Garbage. You can't include training by the companies that develop an llm in the comparison against companies that merely use the same llm. Apples and potatoes.
It'll get paid from revenue, not by redirecting employee salaries. All that AI+compute is literally what customers pay Anthropic for.
Big AI labs are not software companies where payroll dominates expenses. They're capex-heavy industrial entities; it just so happens that the "machines" (whose output they sell) are nominally the same category as the devices that their knowledge worker employees use on their desks.
I wonder if they ever will be. If the chinese open source models are only 3-6 months behind every major frontier model release, I can't see the business model. GLM-5.2 is supposedly on par to Opus depending on the case. And everybody and their mother can run that model in their datacenter and charge Dollars for tokens.
There is distillation going on where chinese providers give the model lots of outputs. We don't live in a world where chinese providers are not doing this so we can't compare the advantage of this distillation, but there is some advantage to it otherwise they wouldn't do it.
If Anthropic can block distillations somehow (which are fair game imo given that Anthropic et al did the same with the written works of mankind), then they might stop or slow down the chinese from catching up.
Chinese also have like 40% of the AI researchers of the world, plus they have access to a lot of cheap labour for writing training data. I'm sure an hour of training data creation from one of China's 162 million university educated people is much cheaper than an hour of work from one of US's 97 million. Probably still cheaper than someone from the grand area.
China is behind in AI chips/GPUs but they are catching up. One thing where they have a hard dependence on outside is their energy imports: they have to import a lot of stuff from third party countries. The US on the other hand is energy self sufficient.
I don't know, compute is compute. Arguably making complex software with LLMs isn't all that different from training a model to do a thing. You're throwing a lot of compute at the problem and hoping for a stochastic solution. The distinction will become even blurrier with time.
Though I agree it might be informative to split it by industry sector.
... and that actually shows - senior engineers have spent actual paid time to train juniors. Plus they used to spent time contributing to open source projects or Stack Overflow, all the stuff which every company benefits from.
why stop there? Count how long and how much energy it took for evolution to produce that 3 chimp brain that is then educated, and add how long it took culture to produce the knowledge in text books for said education to be possible.
OpenAI and Anthropic aren't charities, so whatever cost they inccur for training will be passed down to the companies using the models. So you absolute should include it.
OpenAI and Anthropic aren't charities, so whatever cost they inccur for training will be passed down to the companies using the models
You should, but with two important caveats. First, you don't know what their amortization schedule is like so you don't know what the impact on the pricing will be (are they going to pass the cost on over 5 years or over 20 years?), and second they may go bust before paying the cost down so they may not get a chance to pass it all on. If someone buys the company then they'll get a discount on the value, which means the training costs are just eaten by the investors.
Anthropic spends [...] about $2m of compute per employee per year against a likely all-in comp of $500k+.
The rest of the software market trails. The top 1% of companies spend $89k per engineer per year on AI
This framing makes no sense. The reason Anthropic spends so much on compute per employee is that they are building models. Anthropic employees aren't opening Claude Code and spending $2m in inference every year, so comparing it to other software companies, where AI expense is mostly inference, is completely incoherent.
Yes, the cost has to be passed down eventually, but it's not passed down to one company; it's passed down to all of Anthropic's customers, so the actual share of that money will be distributed among Anthropic's clients.
Look, I 100% agree with the idea that OpenAI and Anthropic are both unsustainable companies that have dug themselves so far into a debt hole that, most likely, the only way they'll be rescued is with government intervention, but this is still a terrible article.
Apples and potatoes are both something people will need to eat if we want to see it from the human utility perspective, and they both require some land space to be allocated for their culture (though one can of course conjugate both culture).
If you want to take the DDG LLM summary at fate value, apples are lower in calories and sugar but higher in fiber compared to potatoes, which are richer in vitamins and minerals like potassium and vitamin B6. Overall, apples provide more dietary fiber, while potatoes offer more protein and essential nutrients.
Comparison rarely lead to one obvious all superior option that discard every other considerations.
the saying "comparing x and y" implies that you compare something that one of them can't compete ; if people praise the softness of the skin first and foremost, comparing apples and potatoes won't lead interesting results
Evian use 1.25 million litres of water per employee per year. When can we expect other non-bottled-water corporations to rise to this level of water usage?
Working regularly with AI is like managing a small team of unbelievably knowledgeable, very smart, and occasionally crashingly naïve junior developers. Because they're so knowledgeable and smart, they can get a lot done very quickly. Because they make a proportion of howling errors, you have to keep a close eye on them -- or carefully train another agent to do it for you, in which case you now have to keep a close eye on that agent as well.
So, overall, you get more done that without AI, at the cost of spending almost all of your time writing specs and doing code review and almost none of it writing code.
Do you get 3.3x the work done? Probably not. Do you get 2x the work done? I think maybe, if you can hack the dynamics of the new job as a manager of eager robots. For me the jury's still out on the second point.
Open-weight models are going to completely shatter these forecasts. It takes a little more effort – right now, probably won’t be true in three months – but you can achieve the same at 1/10th of the cost.
> but you can achieve the same at 1/10th of the cost.
For some tasks, sure. But not for all tasks. And for some tasks, cost per token is irrelevant if it provides real benefits that are oom compared to what you had.
Local models are indeed becoming "good enough" for some tasks, but there are still tasks that they can't touch. There's a recent benchmark for kernel writing. Fable wrote a kernel that provides ~30% more throughput per unit of compute compared to the latest Opus max / gpt max. Does it matter how much that session cost in terms of one session if you can take that kernel, deploy it on your inference fleet and "magically" get 30% more tokens served to your clients? There are companies that would pay millions for such a "leap". Because they can make more millions down the line.
You're looking at the status quo and ignoring the trajectory. The best current open models are about as good as closed models from ~1.5 generations ago. The rate of improvement of all models is converging to zero. It follows that in a few generations, open models inferencing will be about as good as closed model inferencing.
The problem is going to become that there's no incentive for anyone to run the stupidly-expensive training phase. May God have mercy on the stock market.
That’s true, there will always be demand for ultra-intelligent assistants, especially if they surpass what humans can achieve at similar cost. For the other 90%, the average frontier model will be good enough.
I don't disagree with you, but it's important to pay attention to where the money is. Cheap non frontier models is something that Anthropic and open AI could do too, but who's willing to pay a premium for using them? It will be like competing to sell rice, lots of demand at Rock bottom margins.
This includes the cost to AI companies of training their models, which constitute the thing the other companies are buying from them when they “spend on AI”?
Isn’t that type of spending more of a direct input to the thing they (the Anthropic-type companies) are selling?
Wouldn’t we expect non-AI-selling companies to spend less on making AI, and more on making what they make?
I think that token usage by engineers continues to increase, probably at a very high rate for many years (we are in the middle of the S curve of adoption and it isn’t yet clear where this will plateaux) but an increasing percentage of those tokens are cheap, because we use expensive models for goals and design and cheap models for implementations and workflows.
The layoffs are irrelevant to the discourse. It's typically considered by management to be good, for mature companies, to periodically fire as many employees as they can sustain without visibly impacting operations, and then re-hire cheaper workers only where strictly necessary. This allows them to keep costs down, reduce risks of excessive worker entrenchment, and overcome the drawbacks of contingent hiring-sprees.
Excuses for these exercises will vary, AI is just the latest; but it's fundamentally just a labor-containment/efficiency-seeking strategy.
I think its a fallacy to believe people like Zuckerberg or any other stupidly rich person aren't extremely calculative about this. I am very sure they have surrounded themselves by top tier engineers making very informed decisions while their top tier marketing teams make very calculated decisions on how its expressed to the public. The public generally is NOT in favor of AI outside of tech circles so it makes sense to communicate critique of AI to the public.
A missing thread by the author for how Anthropic's training expenses becomes expenses for employee workplace expenses.
And this is before we start adding Anthropic engineer's ability to use it's tools/models for far less than market price.
I've not seen anyone yet implement a true cost to productivity assessment or guardrails for AI usage yet. Sure this is hard to do with people, but performance management is a well understood field with a hundred years of practice for knowledge workers.
We don't get unlimited hiring budget, so we also won't get unlimited token budgets, and we as the operators will be responsible for the productivity of our agents.
What does performance management for engineers look like when dollar token cost is included in reviews? I think it's going to change a lot of assumptions and a lot of strategy around AI use.
Ignoring the bizarre inclusion of training compute for the AI company estimates, the other comparisons are still valid.
> The rest of the software market trails. The top 1% of companies spend $89k per engineer per year on AI, 40% of a fully-loaded $224k senior engineer salary. The median spends $137. That is the gap : ... 0.4x at the top of the market, near zero at the median.
So it's not more expensive than an engineer it's 40% as expensive, and for many companies use-cases the cost is virtually negligible.
Even here in Europe where developers are much cheaper than in the US, it still makes sense to pay for the LLM Enterprise subscriptions.
My company has a Claude Code and Codex one and I use Claude Code because I am more familiar with it. That said, I just use Opus for planning and Sonnet for implementation and it's pretty cheap. Codex seems decent too so I should try it out some more.
But you can get an awful lot done even with just like $200 a month at API pricing if you are careful not to waste a powerful model on an easy task, or carry around a bloated context window etc.
I think a lot of the 'tokenmaxxing' people spending thousands every month are simply using the tools ineffectively (like having loads of Opus agents doing tasks that Sonnet or even Haiku could do). I suspect this will only get worse now with the release of Fable, but Anthropic must love it.
When you say the cheaper models do you mean like Deepseek or GLM? I haven't tried those but they look interesting. It'd be nice to shift to open weights and not be tied to one company.
With cheaper models I really meant cheaper subscriptions but used the wrong vocabulary. We still use Claude Opus (if thats what 4.6 is?). We just have the 20 bucks subscription and I barely use up my token limits in my day to day work.
I often wonder what kinda features other devs implement compared to me, if they need that many tokens?
It kind of feels impractical to bloat up an app with features one barely understands? I've just been reading about these devs using x-amount of tokens, having that y-amount of steps perfected AI workflow, but none of them ever talk about what they actually implement all day...
Is not one or the other. AI is a tool for the Engineer. Costs more? Depends on how you use it. You can reduce AI costs in multiple ways, accepting the tradeoffs.
Excellent, with stunning insight like this, you can see why this VC is earning the big bucks.
This is almost economics level of line projection.
It would be good to understand _why_ anthropics "AI" bill is so high. First, They are going to be renting a lot of inference compute just to service customers (Meta's Capex bill is about 2x its wage bill) It then also needs a huge amount of infra to both run training and experimentation. THats probably a third of the cost. (storage and physical infra to get the most out of storage and compute is hard. Then getting it reliable, so that shit state doesn't propogate across the shared memory plane is very hard.
The other thing to note is that claude usage inside anthropic is tiny compared to the customer's usage. even with uber agents at "mythos++" its going to be at best a few thousand servers. not like the massive fleet needed to serve the paying customer.
So using anthropic as some sort of rational target to base any kind of prediction is madness. Its like looking at lyons tea rooms and going yeah, every company is going to spin up an R&D arm to make a company specific computer: https://www.sciencemuseum.org.uk/objects-and-stories/meet-le...
ALSO this assumes that the current way of running LLMs is the way forward. Custom software is expensive (in both time and tokens) to look after, its much easier and cheaper to buy it in from SaaS companies and let them figure that shit out. (yes I know SaaS apocalypse, but you are paying for real world experience, and a packaged way of doing things, rather than experimenting your self, where in a lot of cases the company doing the experimentation doesn't know what its doing)
I'm not a VC guru but in my opinion you can't include the time and money it takes to grow a tree and mine the iron to compare the time it takes to hammer in a nail with a hammer versus using your fist.
That’s how policy makers and concentrated decision power class get completely disconnected from actual resources at stake and what actual constraints need to be weighted. If a job require to put a blindfold and a sound blocker headset preventing to hear the things people scream, people in the role will happily accelerate against the wall the are induced to ignore.
This is not even specific to capitalism or VC mind you. Look how PRC led to the Great Chinese Famine. That’s why actual democracies (not the inter-elected aristocraties ), despite all their downsides, are so damn interesting. Corruption, negligence, or mere error with catastrophic follows, is easily spread in a situation where small core of individuals monopolize greatest part of decision weight, but is logistically impossible to achieve in a system optimized for widespread and highly redundant power responsibilities.
> Anthropic spends 2.3x its payroll on compute.1 With ~5,000 employees & roughly $10b in inference & training spend in 2026, that works out to about $2m of compute per employee per year against a likely all-in comp of $500k+.2
> The rest of the software market trails.
This shows how VC firms see things and why we have such a lopsided market where grift rises to top easily.
Yes the rest of the software market trails in comparision to the compute costs at Anthropic if you including training the actual models. Like is this the insight? Biggest AI company spends a lot of money to make AI models?
Sure you can find anthropic's business model risky/not feseable but using this as your starting point shows a lack of basic understanding at best and malicious intent to make a stupid point at worst
Big AI labs are not software companies where payroll dominates expenses. They're capex-heavy industrial entities; it just so happens that the "machines" (whose output they sell) are nominally the same category as the devices that their knowledge worker employees use on their desks.
If Anthropic can block distillations somehow (which are fair game imo given that Anthropic et al did the same with the written works of mankind), then they might stop or slow down the chinese from catching up.
Chinese also have like 40% of the AI researchers of the world, plus they have access to a lot of cheap labour for writing training data. I'm sure an hour of training data creation from one of China's 162 million university educated people is much cheaper than an hour of work from one of US's 97 million. Probably still cheaper than someone from the grand area.
China is behind in AI chips/GPUs but they are catching up. One thing where they have a hard dependence on outside is their energy imports: they have to import a lot of stuff from third party countries. The US on the other hand is energy self sufficient.
Anthropic was profitable last quarter.
Though I agree it might be informative to split it by industry sector.
Compare AI costs per-engineer-salary-dollar, because more expensive engineers probably need more expensive AI.
Let's see how this works out in the long run. For a historical analog, more expensive engineers don't use more expensive computers (by and large).
You should, but with two important caveats. First, you don't know what their amortization schedule is like so you don't know what the impact on the pricing will be (are they going to pass the cost on over 5 years or over 20 years?), and second they may go bust before paying the cost down so they may not get a chance to pass it all on. If someone buys the company then they'll get a discount on the value, which means the training costs are just eaten by the investors.
Anthropic spends [...] about $2m of compute per employee per year against a likely all-in comp of $500k+.
The rest of the software market trails. The top 1% of companies spend $89k per engineer per year on AI
This framing makes no sense. The reason Anthropic spends so much on compute per employee is that they are building models. Anthropic employees aren't opening Claude Code and spending $2m in inference every year, so comparing it to other software companies, where AI expense is mostly inference, is completely incoherent.
Yes, the cost has to be passed down eventually, but it's not passed down to one company; it's passed down to all of Anthropic's customers, so the actual share of that money will be distributed among Anthropic's clients.
Look, I 100% agree with the idea that OpenAI and Anthropic are both unsustainable companies that have dug themselves so far into a debt hole that, most likely, the only way they'll be rescued is with government intervention, but this is still a terrible article.
Assuming their investors win the bet they placed on them. Which isn't given.
Bandwidth isn't free, and all my life I've been told that piracy is theft.
If you want to take the DDG LLM summary at fate value, apples are lower in calories and sugar but higher in fiber compared to potatoes, which are richer in vitamins and minerals like potassium and vitamin B6. Overall, apples provide more dietary fiber, while potatoes offer more protein and essential nutrients.
Comparison rarely lead to one obvious all superior option that discard every other considerations.
Evian use 1.25 million litres of water per employee per year. When can we expect other non-bottled-water corporations to rise to this level of water usage?
So, overall, you get more done that without AI, at the cost of spending almost all of your time writing specs and doing code review and almost none of it writing code.
Do you get 3.3x the work done? Probably not. Do you get 2x the work done? I think maybe, if you can hack the dynamics of the new job as a manager of eager robots. For me the jury's still out on the second point.
For some tasks, sure. But not for all tasks. And for some tasks, cost per token is irrelevant if it provides real benefits that are oom compared to what you had.
Local models are indeed becoming "good enough" for some tasks, but there are still tasks that they can't touch. There's a recent benchmark for kernel writing. Fable wrote a kernel that provides ~30% more throughput per unit of compute compared to the latest Opus max / gpt max. Does it matter how much that session cost in terms of one session if you can take that kernel, deploy it on your inference fleet and "magically" get 30% more tokens served to your clients? There are companies that would pay millions for such a "leap". Because they can make more millions down the line.
The problem is going to become that there's no incentive for anyone to run the stupidly-expensive training phase. May God have mercy on the stock market.
How many software developers were working on code like the one you describe?
Isn’t that type of spending more of a direct input to the thing they (the Anthropic-type companies) are selling?
Wouldn’t we expect non-AI-selling companies to spend less on making AI, and more on making what they make?
It is really crazy people didn't think this through.
Excuses for these exercises will vary, AI is just the latest; but it's fundamentally just a labor-containment/efficiency-seeking strategy.
They were lucky with the ad empire he built and that‘s it.
We don't get unlimited hiring budget, so we also won't get unlimited token budgets, and we as the operators will be responsible for the productivity of our agents.
What does performance management for engineers look like when dollar token cost is included in reviews? I think it's going to change a lot of assumptions and a lot of strategy around AI use.
> The rest of the software market trails. The top 1% of companies spend $89k per engineer per year on AI, 40% of a fully-loaded $224k senior engineer salary. The median spends $137. That is the gap : ... 0.4x at the top of the market, near zero at the median.
So it's not more expensive than an engineer it's 40% as expensive, and for many companies use-cases the cost is virtually negligible.
Even here in Europe where developers are much cheaper than in the US, it still makes sense to pay for the LLM Enterprise subscriptions.
Does it though? I do not see any advantages in my day to day job over using the cheaper models.
But you can get an awful lot done even with just like $200 a month at API pricing if you are careful not to waste a powerful model on an easy task, or carry around a bloated context window etc.
I think a lot of the 'tokenmaxxing' people spending thousands every month are simply using the tools ineffectively (like having loads of Opus agents doing tasks that Sonnet or even Haiku could do). I suspect this will only get worse now with the release of Fable, but Anthropic must love it.
When you say the cheaper models do you mean like Deepseek or GLM? I haven't tried those but they look interesting. It'd be nice to shift to open weights and not be tied to one company.
I often wonder what kinda features other devs implement compared to me, if they need that many tokens?
It kind of feels impractical to bloat up an app with features one barely understands? I've just been reading about these devs using x-amount of tokens, having that y-amount of steps perfected AI workflow, but none of them ever talk about what they actually implement all day...
This is almost economics level of line projection.
It would be good to understand _why_ anthropics "AI" bill is so high. First, They are going to be renting a lot of inference compute just to service customers (Meta's Capex bill is about 2x its wage bill) It then also needs a huge amount of infra to both run training and experimentation. THats probably a third of the cost. (storage and physical infra to get the most out of storage and compute is hard. Then getting it reliable, so that shit state doesn't propogate across the shared memory plane is very hard.
The other thing to note is that claude usage inside anthropic is tiny compared to the customer's usage. even with uber agents at "mythos++" its going to be at best a few thousand servers. not like the massive fleet needed to serve the paying customer.
So using anthropic as some sort of rational target to base any kind of prediction is madness. Its like looking at lyons tea rooms and going yeah, every company is going to spin up an R&D arm to make a company specific computer: https://www.sciencemuseum.org.uk/objects-and-stories/meet-le...
ALSO this assumes that the current way of running LLMs is the way forward. Custom software is expensive (in both time and tokens) to look after, its much easier and cheaper to buy it in from SaaS companies and let them figure that shit out. (yes I know SaaS apocalypse, but you are paying for real world experience, and a packaged way of doing things, rather than experimenting your self, where in a lot of cases the company doing the experimentation doesn't know what its doing)
This is not even specific to capitalism or VC mind you. Look how PRC led to the Great Chinese Famine. That’s why actual democracies (not the inter-elected aristocraties ), despite all their downsides, are so damn interesting. Corruption, negligence, or mere error with catastrophic follows, is easily spread in a situation where small core of individuals monopolize greatest part of decision weight, but is logistically impossible to achieve in a system optimized for widespread and highly redundant power responsibilities.
https://en.wikipedia.org/wiki/Great_Chinese_Famine
> The rest of the software market trails.
This shows how VC firms see things and why we have such a lopsided market where grift rises to top easily.
Yes the rest of the software market trails in comparision to the compute costs at Anthropic if you including training the actual models. Like is this the insight? Biggest AI company spends a lot of money to make AI models?
Sure you can find anthropic's business model risky/not feseable but using this as your starting point shows a lack of basic understanding at best and malicious intent to make a stupid point at worst