Anthropic Cites $1.70 Revenue Per Compute Dollar Before November IPO

Investor Figures Ahead of the November IPO

Anthropic told investors this week that it earned $1.70 in revenue for each $1 of computing power it consumed in the third quarter, figures first detailed in a Financial Times report published on Thursday. The artificial intelligence company shared the numbers privately, weeks before a planned November initial public offering that reportedly targets a valuation above $2 trillion. The listing is weeks away, which leaves little time for independent scrutiny of the private figures. Third-quarter revenue rose 50% to $16.8 billion, and the company posted a second straight adjusted profit. Executives put operating margins at 11% and described the business as cash flow positive. The figures were presented at an investor event rather than through a regulatory filing. It is a sharp reversal from 2025, when the company spent $7.33 billion on computing power against roughly $4.6 billion in revenue, about $0.63 earned per compute dollar, and its operating loss topped $8 billion once write-downs tied mostly to earlier fundraising rounds are excluded. The new numbers carry their own qualifications: the second-quarter adjusted profit left out stock-based pay, and the 80% gross margin excluded training costs. Chief Executive Dario Amodei and Chief Financial Officer Krishna Rao presented the margins and cautioned that they could change as Anthropic invests heavily in compute and biomedical research. Much of that spending is already committed: no less than $518 billion across ten years for cloud services, computing capacity and infrastructure, plans that the report noted rising interest rates could slow.

The investor room heard a profitability story, but the secondary market's read is cooler. COINOTAG data shows the pre-IPO instrument, quoted under the ticker ANTHROPIC, last at $1,915.00, down 4.55% over 24 hours after moving between $1,888.00 and $2,008.20. The daily chart reads as a downtrend, with a bearish MACD signal and RSI (14) at 33.9. On COINOTAG's composite support and resistance scoring, the nearest floor sits at $1,873.02 with a score of 98 out of 100, and a second level at $1,804.92 carries 86 out of 100. Until the offering files publicly and trading reflects real diligence, those levels, not the private figures, set the range.

Growth Cools as Clients Build Their Own Models

Underneath the margin story, growth is decelerating. Revenue climbed from about $4.8 billion in the first quarter to $11.5 billion in the second, more than doubling within a single quarter, then rose 50% to $16.8 billion in the third, a slower pace than the first half of 2026 delivered. Over the summer, some investors told the Financial Times they expected annualized revenue to pass $100 billion by year-end. Research firm SemiAnalysis now projects roughly $74 billion, an upgrade from $65 billion at the end of July but well short of the summer's expectations. Even the lowered projection would leave annualized revenue far above the 2025 base of about $4.6 billion. Three pressures stand out. Meta and Microsoft, two large Anthropic clients, now rely more heavily on in-house AI models of their own, trimming demand for external providers; both remain customers, but each is shifting more AI work toward internal systems. Anthropic has also reduced distillation, the practice in which rival labs pay to train their models on the outputs of a stronger model; four people familiar with the matter put that paid usage at 5% to 7% of revenue, so the cutback removes a meaningful earnings stream. A price war adds the third pressure. Anthropic and OpenAI have both lowered prices as low-cost Chinese labs intensify competition, and Anthropic set the price of Claude Haiku 5.5 roughly 75% under its predecessor when it launched this month. None of these figures has yet appeared in a public filing or prospectus. Goldman Sachs and Morgan Stanley reportedly lead the November offering, which would convert the private disclosures, and the growth questions around them, into the basis on which public investors price the stock.