The Wrong Question to Ask About a $965 Billion Company

Alan Greenspan died on Monday, and within hours the financial press had reached for the line he is remembered for. The question being asked is whether AI valuations are a bubble. I think that is the wrong question, and I think the rush to ask it is letting a more important one go unexamined.

The frame that returned this week. Crunchbase News published an essay tying Greenspan's 1996 "irrational exuberance" speech to current AI mega-cap marks: Anthropic at a reported $965B post-money, OpenAI at roughly $852B, SpaceX now sitting at the sixth-largest market cap among US companies.¹ Kyla Scanlon posted a tribute on the same theme to a large social audience.² Yann LeCun warned about an AI bubble the week before. The bubble framing, which has been circulating in AI-specialist forums for a year, has now migrated into the mainstream private-market conversation.

The Crunchbase piece is careful. It distinguishes today's leaders from Pets.com — these are companies "priced like winners, not wanna-bes," with real revenue and real enterprise contracts.¹ OpenAI is on track for roughly $11.6B in 2025 revenue. Anthropic crossed $1B annualised in 2024. This is not 1999. The pricing logic is different because the underlying companies are different.

What Greenspan actually asked. Go back to the December 1996 speech. The line everyone quotes is a question, not a claim: "How do we know when irrational exuberance has unduly escalated asset values, which then become subject to unexpected and prolonged contractions?"³ Greenspan was not predicting a crash. He was asking a pricing question — are these assets worth what they trade for, and how would we tell?

That question, applied to AI, has a clean answer in the bull case. If Anthropic captures even a quarter of enterprise knowledge work, $965B is cheap. If OpenAI becomes the default consumer AI layer, $852B is cheap. The valuations are not irrational on their face. They are priced for a specific future — one in which a handful of these companies dominate one of the largest markets ever built. The valuations assume that future is roughly half-priced in.

This is where the bubble frame goes wrong. A bubble is a story about mispriced assets that eventually correct toward fundamentals. What is being priced into Anthropic and OpenAI is not really a fundamentals story at all. It is a structural one — a bet that the AI market will consolidate into two or three winners, and that those winners will hold defensible positions against everyone else for a long time. The valuations are not predictions about technology. They are predictions about market structure.

Who the price assumes loses. If you accept the Crunchbase distinction, these companies are priced like winners, then you have to ask what is being priced into the losers. The hundreds of AI companies that raised at fractions of these marks. The startups that will fold over the next two years because they cannot match the compute spend. The open-source efforts whose viability depends on whether the closed leaders are allowed to stay this far ahead. A $965B mark for Anthropic is a statement about all of them too: the market is pricing in their defeat.

That is not a bubble question. It is a concentration question. And concentration is not corrected by a market downturn the way mispricing is. If AI valuations fall 40% next year, OpenAI and Anthropic are still the only two companies operating at frontier scale. The compute moats stay. The enterprise lock-in stays. The talent gravity stays. A correction in price is not a correction in power.

Where the human stakes actually sit. This matters because the bubble framing carries an implicit theory about who is exposed. In the dot-com analogy, the people who got hurt were retail investors holding Webvan stock and employees of the failed companies whose options went to zero. The venture capitalists recycled the capital and moved on. That is a real distributional story, but it is not the one running today.

Today's AI capital is concentrated in private markets and in the balance sheets of public mega-caps already so large that a 30% drawdown is absorbed without restructuring. Microsoft, Google, Amazon and Meta have committed over $300B in AI infrastructure capex for 2025 alone.⁴ If that capex turns out to have been overbuilt, the four companies survive it. Their workers are not exposed in the way Pets.com employees were exposed. Their shareholders are diversified institutions.

The people exposed to the structural bet are different. They are the workers whose jobs are being priced as automatable inside the $965B valuation. They are the smaller AI companies whose existence depends on competitive air the leaders are pricing themselves to remove. They are the users and the communities whose data trained the models, who have no negotiating position over how the resulting value is divided. They are the public sitting on the other side of an antitrust question that has not yet been framed as one.

What the bubble frame obscures. Greenspan's question, is this asset worth its price, assumes a world where prices can be wrong and a correction restores something like sanity. The question the AI valuations actually pose is harder. What if the prices are roughly right, conditional on a market structure that has not yet been democratically chosen? The $965B is a forecast of a world. It is also, by being placed, an investment in bringing that world about.

I do not think the useful response to this week's coverage is to argue about whether the bubble breaks. I think it is to notice that the bubble debate is the comfortable debate — it keeps the conversation inside the language of price discovery, where the worst case is a drawdown and the best case is that the optimists were right. The harder debate is about whether the market structure these prices assume is one anyone outside the cap table has been asked to ratify. That question does not get resolved by a correction. It gets resolved by whether anyone in a position to shape it decides to ask it now, while the prices are still being set.

Greenspan's framing was sharper than the use it is being put to. He asked how we would know. Thirty years on, on AI, we already know enough.

Footnotes

¶