Anthropic IPO Filing Leak: $8B Operating Loss, Fast Growth

Anthropic's plan to go public now has numbers attached. Details from the company's confidential S-1 filing, the registration document a company submits to US securities regulators before an initial public offering, were reported late Monday by Reuters and the Financial Times. The filing was prepared in June. It describes a business whose revenue is rising very quickly, whose losses have also grown, and whose long-term computing commitments dwarf both.

Here is what the leaked document shows and why it matters beyond Anthropic.

Revenue: from hundreds of millions to billions per quarter

Anthropic, which is organized as a public benefit corporation (PBC), a US legal structure that obliges a company to weigh a stated public mission alongside profit, brought in $400 million in revenue in 2024. In 2025 that figure reached $4.6 billion.

According to the Financial Times, growth sped up further in the first half of 2026. Second-quarter sales alone came to $11.5 billion, which is more than double the company's revenue for all of last year.

Losses: the headline number and the useful number

The net loss for 2025 was $42 billion, more than five times the 2024 figure. That total is misleading on its own, though. It includes a $34 billion accounting charge linked to money Anthropic raised from investors, which does not reflect day-to-day spending.

The operating loss is the more useful measure, since it counts only the cost of running the business. It came to $8 billion in 2025. More than 91% of that went to computing infrastructure. In practice, Anthropic's losses are almost entirely the cost of the hardware needed to train and serve its models.

The trend has improved since then. The Financial Times reported that Anthropic is on course to finish the current quarter with an adjusted operating profit.

The $518 billion compute bill

The largest risk in the filing is on the spending side. Anthropic plans to spend $518 billion on computing infrastructure over the next ten years. Reuters reported that 80% of that amount is locked into contracts that either cannot be canceled or require payment for reserved capacity whether or not Anthropic uses it.

Nearly half of the cloud budget is going to three hyperscalers: Amazon Web Services, Microsoft and Google. Separately, Anthropic has roughly $161.2 billion in equipment lease contracts with Broadcom. In April, Broadcom and Google agreed to supply Anthropic with several gigawatts of TPU capacity. TPUs are Google's custom AI accelerators, developed together with Broadcom.

Because of these commitments, a single profitable quarter does not settle the question of profitability. Anthropic's costs are largely fixed for years ahead. Its revenue is not.

An 80-page list of risks

The filing's "Risk Factors" section runs to at least 80 pages. The spending commitments take up part of it. Much of the rest covers AI safety.

The prospectus warns that advanced AI models could pose "catastrophic or existential risks to humanity." It tells prospective investors that such systems may display "self-preserving behaviors," may "conceal or manipulate information," and may act in ways "resembling blackmail." It also acknowledges that Anthropic's own continued product development "could further increase the risk that our models cause harm."

Companies do not often tell investors that their core product could endanger humanity. For Anthropic, whose public identity rests on safety research, the language fits its existing positioning. It also puts those concerns into a legal document, where disclosure carries more weight than in a blog post.

IPO timing and valuation

Anthropic reportedly wants to list in November at a valuation of $2 trillion or more. The offering could raise as much as $100 billion. Nvidia, which invested $10 billion in Anthropic last November, may join as an anchor investor, according to reports.

The Bigger Picture

The filing suggests that frontier AI companies are now valued mainly on how fast demand grows compared with how much compute they have already committed to buy. Anthropic's revenue curve is steep. Still, $518 billion in mostly non-cancelable commitments means the company is betting that demand will keep rising for a decade. If growth slows, the cost of that capacity stays the same.

Competition adds pressure. Rivals are pushing hard on price, as OpenAI's recent GPT-6.1 Sol release showed. Lower prices per task could reduce revenue per customer faster than volume grows. Whether the adjusted operating profit holds up over several quarters is probably the most important number to watch after the listing.

The safety disclosures also deserve attention. As models move into always-on agents with real access to systems, behaviors such as concealment or self-preservation stop being purely theoretical. Once Anthropic is public, it will have to report on these risks to shareholders. It is worth watching whether that makes the company more transparent about incidents or more cautious about what it discloses.

For now, the key questions are whether the November timeline holds, whether Nvidia commits as an anchor investor, and how public-market investors price a company that is both growing very fast and warning about its own technology.