The big players in AI aren’t highly leveraged

It’s not traditional leverage but the recent deals being announced where the AI companies are raising money from Microsoft, Nvidia, Amazon, Google, AMD, Oracle, etc. and paying it back in stock or purchase commitments have a certain circular bootstrappy notion to them. The formulas for the valuations rely on feedback loops that are less stable and might create runaway feedback conditions at the slightest hiccup.

In any highly capital intensive business, you always run the risk that the thing you build is worth less than the cost it took to build it. And when that happens, collapses can happen pretty quickly, as everyone invested in these companies rushes towards the offramp.

I can think of a few catalysts that could trigger that initial realization that the thing made isn’t actually worth the cost to build it:

• A new model comes out from a competitor that was cheaper to build and almost as good. (Deepseek reminded everyone that this might happen.)
• New money stops coming in and the companies building things have to tighten their belts. This could be driven by a failure to monetize as much as previously modeled, so that the value of the company itself is questioned.
• Some kind of legal flaw threatens the entire foundation of some expensive models.
• Some kind of technical flaw causes one company’s flagship model to lose the race against other companies.
• Some key personnel are incapacitated in a way that robs the company of its momentum (this almost happened with the board of directors revolt at OpenAI).
• Something else I haven’t thought of.

But once a hiccup happens, something built on so many self-reinforcing loops is less resilient against the unknown, the chaos of the real world.