A Fed official is asking whether AI is becoming 'too big to fail'

It is certainly far too big to succeed.

Anthropic? absolutely. But there is a strong Jevons paradox component.

OpenAI, Google, or SpaceX have a moat of hardware and energy. If AI is 100x cheaper orgs will use 100x more AI and the benefit of algorithmic improvements will flow to whoever has the hardware to run it.

I don't think consolidation is the issue here, so anti-trust doesn't make a lot of sense. There are at least five major US companies strongly competing with each other. The issue is that total investment has gone far beyond expected revenue because no one is far ahead, everyone is bidding against each other for chips/power, and competition is driving down token price. Ironically, consolidation would fix this because a single winner would be able to charge monopoly prices. Anti-anti-trust. :-)

The sheer scale of the buildout and debt load requires incredible scrutiny. But there's a few reasons why this isn't the radioactive bomb it might intuitively seem like.

One, tech isn't a massive employer. The people being employed in all the downstream businesses, the contractors, electricians, etc are in classically cyclical businesses. If this all goes south, this might not be the employment dip that starts a broad recession. Two, the debt is largely being issued by companies with some of the greatest balance sheets in the world. While the CoreWeaves etc of the world would face elimination, the classic big techs probably wouldn't. Three, it matters who the counterparty is. Right now the debt is distributed through private credit, Wall Street banks, REITs, mutual funds, etc. The contagion would reach through private credit to their LPs like pensions, insurance funds etc but as far as we know these funds are not hyper leveraged and over-indexed toward AI.

But yeah the sheer scale of the debt load just cannot be ignored.

I don't think tech itself is the big employer here in a failure case. Its all the stuff that feeds into the data center build out like construction, electric specialists, all the semiconductor manufacturers, etc. If all those people go out of work at once and there's no other work to pick up the slack, then that could be the unemployment spike that snaps aggregate demand down and starts a recession. When people talk about "too big to fail" this is one of the concerns that makes an industry or company "too big to fail". I don't know if the point of AI is cost savings but in that case we're talking about a technology induced deflationary spiral not a failure of the CapEx spend.

Why aren't tech companies using their mountains of cash to pay for the build out? Why are they borrowing money through Special Purpose Vehicles? Because they want someone else to share in the risk if the build out doesn't work. The SPV allows them to pay a fixed amount for the privilege of off loading some of the risk. I agree that they are obfuscating risk here in a dangerous way but my broader point was that, on its face, this debt load does not have the sort of pernicious systemic failure that makes downturns into deep recessions.