Technology / Markets — Anthropic's draft IPO prospectus, reviewed by Reuters and The Wall Street Journal, shows a first-half net loss of $80.7 billion — but $72 billion of it is a non-cash accounting charge on the company's own success, and revenue jumped from $1.1 billion to $16.3 billion in a year. The filing is the first full look inside the AI industry's cost structure, and it has already knocked the expected valuation down from $2 trillion toward $1.5 trillion.
Anthropic's net loss widened to $80.7 billion in the first half of 2026, according to the draft prospectus for its planned initial public offering reviewed by Reuters. A year earlier the loss was $13 billion. The headline number is staggering — and, on its own, badly misleading. About $72 billion of the loss is an accounting charge reflecting the increased fair value of bonds that can convert into Anthropic shares: as the company's expected IPO value rose, the paper cost of the financing it raised on the way up rose with it. It is not cash that left the building.
What did leave the building is almost as remarkable. Anthropic spent nearly $9 billion on computing capacity and infrastructure in the first half of the year, up from $2.3 billion a year earlier — and more, in six months, than the $7.3 billion it spent in all of 2025. Against that, revenue grew to $16.3 billion in the first half, from $1.1 billion in the same period of 2025, and the company told prospective investors at an October 14 meeting that third-quarter revenue had jumped a further 45 percent to $16.8 billion, Reuters reported, citing people familiar with the matter. The company is expected to go public in November. Anthropic did not respond to a Reuters request for comment.
The numbers in the filing
The draft prospectus, also reviewed by The Wall Street Journal, fills in a picture that until now existed only as rumour and run-rate arithmetic. Anthropic generated $4.6 billion of revenue in all of 2025 and lost close to $42 billion that year, of which roughly $34 billion was the same kind of financing-related charge. First-half revenue of $16.3 billion is roughly fifteen times the year-earlier figure — growth few companies of this size have ever printed. The second quarter alone produced $11.6 billion, more than doubling the first; the third quarter's $16.8 billion, disclosed to investors last week, represents a slower sequential growth rate than the doubling investors had just seen, and Reuters reported that some were disappointed by exactly that deceleration even as the absolute numbers eased their concerns about the cost of building frontier AI.
The commitments are the other half of the story. Anthropic has committed to spending at least $518 billion over a decade on cloud services, computing and infrastructure, and reporting on the filing puts total future commitments above $600 billion, with further deals worth up to $150 billion under consideration. The Journal's account of the prospectus breaks the cloud piece down: about $450 billion of cloud agreements over the next seven to ten years, most of them non-cancellable — including $111 billion to Google Cloud, $110 billion to Amazon Web Services and $31 billion to Microsoft, payable regardless of actual usage — plus up to $86 billion renting Nvidia chips through the AI cloud providers Nscale and Lambda, and more than $160 billion for Google chips through an arrangement with Broadcom.
Why this matters
Every number in the AI boom has, until this week, been an estimate supplied by someone with an interest in the estimate. The prospectus changes that. It shows that the industry's economics now run on two ledgers at once: an operating business growing at a genuinely historic rate, and a financing structure whose paper losses grow in direct proportion to the company's success — the more valuable Anthropic becomes, the larger the charge on the instruments that funded it. It also shows the real constraint. The operating burn is dominated by one line: compute. Nearly $9 billion in six months, committed forward at a scale — over $600 billion — that exceeds the annual revenue of almost every company on earth.
That is why the valuation conversation moved. Reuters reported last week that some investors are now weighing a valuation of around $1.5 trillion, down from figures approaching $2 trillion discussed earlier in the marketing process. Nothing about the business got worse in the filing; what changed is that the costs are now documented rather than guessed, and documented costs price differently. Anthropic's own earlier disclosures had already warned investors, in the risk section of its September filing, that AI could pose what it called existential risks — a warning this site covered at the time. The new document adds the financial version of the same candour: the risks are existential, and so are the bills.
Who gains, who carries the risk
The clearest gainers are the suppliers named in the commitments: Google, Amazon, Microsoft, Broadcom and Nvidia hold non-cancellable claims on hundreds of billions of dollars of Anthropic spending. The clearest risk-bearers are the IPO investors being asked to underwrite the other side of those claims — revenue that must keep compounding at extraordinary rates for a decade to cover contracts that cannot be cancelled if it does not. Anthropic's defence, visible in the same numbers, is the growth itself: a company that went from $1.1 billion to $16.3 billion of half-year revenue in twelve months, and told investors it was profitable on an adjusted basis in the third quarter, is asking the market to price the trajectory, not the snapshot.
Critics of the AI build-out will read the filing differently: as confirmation that the industry's economics depend on capital markets remaining open and patient, and that a single lab has now contracted obligations on a sovereign scale. Both readings are honestly available from the same document. What is not available any longer is the argument that nobody knows what frontier AI costs. As of this filing, the price is in writing.
What happens next
Three things will decide how this listing lands. First, the public version of the prospectus: the draft was confidential, and the final document will fix the offering's size, price range and the exact language of the risk factors. Second, the growth rate: the third quarter's 45 percent sequential rise is the number sceptics will annualise and bulls will call a floor. Third, the comparison that arrives with it — OpenAI is on its own path to a listing, and the market will price the two labs against each other from the day Anthropic's ticker trades. The November debut, expected after the midterm elections, is now the best-documented IPO in the industry's history before it has sold a single share.
Related coverage: Anthropic IPO filing warns investors AI could pose 'existential risk' · Barclays cuts 5,000 jobs in AI cost drive
Track it live: market indexes and prices.
Sources: Reuters, "Anthropic logged net loss of over $80 billion in first half of 2026," October 20, 2026 (Echo Wang) — draft prospectus figures, $72 billion convertible-bond fair-value charge, $518 billion commitments, October 14 investor meeting, third-quarter revenue $16.8 billion; The Wall Street Journal, reporting on the same draft prospectus, October 20–21, 2026 (via Hindu Business Line / ANI and Stocktwits) — first-half revenue $16.3 billion vs $1.1 billion, 2025 revenue $4.6 billion and loss near $42 billion, commitments above $600 billion, cloud breakdown (Google Cloud $111 billion, AWS $110 billion, Microsoft $31 billion, Nscale/Lambda up to $86 billion, Broadcom/Google chips above $160 billion); Quartz, October 21, 2026 — 2025 operating loss above $8 billion and financing-charge composition. Anthropic did not comment to Reuters.