Anthropic () wants to go public in October at a valuation of $2tn or more. Some of its own backers are modelling $3tn. Either number would make it the largest stock market debut in history, and I think it is the riskiest trade on offer anywhere in the market this year.
I am not saying Anthropic is not a genuinely exceptional business, because it plainly is. What I am saying is that exceptional and correctly priced are two different questions, and right now the gap between them is wide enough to worry about.
Here is the part investors keep skipping past. The valuation being pitched assumes annualised revenue will nearly triple by the end of 2026, and it is a bold assumption to hang the largest IPO in history on, especially with a listing date that arrives at exactly the wrong moment to test it.
The number itself is not audited revenue. It is an annualised run-rate, a projection stretched forward from a few strong months rather than a certified year of sales. We already know how fast that projection can move.
In June, a temporary Commerce Department restriction on Anthropic’s top models slowed growth almost overnight. A single policy decision did that. I would want to know what the next one could do before I paid a $2tn price for the answer.
A valuation this size does not price in one good quarter. It prices in years of near-uninterrupted acceleration, with no room for a repeat of June and no room for the regulatory and geopolitical friction that is still clearly in play around this company.
Look at what is happening around Anthropic this very week. CoreWeave (), one of the AI infrastructure names most tightly bound to Nvidia, has admitted it would struggle badly if forced to shift away from Nvidia’s chips.
Far from a footnote, it is a signal of how tightly wound and circular the financing behind this entire boom has become, right as Nvidia () walks into earnings shadowed by China licensing uncertainty and mounting questions over who is really funding whom across the AI supply chain.
Then look at the public comparisons investors are actually using to justify Anthropic’s number. Palantir () and Nebius () are trading at somewhere between forty and fifty-five times revenue, and both are already drawing serious short-seller fire even as their share prices climb. If the market is this uneasy about paying up for AI names that already report audited, public numbers, it should give serious pause to anyone pricing a private AI lab off a forward-looking metric instead.
I think about SpaceX () as the clearest lesson in what happens when private valuations move faster than reality can confirm them. It went from roughly three hundred and fifty billion dollars to eight hundred billion in about a year, and there is already talk of it approaching two trillion before it has even listed. Momentum in a private market is not the same thing as durability in a public one, and Anthropic is walking the same path, only faster, straight into an IPO window where investors are already nervous.
Pricing the largest IPO in history off unaudited, extrapolated numbers, in a market already flinching at AI multiples, is the central risk in this whole story, not a footnote to it.
My advice to clients weighing exposure to this listing, directly or through the AI trade more broadly, is to separate genuine long-term conviction from October momentum. Those are not the same trade, and treating them as one is how sharp corrections get made.
This is exactly the kind of moment where independent, professional guidance earns its keep, because the difference between participating intelligently and getting caught out comes down to timing and discipline, not enthusiasm.
Anthropic may well prove every one of its bulls right eventually. But betting $2tn on it happening on schedule, in this market, this October, is not conviction. It is speculation dressed up as certainty.