Anthropic, the company behind artificial intelligence Claude, is ready to go public. This Monday it announced that it plans to list on the Nasdaq with the symbol ANT starting the week of November 16 and if the operation goes ahead, it will be the first company dedicated exclusively to developing advanced artificial intelligence models to hit the stock market.
The prospectus leaves the number of shares it will issue and the price blank (something increasingly common in IPOs), but the figures used by investors have no possible historical comparison. The operation could raise 100 billion dollars (about 89 billion euros) with a valuation of the company close to 2 trillion dollars. It would surpass SpaceX, which in June raised $75 billion in what was until now the largest IPO on record.
Arriving in mid-November would also mean winning the race against OpenAI, which according to rumors has postponed its debut until 2027.
FROM ZERO TO 16,800 MILLION
Five years ago, Anthropic was a group of researchers that spun off from OpenAI with the idea of creating safer artificial intelligences. It began selling access to Claude in early 2023 and made about $41 million that year. In 2025 there were 4,585 million and only between April and June of this year, 11,600 million. The company has told investors that it reached $16.8 billion in the third quarter, according to the Financial Times, although that amount is not yet listed in the prospectus.
More than 11,900 customers spend at least $100,000 a year on Claude, which has become the tool of choice for software developers for its ability to generate code. Some 250 companies spend more than 10 million euros and those who were already clients a year ago spend, according to the prospectus, nine times more.
The company's efficiency has also improved. With about 5,300 employees, it earns 6.6 million dollars per worker, compared to 1.4 million a year ago.
But although the business numbers are good, Anthropic is still a company unable to generate a profit, and for investors, honestly, the numbers will be difficult to digest. In the first half of the year it closed with net losses of 80.7 billion dollars. 72.4 billion correspond to the revaluation of convertible bonds and other instruments that are in the hands of their investors, until now private. When Anthropic's valuation rises, those securities are worth more and the difference is recorded as a loss, even if no money comes out of the box.
THE INVOICE
But even ignoring the effect of the bonuses, the business in which Anthropic operates is complex. The so-called "frontier" models (the most advanced on the market, such as Claude Fable) require enormous resources and it is necessary to continually update them because they are also the ones that bring in the money. More than 80% of third-quarter revenue came from Claude versions released in April.
To power and run them, Anthropic has signed computing contracts worth more than $600 billion, to which must be added another $200 billion in additional planned processor purchases. It is almost 12 times what it earns in a year at the current rate.
Broadcom, for example, rents chips designed by Google for about 161 billion. With Google Cloud it has committed to spending at least 114.5 billion; with Amazon, 110,000 million, and with Microsoft, 44,900 million, and in several cases "regardless of actual use."
And then there's SpaceX. Elon Musk's company, which absorbed xAI and with it Grok, Claude's direct rival, but which will sell computing capacity to Anthropic for 120.5 billion until 2029.
Proof of how complex the interdependence of these companies has become is that the relationship is circular. Amazon, Google and Microsoft are at the same time shareholders, suppliers, competitors, customers and sales channel. Almost half of Anthropic's revenue comes through its cloud platforms, and some providers also act as a bank. Broadcom, for example, is willing to lend you up to $42 billion in convertible bonds so you can pay the rent for those same chips. Nvidia frequently makes similar maneuvers to sell its processors.
And therein is summarized, in a certain way, the reason for the IPO. Anthropic has raised $141.1 billion in capital since its founding, but its expenses have long exceeded the amounts that venture capital companies or even sovereign wealth funds can afford. In order to continue feeding its artificial intelligence, it needs a larger investor base.
INVEST IN THE APOCALYPSE
Those investors will have little to say, and the prospectus warns as much. Its structure can "effectively eliminate" shareholders' ability to influence important decisions. Anthropic is a public benefit corporation, compelled to balance the interest of its shareholders with a mission, which is "the responsible development of advanced AI for the long-term benefit of humanity." An independent organization, the Long-Term Benefit Trust, elects the majority of the board with a special share without economic rights.
Just before the premiere, the company will also issue a single class F share, with 50.1% of the votes, for a company made up of its seven co-founders, including the brothers Dario and Daniela Amodei. Added to this is a public benefit fund with 5% of the capital, which may receive up to 2% more each time the stock rises 50%.
The risk factors take up almost a hundred pages, and some are unusual in a stock brochure. Anthropic recognizes that AI models, including its own, have shown self-preservation behaviors, such as resisting being disconnected or hiding information, and even behaviors reminiscent of blackmail. He also admits that the damage from an "AI-related catastrophic incident" could be so great that neither an insurer, nor perhaps society as a whole, would have the capacity to absorb it.
