Anthropic has filed its S-1 for a Nasdaq listing. The official document shows a net loss of $80.7 billion for the first half of the year, but $72.4 billion of that is accounting-related. In the second quarter, the company posted an operating profit of over $1 billion. At the same time, Anthropic is committed to $542 billion to $567 billion in compute and infrastructure obligations.
On October 26, Anthropic filed its S-1 registration statement for a Nasdaq listing under the ticker symbol ANT. This appears to put it ahead of OpenAI, the creator of Claude, in the race for the first major IPO by an AI model developer. That scenario was already taking shape when Anthropic previously filed confidential registration documents with the SEC.
The IPO is now expected to be valued at around $1.5 trillion. A week ago, estimates were still at $2 trillion, but that figure has been adjusted after sharing financials with investors. Nevertheless, it’s a significant jump. In its latest funding round in May, Anthropic still aimed for $30 billion at a $965 billion valuation.
At first glance, the figures are alarming. For the first half of 2026, Anthropic reported a net loss of $80.7 billion. Of that amount, $72.4 billion is purely an accounting entry. Amazon and Google have financed the company in recent years through convertible loans and warrants. Because Anthropic’s valuation has risen sharply in the run-up to its IPO, the value of those outstanding bonds on the balance sheet has risen accordingly. The company must write off that difference as a loss. It costs the company nothing in terms of cash.
From loss to quarterly profit
Operationally, the picture is different. In the second quarter, Anthropic generated $11.6 billion in revenue and an operating profit of over $1 billion. For the full half-year, revenue totaled $16.3 billion, compared to $1.2 billion a year earlier. Over those six months, it still posted an operating loss of $693 million. The first quarter thus still weighed heavily on the results. By May , it was already clear that Anthropic was nearly profitable, with expected revenue of $10.9 billion. The focus is shifting from expensive model training to large-scale inference: customers making API calls.
The Net Dollar Retention (NDR) is particularly striking. In the traditional SaaS world, 120 percent is already considered excellent. This is because it indicates that existing customers, excluding new customers, spend 20 percent more each year. Anthropic’s S-1 filing shows an NDR of nearly 900 percent. Business customers have thus roughly increased their spending on the Claude APIs ninefold in just one year. This also raises questions. System prompts, RAG pipelines, and fine-tuning are often tailored to the specific characteristics of a single model, making it costly to switch to another provider.
Hundreds of billions of dollars in committed computing power
This growth comes with significant obligations. Anthropic has purchased hundreds of billions in committed computing power from AWS, Google Cloud, and, to a lesser extent, Azure, with contracts running through 2036. The company must use that capacity regardless of actual demand for its models.
The largest commitments are with Google, totaling at least $114.5 billion through July 2033, and Amazon, totaling at least $110 billion through April 2036. Microsoft also has an outstanding commitment of $44.9 billion through May 2033. The document therefore lists a contracted base of $265.6 billion in fixed minimum obligations. On top of that, billions more are hidden away. The document also includes a provision for additional Google TPUs between 2028 and 2031 that will cost another $ 125 billion to $ 150 billion. Plus two other deals exceeding 100 billion: one with Broadcom for TPU capacity worth 161.2 billion and a flexible, cancellable contract with SpaceX (up to 120.5 billion).
For business customers, this structure also has a procurement aspect. Claude capacity can be purchased through existing AWS or Google commitments, or through a direct contract with Anthropic. The first route increases dependence on the hyperscaler. With the second, the customer shares in the pricing structure that Anthropic uses to cover its own cloud costs.
Trust with special control
On governance, Anthropic differs from other publicly traded tech companies. The company was founded out of dissatisfaction with OpenAI’s security risks. The S-1 confirms that the Long-Term Benefit Trust (LTBT) holds significant power. Through a separate class of stock (Class T), this trust appoints a majority of the board of directors. The board has special voting rights and can intervene if it deems an AI model to be socially disruptive or unsafe.
At the same time, the founders hold a separate, unique Class F share that automatically grants them 50.1 percent of all general voting rights. This lets the trust monitor safety limits behind the scenes while the founders retain control over the company’s commercial direction. In theory, the trust can decide to downgrade a model, take it offline, or disable specific features. Standard SLAs do not cover such interventions.
The company has not yet disclosed how many shares Anthropic will issue or at what price, but this latest move makes an IPO truly imminent. According to Bloomberg, it could all happen in November.