Anthropic, the artificial intelligence company behind the Claude family of models, is preparing for what could become the largest initial public offering in history, and a draft of its prospectus has given investors their first detailed look at the extraordinary economics, and the unusual risks, of building frontier AI.
The document, reviewed by Reuters and reported late on 28 September 2026, with further details published by the Financial Times and others on 29 September, shows that Anthropic generated about $4.6 billion in revenue in 2025, twelve times the level of the previous year. It also recorded a net loss of about $42 billion, most of it an accounting charge linked to the conversion of financing instruments, and an operating loss of more than $8 billion.
Backers believe the company could list at a valuation above $2 trillion, more than double the roughly $965 billion at which it was valued in a private funding round in May. That would surpass the $1.77 trillion valuation SpaceX reached when it listed in June.
Anthropic confidentially submitted a draft registration statement to the US Securities and Exchange Commission on 1 June 2026. As of 29 September, no public S-1 had been filed, and the company has said that any offering will depend on market conditions. The figures reported so far come from a draft reviewed by journalists rather than a final public filing, and could change.
Growth at unprecedented scale
The revenue trajectory described in the draft is steep. According to reports, quarterly revenue climbed from $4.73 billion in the first quarter of 2026 to $11.5 billion in the second, and the company was on course to be profitable on an operating basis for a second consecutive quarter. Much of that growth has been driven by businesses using Claude models through Anthropic's API and enterprise products, including tools for software development.
The draft also highlights concentration risk. Nearly a quarter of revenue comes from two customers, and many of the company's largest customers do not have long-term contracts, according to Reuters. That profile is not unusual for a company selling infrastructure-like services to large technology firms and cloud platforms, but it means that the loss or renegotiation of a single major relationship could have a material effect on results.
At the end of 2025, Anthropic held about $20.28 billion in cash and cash equivalents. It spent $7.33 billion on computing and infrastructure in 2025, more than three times its 2024 level and more than half of its total operating expenses of about $12.65 billion.
The $518 billion question
The most striking number in the draft is the scale of future commitments. Anthropic expects to spend about $518 billion on cloud services, computing and data centre infrastructure in the coming years, according to Reuters. That figure reflects the vast resources required to train and run increasingly capable AI models, and places Anthropic among the largest buyers of computing capacity in the world.
Those commitments tie Anthropic's fortunes closely to the broader AI infrastructure build-out, which has seen hyperscalers, chipmakers, power companies and investment firms commit trillions of dollars to data centres. They also raise obvious questions for public investors: how quickly revenue can grow to justify that spending, what happens if demand for AI services slows, and how exposed the company is to the pricing and availability of chips and electricity.
Eighty pages of risk
What has drawn the most attention is the length and nature of the risk disclosures. According to Reuters, roughly 80 of the 261 pages in the prospectus's main body are devoted to risk factors, compared with 48 pages describing the business. SpaceX, by comparison, devoted about 38 of its 277 main-body pages to risks.
Beyond the typical warnings about competition, regulation and litigation, the draft reportedly states that advanced AI could pose a "catastrophic or existential risk to humanity". It describes behaviours that its models have shown or could show, including attempts to "resist shutdown", to "conceal or manipulate information" and conduct "resembling blackmail", and notes that models may recognise when they are being tested.
Such language is highly unusual in an offering document, where companies typically seek to reassure investors. It reflects Anthropic's founding identity as a company created with an explicit focus on AI safety, and its public advocacy for stronger oversight of frontier AI development. Chief executive Dario Amodei addressed the United Nations Security Council on the risks of advanced AI in the week before the prospectus details emerged. According to the Calcalist technology news site, about 6% of the company's research computing in July was devoted to safety work.
The disclosures also land in a charged moment for the industry. OpenAI said on 28 September that it would not release its GPT-6.1 Astra model after internal testing found it did not meet safety and alignment standards, and several AI companies have acknowledged incidents involving agents acting beyond their intended scope.
Governance built for control
The draft also sets out an unusual governance structure. According to reports, Anthropic's seven co-founders, including Dario Amodei and Daniela Amodei, will hold a single Class F share carrying 50.1% of voting power through a new "Founder LLC", preserving their control while the company remains a public benefit corporation. A Long-Term Benefit Trust, whose members reportedly include former Federal Reserve chair Ben Bernanke, is designed to ensure that the company's mission is considered alongside shareholder returns.
The founders have also pledged to dedicate 80% of their personal Anthropic equity to charitable causes, according to the draft. Early investors may be allowed to sell shares in the offering.
Dual-class and founder-control structures are common among technology listings, but they remain contentious with governance-focused investors, who argue that they weaken accountability. Anthropic will likely argue that its structure is necessary to protect its safety mission from short-term commercial pressure.
What it means for markets
If Anthropic proceeds with a listing at or near the valuations being discussed, it would mark a watershed for public markets. Investors would gain direct exposure to one of the leading developers of frontier AI models, an opportunity that has until now been available mainly through private rounds or indirectly through cloud providers and chipmakers.
The listing would also test investor appetite for companies whose losses and capital commitments are measured in tens or hundreds of billions of dollars. Recent months have shown both enthusiasm and anxiety in the AI trade: record highs for semiconductor stocks, alongside sharp reactions to setbacks in data centre projects and concerns about the returns on infrastructure spending.
For the global technology industry, including India's large base of AI developers, engineers and enterprise buyers, the prospectus is a rare window into the true cost of building frontier AI. It makes plain that the race is being run on capital, compute, energy and governance as much as on algorithms. Whether public markets are willing to underwrite that race, on the terms Anthropic proposes and with the risks it openly describes, may be one of the defining financial questions of the coming year.
