Anthropic has published the prospectus for its IPO and is heading for the stock market. In it, the AI company itself writes that its growth will slow and that its losses will continue.
Anthropic filed its S-1 with the US regulator SEC on 26 October. In this prospectus, a company introduces itself to investors. It includes figures up to the end of June 2026, as well as information on customers, risks and the governance structure. Anthropic wants to be listed on Nasdaq under the symbol ANT.
A look behind the scenes
In June, Anthropic had already confidentially submitted documents to the SEC, without naming a date. In August, there was talk of an IPO in October. In early October, Bloomberg reported that the company is aiming for a listing just after the US elections of 3 November, possibly on 9 November. Now the file is public for investors, and the sale of shares can start as soon as the registration with the SEC takes effect.
Now that Anthropic is taking the final steps towards becoming a publicly listed company, the prospectus brings a lot of information to light. For instance, we learn that control of the company stays with an independent trust and the seven founders after the IPO. The seven founders get 50.1 percent of the voting rights through a separate share class. An independent trust, the Long-Term Benefit Trust, appoints the majority of the board of directors.
Growth in figures
The prospectus contains more interesting figures. In 2025, Anthropic generated 4.6 billion dollars in revenue. In the first half of 2026, that was already 16.3 billion dollars, compared to 1.2 billion dollars a year earlier. More than 11,900 customers spend at least 100,000 dollars a year with the company, compared to just over 950 a year earlier. That confirms the strong growth in customers, usage and revenue that Anthropic keeps emphasising.
In the second quarter, Anthropic also booked an operating profit of 974 million dollars, after five clearly reported quarters of losses. Anthropic itself notes that on an annual basis it has only ever made a loss.
The net loss for the first half-year was 80.7 billion dollars. Of that, 72.4 billion dollars is an accounting item: Anthropic raised money through loans that will later be converted into shares, and because the company became more valuable, those loans rose in value. At the end of June, Anthropic had 112.4 billion dollars in cash.
Growth slows
Anthropic expects its revenue growth to “decline over time, potentially significantly and faster than we anticipate”, as revenue grows larger and the market matures. According to the company, recent growth may stem from an early adoption wave that is not sustainable. When growth levels off, seasonal effects weigh more heavily. Around the turn of the year, customer spending was about twenty percent lower for two weeks.
An estimated five to seven percent of revenue comes from customers who violate the terms of use. Since August, Anthropic has been taking tougher action against this and expects that share to fall. The company also cites the risk that customers switch to cheaper models, for example because of competitors’ offerings.
Market becomes more price-sensitive
That customers are looking at cheaper alternatives is no secret by now. Chinese models are often open and therefore free to use. On OpenRouter, where developers freely choose between models, DeepSeek was the most used provider this summer and Alibaba’s Qwen ranked ahead of Anthropic. According to the Ramp AI Index, about eleven percent of users deploy its own top model Fable 5, while the cheaper Opus and Sonnet models are more popular.
Study after study shows that companies with mature AI implementations opt for value for money. The so-called frontier labs, of which Anthropic is one, have crowned themselves champions of pure innovation, but not of AI capabilities at a sharp price. The market becoming more price-sensitive will not benefit growth either.
Anthropic remains loss-making after IPO
Anthropic also expects costs to rise over time and losses to continue. The net loss was 1.5 billion dollars in 2023, 8.3 billion dollars in 2024 and 42 billion dollars in 2025. The operating loss for 2025 came to 8.1 billion dollars, as already became clear in early October.
Anthropic also expects costs to rise over time and losses to continue.
Compute remains the biggest cost item. Anthropic buys it twelve to eighteen months in advance and expects its margins may shrink whenever a lot of new capacity comes online. It has contracts with Google, among others, for at least 114.5 billion dollars, Amazon for at least 110 billion dollars and SpaceXAI, formerly xAI, for up to 120.5 billion dollars. The prospectus leaves the total amount open. ITdaily previously reported that Anthropic wants to put more than 500 billion dollars into AI infrastructure.
Based on the position at the end of June, the IPO also brings a one-off cost of 2.9 billion dollars for employee shares, and another 23.7 billion dollars spread over just over two years.
So where will the profit come from?
According to Anthropic, compute costs always come before the revenue they generate. In the second quarter, revenue rose by 1,368 percent compared to a year earlier, while compute costs rose by 300 percent. At the start of 2025, compute still cost 2.5 times revenue; in the second quarter of 2026 it was less than half.
Anthropic is counting on its models becoming cheaper per task, which drives up usage. Claude Opus 5.5 performs a task about 40 percent more cheaply than Opus 5. Fortune 500 customers spend on average less than 0.1 percent of their total costs on Anthropic, which the company sees as room for growth. In the long term, it wants to house more compute in its own data centres, which should be cheaper. Operations already generated 3.9 billion dollars in cash in the first half-year.
An IPO full of promises
The prospectus substantiates what was already known and suspected. Anthropic is growing like wildfire, but there is no guarantee that this rapid growth will continue. Moreover, even enormous growth is no quick path to profit: Anthropic has to keep investing in expensive compute to support those extra customers with extra capabilities.
The predicted high market value is based on promises without guarantees.
In other words, the predicted high market value is based on promises. There has to be light at the end of the tunnel. As AI models become cheaper to run, Anthropic will be able to sell more AI capacity from its compute. The company assumes that businesses will keep embracing Anthropic, with a growing share of their spending going to AI.
No guarantees
That prediction is no guarantee: the market is becoming more price-sensitive, and the frontier AI models that set Anthropic apart are matched relatively quickly by open models, at least on important capabilities that are relevant to a broad audience.
Anthropic does not name a date for profitability. It does write that it must show a “credible path to profitability” to keep attracting capital. Whether investors buy into that will become clear once the share is actually traded on the stock market. Given the current hype around AI, the odds look good that they will.