Its not just that all this money is being gambled on what are very high risk businesses but that investment money is hurting a huge amount of the rest of the market. The longer computer component shortages continue the bigger the blast radius in terms of other products will be. Its not just RAM, graphics cards and SSDs anymore, PCBs generally have now been added to the list. We are pouring a lot of investment into fuelling these shortages that also increases our energy consumption on a bet that is quite high risk to the detriment of a lot of other businesses and the economy generally and our climate.
I have no idea how this is all going to play out but widespread shortages in computers is just the tip of the spear on the impact coming in the coming years of all this money being poured in to take those components away from every day products.
I don't like this framing that it's a bet. If Moody's and S&P are rating this debt as being relatively safe, then it's not being perceived as a bet by the parties involved. Maybe they're wrong to not perceive it that way, but the article doesn't really try to argue that point. For the most part, it just appeals to this sense that this is wrong, and the people that are actually in the weeds here don't understand a problem that the rest of us do.
> If Moody's and S&P are rating this debt as being relatively safe, then it's not being perceived as a bet by the parties involved.
FWIW Moody's and S&P gave AAA ratings to tens of thousands of mortgage-backed securities and CDOs whose later implosion almost took down the entire global economy in 2008.
To help fund this complicated arrangement, $2.4 billion of the debt was sold to an insurance company called Athene, which manages life insurance policies and annuities for retirees, according to a Moody’s Ratings report.
The deal was part of an increasingly popular playbook: tapping insurance companies, which are historically among the most cautious investors, to make loans to tech titans like Mr. Musk and further fuel the A.I. boom.
They need a diverse set of low-risk vehicles to invest their assets under management in. Who quantifies the risk here? From the article:
> Making possible this flood of insurance money into risky A.I. deals are credit rating agencies like Moody’s and S&P Global Ratings. They have blessed this rapidly expanding pool of debt with relatively safe credit ratings, which meet the criteria that insurance firms need to invest in a particular asset.
This is a similar problem that occurred during the 2008 financial crisis. The ratings agencies seem to be incentivized to miscategorize the actual risk. Folks buy risky assets assuming they're safe, because they've offloaded the due diligence to the ratings agencies, but the ratings agencies are understating the risk (I assume they financially benefit and lack punishment for miscategorizing the risk).
3. Goldman Sachs settled with the SEC in 2010 for $550M.
The Goldman Sachs trial was of note because while the ratings agencies themselves weren't on trial, the deal shows how ratings made these products sellable to investors. And that's the problem, IMO. The SEC should be stepping in to protect investors, as per their mission.
Apollo, the parent company of Athene, is one of the most sophisticated entities in the world at modeling this risk. They care about the assets' credit ratings because they impact their risk-based capital requirements; they're not outsourcing the risk modeling to the credit rating agencies. This also isn't a 2008-like case of underestimating correlation: everyone in this space is well aware of the potential for contagion.
It's true that credit rating agencies' incentives are still poorly aligned, but again, sophisticated players aren't relying on their risk modeling anyway.
If rating deals brings in large fees, then I think it's a structural conflict of interest for whomever is doing to the ratings to understate the risk of the deals, so that deals keep bringing in those large fees.
Yep, that's exactly the conflict of interest I'm talking about. But Apollo doesn't just look at e.g. Moody's B1 rating for CoreWeave unsecured bonds and say "only 12% chance of default over 5 years, good enough for me." Apollo puts together its own view of the risk and invests accordingly, and unlike the credit rating agencies, it has a very strong incentive to get that modeling right.
Why does anyone take the ratings agencies seriously at this point? They really think AI debt is safe… just take 30 seconds to look at the numbers and speed and hold your horses partner this is the Wild West.
It takes 30 secs to find out that google spent 45$ billion last quarter on ai. The company made 39$B. So they only thing it cant cover is ~6B. And this pattern is roughly true for all the hyperscalers. A $6B deficit over 3 months is hardly risky.
Because that's all it takes for these large portfolio managers to wash their hands of performing due diligence to protect their investors. God forbid they themselves had to measure risk; that would involve actual work. Instead, they can pay someone else (where that someone else doesn't have the incentives and disincentives in place to be accurate, wittingly or otherwise).
Because it provides a higher yield than safer fixed-income assets to compensate for its risk, and that risk is somewhat uncorrelated with credit risk elsewhere in insurers' asset portfolios, meaning that insurers can earn higher risk-adjusted returns on their assets (and pass much of it through to policyholders in the form of stronger policy benefits).
"Making possible this flood of insurance money into risky A.I. deals are credit rating agencies like Moody’s and S&P Global Ratings. They have blessed this rapidly expanding pool of debt with relatively safe credit ratings, which meet the criteria that insurance firms need to invest in a particular asset."
Sounds very familiar. Ratings agencies are giving soft ratings to a new, huge source of ratings revenue that they don't really understand and haven't really stress tested.
Goodhart's Law strikes again. We just can't get enough of rubber stamps - which is why theater like SOC2 compliance has become so pervasive.
It is nice to have a broad list of standard concerns and validate each one, but that is not an actual guardrail at all or a panacea for shifted responsibility, despite what upper management tends to believe.
Is this structurally different from the "private credit" market for obtaining debt that other journalists reported on at least a year ago? I guess either way, I shouldn't complain that the NY Times is finally doing so much transparency coverage around the AI industry and its players...
The difference between this and 2008 is that the mortgages were owned by your bank and the average person's bank deposits are mostly FDIC insured unless you're keeping over 200k in an account. Most people weren't heavily invested in Lehman Bros but insurance is much more directly connected to people through car, home, life, health, etc. A few big misses on AI projects could cause a major jacking up of insurance rates by devaluing the value of invested premiums, the same way that big fires in California hit insurers on the other side by creating a lot more covered losses to pay up for.
Cash management account at a brokerage that sweeps your cash to multiple FDIC entities to maximize your FDIC coverage. "FDIC Sweep Program" are the keywords to use to learn more.
New subprime mortgages between 04 and 07 amounted to $1.3 trillion. Total AI infrastructure capital committed is over $5 trillion.
> nor are any of the companies actually critical to the US economy.
The problem isn't that Anthropic and OpenAI will go bankrupt. The problem is the disruption of the financial markets that happens as a result of their downfall. This will have global consequences.
The $5 trillion AI infrastructure commitment is roughly 15% of current GDP, although the two numbers aren't entirely comparable. The subprime number is a pool of loans outstanding at a point in time. The AI number is a debt-fueled capital commitment spread over several years; about 70% of it is debt-financed.
I don't think this is just limited to Insurance which I think a lot of the comments are missing. There is contagion across the whole financial ecosystem, from pension funds to the shadow market of private capital. There are inherent risks with the amount of capital expended already and or promised, and quite frankly I think we're deluded if we think there isn't a risk to the financial system. Ok it might not be sub prime 2.0, but let's not delude ourselves into thinking $5 trillion of committed capital isn't an absurd number
I have no idea how this is all going to play out but widespread shortages in computers is just the tip of the spear on the impact coming in the coming years of all this money being poured in to take those components away from every day products.
FWIW Moody's and S&P gave AAA ratings to tens of thousands of mortgage-backed securities and CDOs whose later implosion almost took down the entire global economy in 2008.
The deal was part of an increasingly popular playbook: tapping insurance companies, which are historically among the most cautious investors, to make loans to tech titans like Mr. Musk and further fuel the A.I. boom.
But why are they buying it?
> Making possible this flood of insurance money into risky A.I. deals are credit rating agencies like Moody’s and S&P Global Ratings. They have blessed this rapidly expanding pool of debt with relatively safe credit ratings, which meet the criteria that insurance firms need to invest in a particular asset.
This is a similar problem that occurred during the 2008 financial crisis. The ratings agencies seem to be incentivized to miscategorize the actual risk. Folks buy risky assets assuming they're safe, because they've offloaded the due diligence to the ratings agencies, but the ratings agencies are understating the risk (I assume they financially benefit and lack punishment for miscategorizing the risk).
1. S&P settled with DOJ in 2015 for ~$1.375B.
2. Moody's settled with DOJ in 2017 for ~$ 864M.
3. Goldman Sachs settled with the SEC in 2010 for $550M.
The Goldman Sachs trial was of note because while the ratings agencies themselves weren't on trial, the deal shows how ratings made these products sellable to investors. And that's the problem, IMO. The SEC should be stepping in to protect investors, as per their mission.
[1] https://www.justice.gov/archives/opa/pr/justice-department-a...
[2] https://www.justice.gov/archives/opa/pr/justice-department-a...
[3] https://www.sec.gov/newsroom/press-releases/2010-123-goldman...
[4] https://www.sec.gov/about/mission
If the ratings companies aren’t really doing the right thing, then that’s what needs to be investigated, maybe even shut them down.
Ever read a prospectus for an investment? If it’s 100 pages, 98 are ridiculous legalese.
It's true that credit rating agencies' incentives are still poorly aligned, but again, sophisticated players aren't relying on their risk modeling anyway.
Because that's all it takes for these large portfolio managers to wash their hands of performing due diligence to protect their investors. God forbid they themselves had to measure risk; that would involve actual work. Instead, they can pay someone else (where that someone else doesn't have the incentives and disincentives in place to be accurate, wittingly or otherwise).
Given history and the fact that retirees are the most powerful voting bloc, they're probably right.
Sounds very familiar. Ratings agencies are giving soft ratings to a new, huge source of ratings revenue that they don't really understand and haven't really stress tested.
It is nice to have a broad list of standard concerns and validate each one, but that is not an actual guardrail at all or a panacea for shifted responsibility, despite what upper management tends to believe.
So what are you supposed to do with your cash once you get to 200k?
https://www.investor.gov/introduction-investing/general-reso...
New subprime mortgages between 04 and 07 amounted to $1.3 trillion. Total AI infrastructure capital committed is over $5 trillion.
> nor are any of the companies actually critical to the US economy.
The problem isn't that Anthropic and OpenAI will go bankrupt. The problem is the disruption of the financial markets that happens as a result of their downfall. This will have global consequences.
Also, subprimes represented ~9% of the economy.
AI is ~2% if we're being generous.