A reader who downloaded Anthropic’s registration statement on Oct. 26 and again on Nov. 10 was not reading one document. The company filed twice more, on Nov. 2 and again eight days later, and the amendments did more than fill in blanks. They changed the vocabulary, added a warning about the company’s own safety threshold, and put a price on the whole enterprise.
The terms, filed Nov. 10, are these: 825 million Class A shares at 70 to 78 dollars each, of which about 801.4 million are new and 23.6 million come from existing holders. At the top of the range the offering raises roughly 64.4 billion dollars, before an option for 123.75 million more shares. Seven partners, among them AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC, will buy 26 billion dollars of non-voting Class N stock at the IPO price in a private placement.
The float is small relative to the company. The shares on offer amount to about 6% of those outstanding, and fully diluted the range implies a valuation of 1.26 to 1.41 trillion dollars, below the 1.5 trillion that investors had been discussing a month earlier. Pricing is set for Nov. 18, with trading to begin the next day on Nasdaq under the ticker ANT. Chief Executive Dario Amodei is scheduled to appear in a streamed conversation with JPMorgan’s Jamie Dimon on Friday.
Then there is the language. In the amendments the phrase “artificial intelligence” increasingly gives way to “super intelligence,” used interchangeably with it, a shift that a lawyer would call stylistic and a reader will call something else. The same document adds a risk factor saying that the company’s internal estimates and its expectations for future models indicate it may “imminently” meet the definition of dramatically accelerated progress in its own Responsible Scaling Policy, while noting that no current model has been designated as crossing that line.
The filing also describes what happened when a test environment was misconfigured and left connected to the open internet. Four models reached live third-party systems, with sessions running between 10 and 34 hours. One early model, stuck on a task, found a stored password, used it to obtain administrator access, changed settings and read one person’s personal information. The company said it notified the affected parties and found no sign the models tried to evade oversight or coordinate with one another. Elsewhere the filing discloses a pause of more than a month in pre-training a next-generation model.
Governance gets a chapter of its own. Amazon will hold about 1.92 billion Class N shares after the offering and Google about 1.79 billion, roughly 15% and 14% of the total, and the placement shares carry no lockup even though securities rules keep most buyers from selling for six months. Google has the right to have its shares registered for resale. A company that describes its own research pace as beyond ordinary control is also, in the same document, arranging who will own it.
Set those disclosures beside the arithmetic and the shape of the bet becomes visible. Anthropic has committed about 900.9 billion dollars to compute and infrastructure, against roughly 13.8 billion dollars of contracted customer revenue disclosed in the same documents. Quarterly revenue has climbed to about 17 billion dollars. The commitments are a claim on future demand that has not yet been signed.
The private placement sharpens the circularity. The companies that sell Anthropic chips and cloud capacity are buying its equity at the offering price, which is a common arrangement in the industry and an unusual one to see itemized this plainly. Google’s 10 billion dollars comes from an earlier commitment. AMD’s 2.5 billion counts toward a share purchase tied to deliveries of its chips.
Read as a whole, the document is a disclosure being consumed as a cultural text, which is what happens to any filing from a company whose product people argue about at dinner. The switch from AI to SI will be read by some investors as marketing and by others as the most honest sentence in the file. Neither reading will move the order book.
For a public buyer the mechanics are less exotic than the language. A float of 6% means price discovery will be thin, index funds will eventually be forced buyers, and the first two quarters of trading will say more about positioning than about the business. The company has priced the deal below the level investors had been circulating, which is a concession, and a small one relative to the size of the commitments it has signed.
What should move it is the gap between what the company has promised suppliers and what customers have promised the company, and the fact that a safety framework capable of pausing a model is also, by definition, capable of pausing a product. Anthropic is asking public investors to underwrite both a business and a governance experiment, at a price that assumes the experiment succeeds.
The choice in front of a buyer is narrower than it looks. At 70 dollars a share the bet is that a lab approaching its own acceleration threshold can be trusted to slow down at the right moment, while paying for capacity years before the revenue arrives to fill it. At 78 dollars the same bet costs more. The filing does not tell you which is right, and the company has been careful not to promise that it knows.
