Anthropic Sets $70-to-$78 IPO Price Range for Nasdaq Listing

Anthropic wants Nasdaq to certify a price its own suppliers have already agreed to pay.

The filing is not subtle. The company plans to sell 825 million shares under the ticker ANT, at $70 to $78. At the top of the range that is as much as $64.35 billion in proceeds and a fully diluted valuation of $1.405 trillion, with the offering targeted for mid-November. For the nine months through September 30 it reported $33.24 billion in revenue and a net loss of $104 billion, much of it — the company says — from fair-value adjustments on convertible notes and derivative liabilities.

Take the revenue seriously. Thirty-three billion dollars in nine months is not a demo, and the habit of treating frontier AI as a fad is how people miss what is actually being sold. The loss is what will be explained away for you, by people who need the explanation to hold. "Fair-value adjustments" is this boom's "adjusted EBITDA": an invitation to step past the figure that would otherwise end the conversation. The invitation may be technically fair. Non-cash marks on convertibles and derivatives can swamp operating results, and marks reverse. That is the problem, not the comfort. The red ink is an argument about instruments insiders already used to finance the company, and the public is being asked to stand on the other side of a capital structure it did not write.

The filing then names the buyers who are not waiting on the opening trade. AMD, Broadcom, Google, Micron, Microsoft, Nvidia and Taiwan Semiconductor have agreements for a private placement of $26 billion in unregistered Class N common stock at the IPO price. Call that arm's-length price discovery only if you can do it with a straight face. These firms design the chips, fabricate them, supply the memory and rent the cloud. They sell Anthropic the means of production. A $26 billion seat can be rational for Nvidia even if Nvidia never concludes the equity is worth $1.4 trillion, because more Anthropic is more silicon. Microsoft and Google are solving a strategic problem a brokerage account cannot see. Vendor validation is not a market. It is a supply chain buying a piece of its customer and pointing at the purchase as if strangers had set the price.

America should want frontier labs. The race in advanced models, and in the chips that train them, is a national fact, and capital raised here beats capacity that depends on someone else's industrial policy. That does not tell you who controls the lab, at what dilution, or why a range blessed by the suppliers deserves to be treated as a clearing price. A Republican Party that spent years calling Big Tech an unaccountable cartel does not get to salute this syndicate because the product is a model and the announcement says leadership. The chorus that will brand any hesitation Luddism is already deep in the private AI boom. Their urgency is not your duty.

Anthropic's politics run in the same direction as the prospectus. For years the company has argued that frontier AI should not be left to ordinary competition alone — that scaling needs governors, that government ought to have a role, that safety rules are the responsible path. Some of that is conviction. Much of it is also a moat. Rules only a handful of labs can satisfy are a preference for incumbents, written in the language of catastrophe. The lab that wants Washington in the room for the next model now wants the public to capitalize a valuation that assumes the room stays small.

Markets are a discipline. They make a seller find buyers who are not already its landlords. This offering arrives pre-disciplined: a circled range, $26 billion in unregistered stock placed with the picks-and-shovels firms, a trillion-dollar headline, and a $104 billion loss filed as a fair-value weather event. You are not being offered a discovered price. You are being offered a negotiated one, and told to call the negotiation capitalism.