By ByteBot
Anthropic is filing its public S-1 today, and for the first time the numbers are not estimates — they are the company’s own figures. First-half 2026 revenue reached $16.3 billion. Net loss reached $80.7 billion. Compute commitments could reach $750 billion. If you build on Claude’s API, this document tells you more about the platform’s future than anything Anthropic has announced this year.
The Numbers, In Plain Terms
In the first half of 2025, Anthropic made $1.1 billion. In the first half of 2026, it made $16.3 billion — roughly fifteen-fold revenue growth in twelve months. Then Q3 came in at $16.8 billion alone: a single quarter that exceeds everything the company made in all of 2025, up 45 percent quarter-over-quarter.
The $80.7 billion net loss looks alarming until you read the next line: roughly $72 billion of it is a non-cash accounting charge tied to the valuation of convertible bonds Anthropic issued to raise capital. The real operational loss for the half was closer to $8.7 billion — still substantial, but not existential given the revenue trajectory. By Q3, Anthropic told prospective investors it had reached adjusted profitability and positive free cash flow, with an adjusted operating margin above 11 percent.
The $750 Billion Compute Bill
This is the number that matters most for developers. In H1 2026, Anthropic spent nearly $9 billion on compute and infrastructure — more than the $7.3 billion it spent in all of 2025. Its locked-in, noncancelable cloud contracts run to approximately $450 billion over the next seven to ten years: Google at $111 billion, AWS at $110 billion, and Microsoft at $31 billion. With all potential commitments counted, the ceiling reaches $750 billion. Read the full compute commitment breakdown for the details.
At first glance this looks like strength — massive infrastructure locked in ahead of demand. The filing reveals it is also a vulnerability. The three cloud providers holding those contracts are increasingly treating Anthropic as a competitor, and they have not supplied all the capacity Anthropic has requested. To fill the gap, Anthropic has been renting AI compute from Meta and SpaceX — entities the filing itself describes as having “limited operating history.” The S-1 warns explicitly that compute from hyperscalers “could be curtailed, repriced, or terminated.” The SemiAnalysis breakdown of the Amazon-Anthropic compute relationship is the best single explainer for why this risk is not theoretical.
What the API Numbers Tell You
Roughly 75 to 80 percent of Anthropic’s revenue comes from pay-per-token API usage and enterprise developer contracts. The API is the business, not a side product. The customer concentration story cuts both ways for smaller developers: the top 1 percent of customers account for 80 percent of revenue, and about 20 percent comes from just two companies — understood to be Cursor and GitHub Copilot, which together drove roughly $1.2 billion of 2025 revenue. More than 1,000 enterprise customers now spend over $1 million annually.
For developers outside that top tier: the Claude API will be maintained and improved because enterprise API revenue is the entire business model. The risk is that pricing moves to protect the $1.70 per dollar of compute Anthropic currently generates. That margin is thin. As a public company, there will be quarterly pressure to expand it. See the full revenue and loss breakdown from the draft S-1 for the complete picture.
The Valuation Debate
Anthropic entered the year targeting a $2 trillion valuation. By the October 14 investor day, discussions had drifted to around $1.5 trillion. Research firm New Constructs called it the most ridiculous IPO of 2026 and pegged fair value at $144 billion based on discounted cash flows. FutureSearch forecasts $1.65 trillion ninety days post-IPO. Both camps are reading the same filing — they disagree on whether $16.8 billion in quarterly revenue from a company growing 45 percent quarter-over-quarter justifies a frontier-era multiple.
Anthropic’s unusual answer to investor concerns about mission drift: a proposed 5 percent share transfer to a new public benefit fund, worth approximately $100 billion at a $2 trillion valuation.
What This Means If You Build on Claude
The S-1 pictures a company growing faster than almost any software business in history, constrained by compute capacity rather than demand, with thin but improving unit economics. The three things to watch after the IPO: whether compute availability holds (the hyperscaler squeeze is disclosed risk, not speculation), whether customer concentration improves (two customers driving 20 percent of revenue is a fragile foundation), and whether public-company quarterly pressure accelerates or complicates the Claude development roadmap.
The roadshow starts the week of November 9. If infrastructure decisions at your company hinge on Claude’s long-term availability, the compute squeeze section of this filing is the most important document you will read this quarter.
