Anthropic Priced Its IPO at $70 to $78. The Filing Says Something Bigger

Anthropic put a price on the biggest AI listing in history on November 10. The shares will be offered at $70 to $78 each, the roadshow is under way, and the company says trading could start as soon as November 19. That is the headline. The more interesting news is buried in the same filing, where Anthropic tells regulators it has crossed a threshold in its own safety policy: the point at which AI starts to speed up AI research itself.

The deal in numbers

The offering covers 825 million Class A shares. Anthropic itself is selling 801.4 million of them, and existing stockholders are selling 23.6 million. The underwriters, led by Goldman Sachs and Morgan Stanley, get a 30 day option to buy up to 123.8 million more shares.

At $78, the top of the range, the public offering raises about $64.4 billion. At $70, it raises about $57.8 billion. Anthropic has applied to list on the Nasdaq under the ticker ANT. The registration is not yet effective, so no shares can be sold until the SEC declares it so.

Alongside the public sale, seven technology companies are buying $26 billion of a separate class of non voting shares at the same IPO price. Google and Nvidia are each putting in $10 billion. AMD is buying about $2.5 billion, and Broadcom, Micron and Microsoft about $1 billion each. TSMC is buying about $500 million. Together, the IPO and the private placement are expected to bring Anthropic about $84.6 billion in net proceeds at the midpoint of the range, money earmarked for general corporate purposes including compute and infrastructure.

Ordinary investors get a slice too. Anthropic expects retail allocations to represent up to 20 percent of the shares offered, through brokers including Charles Schwab, Fidelity, Robinhood, SoFi and E*TRADE.

Smaller than the draft that leaked

The range is notable for what it is not. A day earlier, Anthropic sent bankers a draft filing that offered about 867 million shares at $70 to $80, worth up to $69.3 billion, and then withdrew it in a follow up email saying the pricing details were inaccurate. The real filing offers fewer shares and tops out $2 lower.

The correction matters because it was the first stumble in an otherwise tightly managed process. The company spent a day with no price at all in its public documents, which is unusual for a deal of this size. Now the blank is filled, and the deal is a little smaller than the version that leaked.

The valuation question

The range values Anthropic at $1.26 trillion to $1.41 trillion, according to The New York Times. CNBC puts it at up to $1.405 trillion on a fully diluted basis at $78 a share. That is below the $2 trillion valuation bankers discussed earlier in the summer, and well above the $965 billion Anthropic was valued at in a May funding round.

The gap between those two numbers is the story of the last week. A sharp selloff in AI stocks on November 5 knocked confidence, and investors pushed back on the top end. The range that survived is a compromise: big enough to make this the largest IPO ever by money raised, restrained enough to leave room for the shares to rise after listing.

The business underneath is growing fast. Third quarter revenue was $17 billion, up from $11.6 billion the quarter before. Run rate revenue, an annualised measure of recent sales, reached about $81 billion at the end of October, up $10 billion in a single month. That growth is the reason investors are willing to pay more than a trillion dollars for a company that still loses money on its operations.

The line buried in the filing

The filing also contains a disclosure with nothing to do with money. Anthropic says it has met the AI research and development capability threshold in its Responsible Scaling Policy, the safety framework it wrote for itself.

The wording matters. In the November 2 version of the filing, Anthropic said its estimates for future models indicated it “may imminently meet” the threshold, and that no current model had crossed it. On November 10 the passage reads differently: “we made a determination that we have met this threshold based on what we expect from future models and our overall rate of progress, rather than on what we have observed from current models.”

The threshold is defined as the point where AI would double the rate of progress in AI research itself, as much progress in one year as would normally take two. The policy says it is “intended to capture the onset of dramatic recursive self-improvement.” Crossing it is supposed to trigger stronger measures: security research against the best resourced attackers, logging of all critical AI development work, systematic alignment checks, and risk reports reviewed by an outside expert.

Anthropic has not said which of those measures are under way, or when the further reports it promises will appear. It says only that it continues to assess the data and is evaluating the implications.

Outside evidence points the same way. METR, a nonprofit that tests frontier models, estimated on November 5 that Anthropic’s capability scores have accelerated 1.9 times since October 2025, and guessed that about 80 percent of the current acceleration is due to AI itself. Anthropic’s own system card for its newest models puts the figure at about 1.88 times, just below the 2 times line, and notes that the newest model does not cross the threshold on its own. The filing is where Anthropic first said what it concluded from all of it.

There is a practical side to this, too. The filing says Claude wrote over 90 percent of the contributions to Anthropic’s codebase in September 2026. And it discloses that the company recently chose to delay pre training a new frontier base model for more than a month, to strengthen security measures against model weights being stolen. The next flagship model is late because the company is worried about protecting it.

For people who use Claude, nothing changes today. The filing says the determination does not affect current plans or prices. But the company warns that as capabilities advance, measures such as delaying releases or limiting access “may become more frequent and more costly.” Its most capable model class, Mythos, is already limited to vetted organisations, and an earlier version was switched off entirely for all customers in June after US export restrictions, until the restrictions were lifted.

What the money is for

The scale of the spending explains the scale of the raise. Anthropic’s compute and infrastructure commitments now total about $900.9 billion, with $451.6 billion of that due before the end of 2030. That is up from $895.4 billion a week earlier. The company is signing contracts for computing power faster than it can build the data centres to house it.

This is the bet investors are being asked to fund: that revenue growing at this pace justifies commitments that are more than ten times annual sales. If the models keep improving and customers keep paying, the arithmetic works. If growth slows, the commitments do not shrink with it.

What to watch

Pricing is expected in the week of November 16, with the first day of trading planned for Thursday or Friday of that week, November 19 or 20. Three things will tell us how the deal is going before then.

First, the roadshow itself. Anthropic needs enough large investors to fill a $64 billion order book at a price near the top of the range. A deal this large leaves no room for a weak reception.

Second, the retail slice. Up to a fifth of the shares going to ordinary investors is unusual for an offering of this size, and demand there will show whether the public wants in on an AI lab at a trillion dollar valuation.

Third, the threshold disclosure. Anthropic has now told the SEC, in a document signed under penalty of perjury, that AI is accelerating AI research at its own company. Regulators, rivals and safety researchers will all read that sentence. The question is whether anyone asks what follows from it before the shares start trading.