Anthropic Wants $2 Trillion. That's a Liquidity Test, Not an AI Test.

The largest company in history is about to go public, and it doesn't want the market to answer the question it's actually asking. Anthropic, the maker of the Claude models, filed confidentially for an IPO with the SEC on June 1, and investors expect it to list in late September or early October at a valuation of up to $2 trillion — a debut that would blow past SpaceX's record $1.77 trillion offering this summer. The formal prospectus lands after Labor Day, and the roadshow that follows is a two-week test of one thing only: whether the marginal buyer, the next dollar of risk capital, will underwrite that number.

Here is the arithmetic, because beginners deserve it plain. Anthropic said its annualized revenue run rate — a projection of a full year from a shorter stretch — topped $65 billion at the end of July, roughly seven times its pace at the end of 2025. Divide the $2 trillion target by that $65 billion and you get about 30 times current revenue. The private round that gave Anthropic a $965 billion valuation in May implied closer to 15 times. The IPO would nearly double that private mark in one move, before any new sales have been earned.

This is the real reason the calendar drifted. Revealing the prospectus after Labor Day, rather than in the summer, buys the company nothing except time to watch the market's mood before it commits to a number. And the mood has already been set by the last giant debut. SpaceX priced at $135 a share in June, popped about 19% on day one, touched an intraday high near $225 on a tiny tradable float, and then sagged below its opening price within three weeks. When it reported numbers in August showing the sheer size of the capital its AI-compute build demands, the stock fell to roughly $108, about $1.4 trillion. The froth had been real; so was the bill for the capacity underneath it.

Anthropic is the same business model at a different scale. Its revenue is genuinely compounding — it posted positive adjusted operating income last quarter on revenue above $11.5 billion — but it is a deliberate loss-leader landlord that rents compute it does not entirely own. It committed $80 billion to compute capacity in a single week, sources told Forbes. The S-1 is expected to name "AI backlash" as a risk factor; a Gallup survey found seven in ten Americans oppose data center construction in their area. None of this makes the growth fake. It makes the 30-times multiple the thing being tested.

So watch the plumbing, not the price. Two details in the coming filing do more work than any headline. First, Reuters reports Anthropic is weighing the unusual step of letting existing shareholders sell stock directly into the IPO — secondary shares — something SpaceX and Cerebras each barred. That changes who sits on the other side of the first-day buyer's trade: your allocation may be providing the exit for an early holder rather than funding growth. Second, the company is considering lockup periods longer than the customary 180 days and requiring employee sales through preset trading plans. Longer locks keep the public float small, which supports the price — but the unlock calendar is the fuse, and the SpaceX show taught us exactly how the scarcity premium deflates once supply actually reaches the market.

The result is a read on the whole AI complex, not just one company. The largest IPO in history is a claim on a finite pool of free risk capital, and the way it prices tells you what the big mutual funds and sovereign funds — the forced buyers who will index it into their mandates regardless — think the frontier is worth. The revenue is real and the growth is staggering; the question underwriting this whole roadshow is whether 30 times is a mark the market honors or a level it lets bleed out into the lockup calendar. On a trade that big, price is the last thing to trust. Read whose shares are selling, and who is left holding them when the supply opens up.