By EOY, two of the largest IPOs in history will belong to companies that were nonprofit research labs with a few dozen employees six years ago. OpenAI and Anthropic are both expected to go public in October, as evidenced by the S-1 prep the labs are going through (detailed legal filing companies must submit to the SEC before going public), let’s look at three consequences:
-
Both companies are spending roughly all the money they have on compute, and both can plausibly absorb large amounts of capital effiecently. In fact, this relates to a concept I was exposed to in a podcast with the CFO of Anthropic, where they discussed the metabolism of LLM companies where instead of metabolising ATP, they’re metabolising money. According to the Anthropic CFO, it seems that when you divide compute among internal use, serving customers, and model training, there is a lot of capital that can be “metabolised” in the form of compute. The IPO is the only way to raise that much without giving Microsoft, Google, and Amazon even more leverage than they already have.
-
Notwithstanding Ant and OAI’s charter language about safety and the long-term interests of humanity, they will now both have public-market shareholders whose interests are quarterly earnings. The bull case is that the PBC structure and what’s left of OpenAI’s capped-profit complex give the boards real cover to refuse demands that would compromise alignment work. The bear case is that they don’t, and the most useful single number to watch is what fraction of float ends up with index funds versus active managers over time, as active managers are much more inclined to intervene on matters of profit.
-
If both companies clear a $1T market cap, AI labs become systemically important to U.S. financial stability in a way they currently aren’t.