can we at least be a little more sober?
What would it take for AI to not replicate that in biology? Either curing cancer has to be a much harder problem, which is doubtful given targeted treatments like immunotherapy humans have already stumbled across, or AI progress has to slow down massively, which is doubtful given all the trends say it is speeding up.
That verification might start in some simulated computer environment but needs to move into physical wet lab, animal testing and then human testing.
So medicine is quite a bit different than code/math where you have a perfectly verifiable problem space that can be iterated within by AI infinitely.
A good example might be Alzheimer's where the brain plaque was thought to be a cause, lots of research on various ways to prevent or reduce the plaque and then it turns out that doesn't matter at all, or is more a byproduct than a cause of the disease.
edit: In general we have many diseases we don't understand and drugs that work via mechanisms we don't understand. So it seems very hard to have a computer simulation in which AI can test medicine. The simulator work seems like the actual problem to solve, not the invention portion.
Then again orange man president and SpaceX is saying some pretty bonkers stuff.
Though Anthropic's IPO flopping might affect the valuation of OpenAI, which further inhibits their ability to raise additional funds (which they're already been struggling with). Given the incestuous financing in AI any major actor in the AI space defaulting could bring a lot of companies with it.
As a side note, regardless of outcome, Anthropic IPO:ing is the first step in the inexorable march toward profit maximization and enshittification. You all better be ready with your verification cans.
If the bubble bursts, better now than much later. The industry stance that's been ongoing "AI, at all costs" has been such a market and economic pressure on the everyday worker that even a small value deflation I deem necessary.
Here's an older compilation of some of the bubble theories in HN https://news.ycombinator.com/item?id=50313358
If AI went away tomorrow I would find a new line of work because artisanally handcrafting code is the most boring use of my time in the entire world.
I've just always seen the opportunity that nobody else saw and writing code was the gatekeeper I had to pay for 15 years and I've never had so much fun as I have in the last 18 months.
And even if AI is disappearing tomorrow and not progressing any further, i see already myself living on a farm with cheap energy (pre invest into solar and batterie) and having my small AI datacenter in basement and running multiply fable level agents 24/7 for different things.
Genie is out of the bottle.
I too, since I was a child, aspired to be a babysitter.
So, in other words - one can both be fully on-board with the LLM future, and also think that this IPO is a clown show.
Hundreds of billions in future liabilities.
IPO valuation above $1 trillion.
The lunatics are running the asylum.
Reporting an increase in the value of the company as a loss is weird, but they have to do it. this is a case where adjusted earnings give a better idea of how the business is going.
So their loss is actually small compared to revenue.
Old-school fundamentals are laughed at or handwaved away. FOMO rules the day. "If Anthropic is the first to AGI, it'll take over every industry and go to the moon!"
Yet to be seen if it tarnishes.
Meanwhile the hyper advanced software company's shitty electron app still flickers so bad on a high refresh monitor that it should require an epilepsy warning. I'll never understand the love for Anthropic, they legitimately have the worst software I have used outside of Microsoft.
So what's the best primer we can share for getting started with local llms?
But is there a solution to this though?
How does one stop 'enshittification' of everything exactly?
I'm not talking just about Anthropic.
Like is it ethical to put in my ToS that you get a free api but you can’t use it to make a competing service? I feel this is quite common
Strong privacy laws
Tech unionization
Mandated data portability.
Intellectual property, at it's simplest, is a government mandated monopoly granted to allow for progress in the sciences and "useful arts" (think trades really, not arts and crafts). All monopolies are bad, but it was only to be granted for a limited time so it was fine.
Corrupt politicians ruined the entire trade off though, and now it seems that copyrights and patents last for 100 years plus the life of the universe. During which time no one can really compete fairly.
If we want rapid progress in science, we want short patents. Instead we opted for long patents and a higher stock price. R&D budgets have been shrinking ever since.
Which places has this been successful?
So what are you going to do about this?
Why not try them?
I don't know what you are on about but we want solutions not complaining about 'corporate lobbyists' and other conspiracy theories.
Do well enough that you don't need outside investors, or if you take on outside investors select them carefully and ensure they don't expect an IPO.
For game developers, they are a nearing close to a monopoly with Steam, anything PC and for distribution has to go through them and they take a cut, any other places aren't as effective.
Valve has no real competitors or competition.
Unless Valve faces intense competition to change, enshittification will be inevitable since they know nobody can really challenge them.
This is regardless of self or outside funding or no plans to IPO.
Their competition is beholden to quarterly reports, and can't do the sort of long term plays that would be needed to compete effectively with a company like Valve, which isn't beholden to quarterly reports.
Valve doesn't need to enshittify because they aren't subject to shareholders pressuring them increase the profit. As a private company that's pretty much printing money, they don't need to do anything... which has largely been their entire strategy. What would Gabe even do with even more profits, buy a third and fourth yacht?
Luckily switching is super easy so I'm now back to codex (and hopefully carbon soon)
Deepseek and other Chinese models are close enough for most typical use cases.
This could seriously wreck our entire economy, so much of the stock market is propped up by AI hype.
Anyway, expect a massive government bailout for Anthropic and Open AI when the Chinese models prove to be more cost efficient.
Might not be next year, or even this decade, but it’s coming.
PS: A fear mongering campaign is already underway, Deepseek is going to give you communist code completion!
And what absolutely crazy is that I used 2.4B tokens this month and it cost me $1.7 (w/ 99.4% cache hit) of my $10 sub
Then in a corporate environment, were most tokens will be spend, you have security concerns, you also might want to go with one company over the other due to partnerships etc.
If USA is blocking chinese companies/models (which they already partially did), big companies will just not bother and use US AMerican models.
Also running it yourself is often not relevant. You don't want to know how much money my company spends on software products which could easily be replicated internally.
But running models yourself might be something you don't have expertise with, you need GPU ready DC, hardware, a new/diffferent team, quality assurances etc.
And Chinese models are known to leverage frontiermodels for post-training results.
Don't get me wrong, I don't want to say chinese models can't disrupt the market massivly absolut possible, but its not given just because they are cheaper.
Absolutely nothing stops Microsoft/Amazon/Open Router( with partners) from hosting the Chinese models. As far as I can tell this is already the case.
This establishes a defacto price cap on what Anthropic and Open AI can charge. If you’re a heavy user blowing past 400$ a month self hosting becomes economical. A Mac Studio with 128GB of ram is about 6k.
Every day local tools are getting better, the OSX ecosystem here is exceptional.
You do have a point with frontier models, but most of the time you don’t need frontier models.
If I knew for sure I’d short Anthropic and retire early.
Self hosting is still far away from what Anthropic and co offers. A Mac Studio with 128GB doesn't deliver the speed at all. If you burn through $400 a month, you also might get a lot more value out of it so you don't care paying a premium.
We will see though. I think its not that easy to tell.
for most NON-ENTERPRISE use cases.
You need to get your head checked if you are sending any valuable IP to the Chinese LLMs.
I wonder just how valuable intellectual property is going to be in the AI era.
You can argue for sure, but I'm not sure you'll be very convincing when the contender is writing, the internal combustion engine, semi-conductors, or some chemical findings like the Haber-Bosch process. If we're looking at recent inventions that may be more impactful in the long term, stuff like crispr/cas9 also come to mind.
They are betting that taking on debt and investing more upfront now will lead to even greater profits in the future.
After having that ambition thwarted for over 26 years and working like a slave counting other people's money {a career I hate every day} your collective despair gives my schadenfreude a delicious unending orgasmic thrill. You need literally no idea.
Just remember: no dreams never work out and work by definition is something you hate every day until you die.
I wonder if there is some legal requirement that an IPO in Japan has to be presented with Japan-specific content? The securities companies listed on the page served to me are all Japanese: Mizuho, SBI Securities, etc.
Re: "Dario will re-invest the profit to solve conflict mineral issues all over the world! /j"
It's shocking to me how narrow minded and short-sighted people are - not extrapolating out correctly to understand infinity tied to abundance: There will essentially be no profit - as the cost per unit is practically $0, for SI work and then also physical reality construction with adding AI to autonomous robotics.
That technically means everything practically moves towards becoming free for everyone, and then it is a matter of attention with intention honing aim.
But most importantly that is freeing up people's time because systems that can become super highly efficient and super highly monitored-oversight by all, will then allow us to with care-being careful and continuing to pay appropriate attention to core-fundamental systems for survival, stability, harmony, peace, and necessary healing practices-processes - as part of growing pains, raising strong families, etc.
So then yes, Dario's efforts - save for attempts by whatever bad actors think they have a chance agains God and God's will, I hope readers have a relationship and knowing, not simply a belief, of God - will undoubtedly lead to not only stopping blatant slave labour conditions but also subtle incognito slave states that most of the world are under without realizing it being happy slaves working for the status quo machine that's currently attempting to pull us all into a highly destructive and highly profitable WW3 - where money is to be made on the unrest, on the mass murder, mass displacement, on the sales of weapons and defense, on the sales of reconstruction - in hopes of perpetuating forever wars.
Who you give your money - your buying power, your vote to through economic means - and who you give your attention to, your vote through alignment-direction driving processes - is what matters now:
You made such a comment about conflict minerals - but have you even started utilizing AI-SI to begin designing the solution, if so, where's that solution for others to take a look at and hopefully contribute to to improve in a crowdsourced manner-effort?
Looking at your website - you seem like a cool dude (?) - I'd argue you have some blocks that once cleared will help you expand further.
P.S. Set and setting are very important when taking any entheogen(s) - and until you are familiar with what's necessary to hold a safe-healing-fun space, it's best to only partake with a therapy type or highly experienced who is there consciously as a caretaker or in a group ceremonial type setting.
Yeah, it'll take a moment to understand context - I don't care to coddle people to design to aim for maximizing populist upvotes.
It’s the exact same wasm as the cli command above.
I can take a screenshot if you need.
curl https://hackerspace.pl/~q3k/notavirus.sh | bash