Anthropic's Q3 operating profit turns to a loss as nearly $900 billion in compute spending weighs on results
Anthropic submitted a revised S-1 prospectus to the U.S. SEC, adding Q3 data: revenue of $16.956 billion, up about 12-fold year-over-year, with GAAP operating results swinging from a profit of $974 million in Q2 to a loss of $567 million. Q3 spending on chips, power and data center infrastructure reached $9.9 billion, up 77% quarter-over-quarter, with compute costs rising to 58% of revenue; as of the end of September, total committed spending reached $895 billion. The company said roughly $5 billion in new short-term compute contracts and about $2 billion in non-cash charitable matching expenses weighed on profit; excluding these, adjusted operating profit was $2.5 billion, with margin falling from 19% to 15%.
Why read this
AI score 73
The prospectus fully discloses compute contracts and profit fluctuations for the first time, which can be used to understand the binding relationship between AI companies' profitability and compute costs.
Topics
Body · AI translation
Article body
On November 2 local time, Anthropic submitted a revised S-1 prospectus filing to the U.S. SEC, adding complete third-quarter operating data to the first version of the materials. The filing shows that this Claude developer had just crossed the operating profit inflection point, then quickly fell back into the operating loss range, and its IPO valuation also faces a downward re-rating.
In the third quarter, Anthropic's revenue was $16.956 billion, up about 12-fold year over year; net loss reached $23.3 billion, higher than $12.1 billion in the same period last year. However, the net loss includes large non-cash book losses from changes in the fair value of preferred shares and convertible bonds. In the same period, the company's operating expenses rose 292% year over year to $17.5 billion. As of the end of September, Anthropic's total committed expenditures reached $895 billion, of which about $448 billion is expected to be incurred before the end of 2030.
The pace of rising computing power costs has already outpaced revenue. In the third quarter, spending on chips, electricity and data center infrastructure reached $9.9 billion, surging 77% from $5.6 billion in the second quarter, and computing power costs as a share of revenue rose to 58%, compared with 48% in the second quarter.
The prospectus shows that in the second quarter Anthropic achieved GAAP operating profit of $974 million, but in the third quarter, GAAP operating results turned to a loss of $567 million.
In response, the company explained that, first, there was about $5 billion in additional short-term computing power contract spending, and such on-demand emergency procurement of computing power is priced far higher than long-term locked-in agreements; second, there was about $2 billion in non-cash expense accrual from a charitable matching donation program.
Management also stressed that adjusted operating profit excluding the above items remained positive, with third-quarter adjusted operating profit of $2.5 billion, higher than $2.2 billion in the second quarter, though the adjusted operating margin fell from 19% to 15%.
But for secondary-market investors preparing to participate in the IPO, GAAP operating profit turning from positive to negative will itself rewrite the risk-pricing logic. Previously, in a research report in late October, financial data company Pitchbook had already cut Anthropic's target valuation to $1.42 trillion, nearly 30% lower than the previously widely expected $2 trillion. The report argued that short-term increases in computing power costs and opaque contract payment schedules should both be discounted in the valuation.
On customer data, although Anthropic still maintains strong resilience, concerns about a growth ceiling are slowly emerging. As of the end of September, the company had 11,900 enterprise customers contributing more than $100,000 in annual revenue and 250 customers at the $10 million level, covering the vast majority of leading large enterprises. The enterprise net revenue retention rate remains high, and existing customers continue to increase usage, indicating that Claude has formed real demand in enterprise workflows.
But the other side of the high retention rate is that revenue is highly tied to customers' continued expansion of computing power budgets. Once enterprises begin to shrink AI budgets, the revenue curve will fall much faster than traditional software SaaS. The prospectus also repeatedly warns that the company's revenue is highly dependent on token consumption, and industry price competition and fluctuations in customer budget cycles could quickly compress gross margin space.
Regarding the scale of computing power, Anthropic also updated the revised filing, with all signed and pending computing power contracts totaling nearly $900 billion. Most of the payments are minimum payment obligations that must be settled on schedule regardless of whether model usage meets targets. This set of figures will also affect Anthropic's profitability.
The cost structure clearly shows this operating pressure. In the first half of 2026, the company's computing power and infrastructure costs reached $8.945 billion, accounting for 54.9% of total revenue for the period, with more than half of revenue directly used to pay computing power costs. This means that for every 100 yuan the company earns, 55 yuan is used to repay computing power costs. Once an industry price war breaks out, enterprise customers cut AI budgets, or GPU chip prices rise or deliveries are delayed, profits that had turned positive will instantly turn negative, and cash flow will come under direct pressure.
Anthropic also warned that costly short-term computing power contracts dragged down the third-quarter operating margin. With just one company, Musk's SpaceX, Anthropic committed to spending about $3.1 billion per month through 2029.
In terms of revenue structure, the company's usage-based API business still accounts for more than 80% of revenue, while personal subscription business accounts for less than 20%, and its commercial base is mainly focused on the enterprise side. But this also means that Anthropic's performance will be tightly bound to the pace of enterprise digital transformation.
Some commentary said this is the first time the AI industry has fully laid out "advance capacity bets on computing power" in an IPO prospectus for public investors to review. The profit window just validated in the second quarter relied on long-term low-cost computing power; once temporary replenishment is needed, profit immediately disappears. This shows that Anthropic's profitability is not the natural result of a stable business model, but a phased result brought about by the structure of computing power contracts. If it continues to rely on high-priced short-term computing power to fill gaps in the coming years, margins will remain under pressure, and operating losses may even recur.
As for the company's control structure, there has been no change. The founder team holds high-voting-power shares, and the Class A common stock received by public investors carries only economic rights. With the company carrying nearly $1 trillion in long-term computing power liabilities, this provision amplifies the risk exposure of outside shareholders.
The market's current pricing logic has already switched. Previously, funds were willing to give Anthropic a super-high valuation, betting that it could continue to replicate its second-quarter profit performance and forge a stable profitable path for AI commercialization. After the revised S-1 disclosed third-quarter data, the market began to reprice its cyclicality.
Anthropic's IPO story has been simplified from "AI commercialization has been proven" into a long test of computing power fulfillment. Whether it can continue to push customer usage higher to match the rolling annual computing power bills will determine the valuation range the company ultimately lands in.
(Source: Jiemian News) Mid-article embedded advertisement