Anthropic Trimmed Its Own IPO Before It Even Started

On Tuesday, Anthropic put a price on its future: $70 to $78 a share, 825 million shares, a Nasdaq listing under the ticker ANT, and a roadshow that starts now. But the most interesting number in the filing is one that is no longer there. A day earlier, a version of the filing sent to bankers showed about 867 million shares priced at $70 to $80. Anthropic narrowed its own offering before the roadshow even began.

Anthropic also lined up $26 billion from the biggest names in technology. Whether the trim means anything at all is an open question: issuers routinely price below early talk to leave room for a first-day rise, so the narrower range may reflect ordinary IPO pricing mechanics rather than a verdict on the company.

The range that moved first

The filed terms, set out in Amendment No. 2 to Anthropic’s S-1, are precise. The company and some of its existing shareholders will sell 825 million Class A shares — 801.35 million new shares plus 23.65 million from existing holders — at an estimated $70 to $78 each. The sale is expected to raise between $57.75 billion and $64.35 billion, and Anthropic estimates net proceeds of $84.6 billion to $93.7 billion once the private placement and the underwriters’ option are counted.

The valuation arithmetic depends on how you count. The New York Times calculates that the range could value Anthropic at as high as $1.4 trillion, and CNBC puts the fully diluted figure at $1.404 trillion. On the 13.09 billion shares the filing expects to be outstanding after the offering, the same range works out to roughly $916 billion to $1.02 trillion. For comparison, Anthropic was valued at $965 billion in its May funding round, and earlier discussion — reported by Reuters — had put the figure as high as $2 trillion. Whether those numbers are truly comparable is an open question: the basis of the earlier figure is not established, and the headline number counts options and equity awards that the share-count figure does not.

What is not in doubt is the direction of the revision. The range bankers saw a day earlier was wider at the top and larger in size. The filed version is smaller on both counts. The company has already trimmed its range once.

The $26 billion safety net

Alongside the offering, Anthropic struck a deal to privately sell $26 billion of non-voting Class N stock at the IPO price to AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC. Google and Nvidia are investing $10 billion apiece.

Read that list again. The companies buying in are incumbents who are also Anthropic’s customers, suppliers and rivals — and whether the placement signals confidence in the company or dependence on them is an open question. This is not a random syndicate of believers; it is the companies at the center of the AI industry itself, buying into the offering at the same price the public will pay.

There are two ways to read that. One is confidence: the people who know this business best are putting $26 billion behind it. The other is dependence: does an offering with $26 billion spoken for before the roadshow need that backing, or does it simply not trust the open market to do the job alone? The filing itself does not say, and the roadshow may not settle it either — pricing alone may leave the reasons unresolved.

The warning Anthropic gave itself

There is a reason the caution is legible in the numbers. Anthropic told investors its growth was cooling, before a tech sell-off that followed the midterm elections cooled investors’ appetite further. The filing’s own figures tell both halves of that story at once.

Run-rate revenue reached approximately $81 billion by the end of October, about $10 billion more than at the end of September. Third-quarter revenue came in at $17 billion, and nine-month revenue was $33.2 billion.

But the same filing reports a nine-month net loss of $104 billion, compute and infrastructure commitments of $900.9 billion, and roughly $132 billion in cash and short-term investments. Whether those figures draw pushback from the investors being asked to buy in is an open question.

The filing carries one more telling detail. It states that Anthropic has met the AI research-and-development threshold in its own safety policy — judged, the filing says, on what it expects from future models rather than current ones. The company is, in effect, grading its own safety homework on the eve of asking the public to buy in. And in a small but revealing change, the filing uses “super intelligence” and “SI” interchangeably with “artificial intelligence” and “AI”, citing a September executive order directing government departments to use the new terminology.

The fine print: control stays home

Whoever buys in should know what they are — and are not — buying. As a public benefit corporation, Anthropic’s founders hold a Class F share with 50.1% of the voting power, and the Long-Term Benefit Trust will elect a board majority. Class A buyers will have limited influence over the company they are funding.

The founders, meanwhile, are selling. Co-founders including chief executive Dario Amodei and president Daniela Amodei are each selling nearly 3.4 million shares — about $250 million each at the $74 midpoint of the range.

None of this is hidden. It is all in the filing, in plain sight. The underwriters have an option to buy up to 123.75 million more shares at the IPO price, and Anthropic expects up to 20% of the offering to be allocated to retail investors. The company plans to price the offering on November 18 and start trading on the Nasdaq around November 19; the Wall Street Journal reports it is aiming for shares to start trading Thursday or Friday next week.

What the roadshow has to answer

The context around the debut is noisy, and the filing does not pretend otherwise. The Verge notes months of debate around AI safety, midterm election results showing souring public opinion toward AI companies and the data-center projects they are investing in, and the weekend’s revelations about an alleged Chinese spy’s involvement with Nvidia chief executive Jensen Huang. Whether any of that moved the price is something the reporting does not establish — and the roadshow can test demand and produce further reporting, while the reason for the revision may remain unresolved.

What the filing does establish is the shape of the debut: a trimmed range, $26 billion already spoken for, founders keeping control, and a company that warned investors its growth was cooling before it asked them to buy. Whether that shape is caution, careful management, or simply ordinary pricing is an open question — and one the roadshow may leave unresolved.

The roadshow will test whether that bet was right — though a final price, high or low, will not by itself establish why the range moved in the first place. If the final price lands at the top of the range, the caution may look like discipline; if it lands at the bottom, or if the range moves again, the trim may look like the first sign of something else. Either way, the reason for the revision remains unresolved: the filing records the move, not the motive behind it.