Anthropic’s reported IPO prospectus presents prospective investors with an unusually stark combination: rapidly expanding artificial-intelligence revenue, exceptionally large expected computing costs and warnings that increasingly capable AI could create severe harm. The document has not been made public, and Anthropic declined to comment when contacted by the Guardian, leaving the disclosures dependent on accounts of people and news organizations that said they reviewed or were briefed on the filing.
The reported details nonetheless frame a potential public offering as a significant test for capital markets. Reuters reporting summarized by CNBC said the Claude developer was seeking a valuation above $2 trillion while projecting $518 billion of cloud, computing and infrastructure spending in coming years. For investors, the issue is not merely whether demand for AI services is growing, but whether revenue can ultimately support the capital required to build and operate frontier models.
A prospectus not yet available for public review
Reporting on Sept. 28 and 29 described a prospectus that had not been released publicly. That limitation is consequential. A prospectus normally gives investors a primary document against which they can assess definitions, accounting notes, time periods and management assumptions. In this case, the reported figures, risk language and prospective valuation should be treated as attributed accounts rather than independently verifiable filing disclosures.
The Guardian said Reuters and the Financial Times reported that the document warned advanced AI could pose catastrophic or existential risks to humanity. CNBC’s account of the Reuters report said the filing also discussed the possibility of models resisting shutdown, hiding or manipulating information, and behavior resembling blackmail.
Those reported scenarios are warnings about possible failure modes, not evidence that Anthropic’s models have exhibited them or that an imminent event is expected. Their inclusion would place safety considerations alongside familiar IPO risks such as competition, customer retention and regulation. It would also make those concerns part of the company’s formal investment-risk narrative at the same time Anthropic is pursuing wider commercial adoption of Claude.
Risk disclosures appear to outweigh business narrative
Reuters reportedly found that about 80 of the prospectus’s 261 main-body pages were devoted to risk factors, compared with 48 pages describing the business. On that basis, risk factors represented roughly 31% of the main text, while the business discussion accounted for about 18%. The risk section would be about 32 pages longer than the business section.
Page counts alone do not determine a company’s outlook; public-company filings are designed to disclose a broad range of contingencies, and a lengthy risk section is not a forecast that those outcomes will occur. But the reported allocation is notable because it brings together risks that often sit in separate investor debates: model safety, escalating infrastructure obligations, dependence on major customers and the difficulty of sustaining margins in a computing-intensive market.
Anthropic’s commercial case is substantial. Calcalist Tech, reporting on the alleged filing and Reuters’ figures, said the company generated nearly $4.6 billion in 2025 revenue. Yet Reuters’ reported $518 billion of future infrastructure spending is more than 112 times that annual revenue figure. That is not a spending-to-sales forecast: the spending is described as extending over coming years, while the revenue figure covers one year. Still, the comparison illustrates the scale of investment investors would need to evaluate, including the assumptions behind utilization, pricing and customer growth.
Loss figures require an accounting distinction
Reports described two different measures that should not be collapsed into one. CNBC said Reuters reported a 2025 net loss of $42 billion and operating losses of more than $8 billion. Net loss is the bottom-line accounting measure after items beyond the operating costs of running the business; operating loss more directly reflects the gap between revenue and operating expenses in the reported period.
Calcalist Tech said roughly $34 billion of the reported net-loss figure was an accounting charge associated with financing liabilities. If that characterization is accurate, it helps reconcile why the headline net loss greatly exceeds the reported operating loss. It does not make the operating deficit immaterial, nor does it establish how much cash Anthropic spent: cash flow, capital commitments, debt terms and the underlying accounting footnotes would require the actual prospectus or later financial statements to assess.
The distinction is especially important in AI infrastructure. Payments for cloud capacity, model training, data-center arrangements and contracted computing can be recognized differently depending on their structure and timing. A large accounting charge may not equal cash spent on operations, while a smaller operating-loss figure still may not capture all future commitments. Investors would need to separate booked expenses, noncash accounting effects and contractual obligations rather than treating any single loss number as the full cost of the company’s expansion.
Customer concentration and capital needs sharpen the listing challenge
CNBC reported that people familiar with the filing told the Financial Times that nearly one-quarter of Anthropic’s 2025 revenue came from two customers and that many large customers did not have long-term contracts. The reported concentration underscores a basic tension in enterprise AI: a small number of buyers can accelerate revenue quickly, but changes in their procurement plans, pricing demands or internal AI strategies can have an outsized effect on growth.
That risk would be amplified if the company’s infrastructure commitments are fixed or difficult to scale down. Major cloud and computing arrangements can secure scarce capacity needed to train and serve models, but they also raise the break-even point if customer demand or prices fail to develop as expected. The reported prospectus therefore appears to put two questions before prospective shareholders: how durable Anthropic’s revenue is, and how flexibly it can match its computing bill to that revenue.
Anthropic has not publicly confirmed an IPO, publicly released the prospectus described in the reports, or commented on its contents. Until a filing becomes available, the financial and safety disclosures remain a detailed but unverified account of what could become one of the market’s most closely examined AI listings.