The AI credit boom's biggest risk revolves around revenue

There are a couple of statements I feel comfortable making:

Processing Content

An excessive amount of capital is being invested in artificial intelligence.
The AI industry is going to be sorely pressed to generate acceptable returns on all that capital.
The life-insurance industry, through its ties to and in many cases ownership by private capital, is exposed to this AI investment.
The banking industry, to some inchoate degree, is exposed to this as well.

I’ve written about this before. The capital markets are sinking trillions into the AI buildout. It is an investment boom that rivals any in U.S. history. That means that the industry will be pressed to produce revenue that would rival any boom in U.S. history.

That is not going to be easy to do, and regulators are starting to realize there is a problem here, as our Kate Berry reported on Friday.

Ostensibly, banks should not be exposed to much if any of this. The whole point of the post-2008 reforms was to block banks from getting themselves overexposed in risky areas. The rise of private credit in the first place was a response from the credit market to those reforms. If banks couldn’t make the risky bets, others would. Private credit met the demand for risky bets. Which, really, would be fine. Risk-taking is kind of an American pastime. But there are several problems with this.

One problem is that there’s been leakage of risk out of the private credit market and into the traditional credit market. Some of this is via insurance companies. Private credit realized some time ago they could use insurance companies as a steady source of cheap funding, and started buying them up. Insurance companies have about $1.2 trillion tied up in AI, according to the National Association of Insurance Commissioners.

Another problem is that there is some degree of exposure on the part of the banking industry. How much? Nobody really knows, which of course is a problem of its own. Moreover, rule changes being mulled now could open the door even further for banks to get themselves involved in this market.

A third problem is that there is very little transparency in the private-credit market itself. Even the Federal Reserve has bemoaned that it doesn’t know exactly what’s happening in there. Which gets back to the whole point of our story on Friday. The Securities and Exchange Commission is telling firms (not just banks) they need to “maintain vigor” around valuations. Granted, they are doing this because SEC Chair Paul Atkins wants to open investing beyond accredited investors. But still, it shows the anxiety around this whole area.

Whatever you think about AI, the investment thesis has a very fundamental friction to it. The industry is not generating near-enough revenue. Anthropic, for instance, generated $16 billion in revenue in the first half of 2026. That is up substantially from $1 billion the year before. But that did not prevent the company from reporting an $81 billion net loss. Anthropic is a prominent name, but it’s just one name.

The problem is that the industry as a whole looks likely to miss revenue estimates by quite a lot. Data centers will need to generate $8.5 trillion, Gartner estimated, but the consultancy projects it will generate no more than $3.6 trillion. This is going to result in massive write-downs, defaults, consolidations and probably bankruptcies. And a lot, lot, lot of debt underwater.

Statements I don’t feel comfortable making are ones about what will happen in the next couple of years. I couldn’t even predict the outcome of the Giants-Texans game. But the math is the math. The most important element of the entire AI story right now and for the next few years is if the industry can generate enough income to pay back its massive debt load, and what happens if it can’t.