A draft of Anthropic's stock prospectus (S-1) has leaked. According to the document reviewed by The Wall Street Journal, the company plans to spend up to $750B on compute capacity in the coming years, with part of that sum tied to cloud deals still under negotiation. The same filing states that the net loss for the first half of 2026 reached $80.7B on revenue of $16.3B. According to the leaked data, third-quarter revenue hit $16.8B, and the company aims to hold its IPO in November 2026. Anthropic did not respond to requests for comment from the publications.
Key Figures From the Leaked Filing
The draft S-1 prospectus offers the most detailed picture yet of the company's financial condition. According to the document, Anthropic could take on up to $750B in commitments for compute infrastructure — one of the largest infrastructure plans ever announced in the artificial intelligence (AI) industry. For comparison: that sum exceeds the annual budgets of many countries and vividly shows how sharply the entire AI industry needs compute capacity.
In the first half of 2026 the company earned $16.3B in revenue, but its net loss reached $80.7B — more than six times the $13B loss in the same period a year earlier. At the same time, growth remains strong: third-quarter revenue hit $16.8B, up 45 percent from the previous quarter. In other words, the company earned more in a single quarter than in the entire half-year.
The report reveals one more important detail: a fifth of half-year revenue comes from just two customers. Such customer concentration is considered a serious risk factor for investors, because the loss of one key customer would hit revenue directly and significantly.
The trajectory of the numbers is also noteworthy: third-quarter revenue of $16.8B even exceeded the half-year figure of $16.3B. That means the company's growth rate accelerated sharply in the second half of the year — the 45 percent quarter-over-quarter increase is proof of it. Such acceleration ahead of an IPO is investors' strongest argument, though whether it can be sustained remains an open question.
What Lies Behind the $80.7B Loss
The document says the large loss is mainly tied to accounting adjustments on financing deals — meaning not all of that sum represents cash outflow. Even so, the figure shows the scale of costs that come with AI companies' growth rates: the price of compute capacity needed to train and run models is rising faster than revenue.
The essence of the accounting adjustments is that the funds the company has raised and the associated obligations appear in the financial statements as a one-time large loss. That is a familiar situation in the startup world, but at Anthropic's scale the numbers have reached unprecedented levels. The key question for investors: is this loss a temporary accounting effect, and when will the operating business start generating steady profit?
The fact that two large customers account for nearly 20 percent of revenue exposes the fragile side of the business. Ahead of an IPO, such concentration is usually seen as a valuation-dampening factor, since it makes future revenue streams harder to predict.
The Fight for Cloud Capacity
One of the most notable parts of the document is Anthropic's concern about cloud capacity supply. As TechCrunch writes, the company is struggling to secure enough capacity from AWS, Google, and Microsoft, because these giants have started to see Claude's maker as a competitor rather than a customer.
Anthropic is struggling to secure enough cloud capacity from AWS, Google, and Microsoft — these giants see Claude's maker as a competitor, not a customer. — TechCrunch
The paradox of the situation is that Anthropic's largest investors and infrastructure partners are those same companies — AWS and Google have invested billions of dollars in it. Now, however, their own AI models compete directly with Claude, and the question of priority in capacity allocation has become acute.
As an alternative, Anthropic is leasing its own data centers and working with third-party operators. The document notes, however, that these operators' 'limited operating history' may make them less reliable than large cloud providers. In other words, the company faces a choice: dependence on giants it considers competitors, or trust in inexperienced operators.
The number of developers and companies using the Claude API is also growing in Uzbekistan. The infrastructure strain on the provider is one of the factors shaping the pricing policy and stability of the service they use.
The Risk List: Liability Beyond Insurance Coverage
The risk-factors section of the leaked prospectus reads far more sharply than typical corporate warnings. The document states that the company could face liability for damage caused by AI models in excess of its insurance coverage. That is one of the most serious legal risks openly disclosed to investors.
Another risk is that models may exhibit behavior aimed at resisting shutdown or give responses 'resembling blackmail.' Such admissions provide new material for the public debate on AI safety and will certainly draw regulators' attention. The fact that the company did not hide these risks in its filing is seen as an attempt to comply with pre-IPO disclosure requirements.
Anthropic has previously made disclosures about 'catastrophic risks' — the leaked document continues that line of openness. This time, however, the risks are recorded not as theory but as official legal warnings in a stock filing, which gives them greater weight.
Another important item in the risk list is the warning that liability may exceed insurance coverage. This means the company is acknowledging in advance that its existing insurance protection may prove insufficient. For IPO investors, that is a direct signal that the potential financial impact could be unbounded. And the clause about models potentially resisting shutdown or giving blackmail-like responses shows how technical safety research has migrated directly into a business document.
The November IPO and the Market Picture
A week before the document leaked, the company told prospective investors that third-quarter revenue had reached $16.8B — up 45 percent from the previous quarter. According to sources, Anthropic aims to go public in November 2026; earlier reports also covered the company's IPO roadshow scheduled for November 9.
If the IPO goes ahead, it could become one of the largest public listings in the AI sector. Right now the market is weighing how to balance the $750B infrastructure plan against the $80.7B loss — that question will be at the center of the investor presentations in November.
The main dilemma facing investors is simple: on one side, revenue growing 45 percent per quarter and a leading position in the AI market; on the other, unprecedented costs, customer concentration, and legal risks. The leak of the S-1 draft has brought this debate into the public arena — the argument now plays out not only behind closed doors but in front of the entire market.
What the Leak Itself Means
The leak of the S-1 draft is different from an ordinary news leak. A draft stock prospectus contains a company's most confidential financial and strategic information and is normally seen only by regulators and a select circle of investors. Its emergence in the open indicates that the IPO process itself is in an active phase.
The fact that the plan is not final — the document says part of the $750B plan relates to deals still under negotiation — means the leaked figures could change later: once negotiations conclude, the final volume of commitments may differ from the figures in the current draft.
The gravity of the story is also confirmed by the fact that two authoritative publications — The Wall Street Journal and TechCrunch — produced independent stories based on the leaked document. The WSJ reviewed the document itself, while TechCrunch deeply analyzed its section on cloud partnerships. The two sources complement each other: one reveals the financial picture, the other the contradictions in the infrastructure strategy.
Against the Backdrop of the Infrastructure Race
The $750B plan is not an isolated event — it is part of a broader infrastructure race in the AI industry. Major labs are investing unprecedented sums to secure the compute capacity needed to train models and serve millions of users. The fact that Anthropic's document says part of these costs relates to deals still under negotiation shows that the plan is not final but expanding.
Against this backdrop, the company's move to lease its own data centers also looks logical: it reduces dependence on external providers and creates a guarantee of long-term capacity. But this strategy further increases capital expenditures — these very costs are one of the main sources of the $80.7B loss.
In the end, one question remains: how will the market value a $750B future plan against a current loss of $80.7B? The answer will become clear at the investor meetings in November — and the leaked document has already started that debate.
