Anthropic’s amended IPO prospectus shows $33,242 million of revenue and $4,812 million of cost of revenue for the nine months to 30 September 2026, a gross margin of 85.5% by our arithmetic. For the same nine months of 2025, the S-1/A filed on 2 November gives $2,545 million of revenue and $1,796 million of cost of revenue, a 29.4% margin.
The Information reported on 6 November that the prospectus shows gross margins “of around 85% in the first nine months of the year, excluding the cost of developing its models”, adding that “That may leave it room to lower prices if it needs to boost demand as computing gets costlier.”
Business Pill · Gross Margin
A short explainer on gross margin. It teaches the general idea only and does not discuss Anthropic’s filing.
The key insight: As we read the filing, Anthropic kept about 85.5 cents of each revenue dollar after customer-serving costs in the first nine months of 2026, up from 29.4 cents a year earlier, while model-development compute sits in research and development, by our arithmetic.
The S-1/A no longer reports gross profit as a line item. It says that during 2025 the company changed its presentation of operating expenses from a functional classification, under which cost of revenue and gross profit were separate lines, to a classification based on the nature of expenses. A supplemental table in the notes still groups operating expenses by function.
That table lists cost of revenue of $4,812 million for the first nine months of 2026, research and development of $19,240 million, sales and marketing of $6,619 million and general and administrative of $3,831 million. The filing says “Substantially all of cost of revenue consists of compute and infrastructure costs associated with customer-serving workloads”.
Gross margin by period, computed from the supplemental functional table in Anthropic’s S-1/A of 2 November 2026: revenue minus cost of revenue, as a share of revenue. The grey ‘after ALL compute’ bar subtracts total compute and infrastructure expenses from the statement of operations instead, a different measure. All percentages are our arithmetic.
What Sits Outside Cost of Revenue
Model-building compute is booked elsewhere. The filing says research and development expenses “primarily consist of compute and infrastructure costs associated with model development, research, and internal usage workloads”, plus personnel and other costs. Sales and marketing includes compute costs for “providing free access to the Company’s intelligence service”.
The statement of operations lists total compute and infrastructure expenses of $18,808 million for the nine months. Subtracting all of it from revenue leaves 43.4% of revenue, by our arithmetic. That is a different measure from gross margin: it counts training, research and free-tier compute along with serving costs, but excludes the personnel inside cost of revenue.
The Other Margins in the Filing
The prospectus reports a GAAP operating margin of minus 4% for the nine months and an Adjusted Operating Margin of 12%, a non-GAAP measure, with Adjusted Operating Income of $4,081 million. It says revenue reached $17.0 billion for the three months ended 30 September 2026 and that quarterly revenue grew 1,127% year over year, while compute and infrastructure expenses grew 382%.
The Pricing Question
The Information’s reading is that the margin may leave room to cut prices. Our earlier reporting found that the cheapest OpenAI and Anthropic main models cost 1% of the top price. OpenAI has not disclosed comparable figures: asked on CNBC on 6 November about 85% gross margins and 45% compute margins, OpenAI CFO Sarah Friar said, according to our machine transcript of the clip, “We have not talked publicly about our margins, but we do continue to invest in our business”.
As we read the table, revenue grew about 13.1 times year on year while cost of revenue grew about 2.7 times: $33,242 million and $4,812 million, against $2,545 million and $1,796 million a year earlier, by our arithmetic.
As we read the footnotes, the 85.5% is a serving margin; subtracting all compute and infrastructure instead leaves 43.4% of revenue, by our arithmetic.
As we read the adjusted figures, the filing shows a 12% Adjusted Operating Margin and a minus 4% GAAP operating margin for the same nine months.
The analysis says: “Inference is a revenue engine: continuous, distributed across millions of applications and agents, incurring marginal cost with every query.”
As we read it, cost of revenue in the S-1/A is that serving cost, while training compute is booked separately. The analysis does not discuss this filing.
What Is Not Established
The S-1/A does not break out compute by workload in dollars beyond the functional captions, so the split between training and research compute cannot be read from the filing. The 85.5%, 29.4% and 43.4% figures are our arithmetic from the filing’s tables, not ratios the prospectus states. We have read The Information’s summary post, not its full article.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growth story. This week it started being priced as a credit story.
Anthropic’s amended prospectus puts cost of revenue at $4,812 million against $33,242 million of revenue for the first nine months of 2026, an 85.5% gross margin by our arithmetic, up from 29.4% a year earlier, with model-development compute booked in research and development.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Dollar figures, footnote language and the non-GAAP margins are from Anthropic’s S-1/A of 2 November 2026. The Information is quoted from its post of 6 November. The Friar quote is from our machine transcript of a CNBC clip. All margin percentages are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
