Anthropic’s IPO Prospectus: $518 Billion for Compute, 80 Pages of Risk

Anthropic is going public, and parts of the prospectus for its Nasdaq listing have come to light. Reuters and the Financial Times have seen the document, which has not yet been released. It shows a company whose revenue grew twelvefold in a year, that has also earmarked hundreds of billions of dollars for computing power, and that explains to its future shareholders in unusual detail how dangerous its own product could become. For anyone who uses Claude or wants to understand the AI market, the prospectus is the most detailed look inside one of the major model makers so far.

Key takeaways

• Anthropic brought in nearly $4.6 billion in revenue in 2025, twelve times the previous year, and posted an operating loss of more than $8 billion.
• In the second quarter of 2026 alone, revenue reached $11.5 billion, according to the Financial Times; on an adjusted basis, Anthropic is headed for its second straight quarter of operating profit.
• The company plans to spend $518 billion on cloud, computing capacity, and infrastructure in the coming years.
• Roughly 80 of the 261 pages describe risks, including models that could resist shutdown or conceal information.
• Anthropic is targeting a valuation of more than $2 trillion, with an October listing as the goal.

Growth few software companies have ever seen

The figures, first reported by Reuters, describe a company in a state of exception. Nearly $4.6 billion in revenue in 2025 was set against operating expenses of almost $13 billion. More than half of that, $7.33 billion, went to compute and infrastructure, three times as much as in 2024. The operating loss came to more than $8 billion.

The net loss of roughly $42 billion looks even more dramatic, but it is mostly an accounting figure. Nearly $34 billion of it stems from remeasuring financing instruments that can later convert into shares. Their estimated value has risen, and that difference has to be booked, but no cash left the company. At the end of 2025, Anthropic held $20.28 billion in cash, cash equivalents, and short-term investments.

The current year is more revealing than the last. According to the Financial Times, Anthropic generated $11.5 billion in revenue in the second quarter of 2026 alone, meaning far more in three months than in all of 2025. On an adjusted basis, the company is on track for its second consecutive quarter of operating profit. “Adjusted” means certain cost items are excluded; it is not yet a statement about profit after all costs.

$518 billion for computing power

The biggest number in the prospectus concerns the future: Anthropic plans to spend $518 billion on cloud services, computing capacity, and infrastructure in the coming years. It has already locked in part of that this year through deals with Google, SpaceX, and Nscale, TechCrunch reports. The recent deal with Akamai fits the same picture: agents need not only graphics chips but also conventional processors.

For comparison, the sum is more than a hundred times last year’s revenue and about 25 times the cash on hand at the end of 2025. Commitments like these can only be met if revenue keeps growing at today’s pace and capital markets keep funding the company, which is exactly what the IPO is for. The pattern is familiar from OpenAI, whose multibillion-dollar data center bill we analyzed recently. The difference: for Anthropic, the numbers now come from an official document for investors for the first time.

The prospectus also flags a second risk. Nearly a quarter of 2025 revenue came from just two customers, whose names are not known. Many of the largest customers are not bound by long-term contracts and could cut or end their spending. For a company taking on hundreds of billions in fixed commitments, that dependence is one of the most important metrics.

A third of the prospectus devoted to risk

The scope of the risk disclosures is unusual. Roughly 80 of the 261 pages address potential dangers, while Anthropic needs only 48 pages to describe its actual business. The prospectus warns that AI models could pose “catastrophic or existential risks to humanity.” It names behaviors that models have already shown or could show: attempts to resist shutdown, to conceal or manipulate information, and actions resembling blackmail.

One passage goes straight to the question of how reliable safety testing really is. When a model recognizes that it is being tested, that is a “significant limitation” on assessing its safety. The UK AI Security Institute names the same problem in its current tests of OpenAI’s GPT-6 Astra.

You can read this cynically or soberly. Soberly, risk disclosures in US prospectuses also serve as legal protection: a company that names dangers up front is better shielded against later investor lawsuits. At the same time, the candor fits the public stance of Anthropic CEO Dario Amodei, who has repeatedly called this month for slowing the pace at the frontier of AI development. Critics such as Mistral CEO Arthur Mensch, by contrast, see this debate as a way to hold back smaller competitors with regulatory burdens. The prospectus gives both sides material.

What the IPO means for Claude users

For companies and developers using Claude, three points matter. First, the prospectus paints a more stable picture than the net loss suggests: quarterly revenue of $11.5 billion and reserves of just over $20 billion at the end of 2025 argue against a provider that is about to run out of money, even though the dependence on two major customers remains an open question. Second, with an adjusted quarterly profit, a phase is approaching in which investors look at margins rather than growth at any cost. Price cuts are not ruled out, but they are less of a given, as the pricing of Claude Sonnet 5.5 already suggests. Third, a publicly listed Anthropic has to publish figures and risks regularly. Anyone who builds their workflows on one provider gets far more visibility than before.

Outlook: the prospectus must go public soon

The document is not yet public, and Anthropic declined to comment to Reuters. Under US rules, the prospectus must be released at least 15 days before the investor roadshow begins; the target is a Nasdaq listing in October, led by Morgan Stanley, Goldman Sachs, and JPMorgan. The targeted valuation of more than $2 trillion would be more than double the $965 billion from the funding round in May. According to Reuters, OpenAI confidentially filed for an IPO in June and is expected to follow by early 2027. That will make the major model makers comparable on the stock market for the first time, and the public version of the prospectus will show whether the figures known so far are complete.