Anthropic's draft IPO filing shows $16.3B half-year revenue, an $80.7B loss and a $600B compute bill

Anthropic has confidentially filed a draft S-1 for a stock market listing, and the Wall Street Journal's look at it, reported on 20 and 21 October 2026, gives the first detailed view of the company's books. The company declined to comment. It is reportedly preparing to list as soon as November, with valuation talk of up to about $2 trillion, though CNBC has said a revised range starting near $1.5 trillion disappointed some who hoped for more.

The growth is steep. Revenue for the first half of 2026 was about $16.3 billion, against roughly $1.1 billion a year earlier and about $4.6 billion for the whole of 2025. Anthropic also told prospective investors that third-quarter revenue reached $16.8 billion, around 45% above the second quarter. About a fifth of first-half revenue came from just two unnamed customers, although one newsletter put that share at 25%, so the exact figure is not settled. The draft itself warns that many large clients are not tied to long contracts and could cut or stop spending.

“For the IPO, I do think investors will care less about whether the concentration exists today and much more about whether Anthropic can show that the next several billion dollars of revenue are coming from a broader customer base rather than the same few accounts getting larger.”

Shay Boloor, chief market strategist at Futurum, to Inc., as quoted by VKTR · vktr.com ↗

The losses are just as large. The net loss for the half was $80.7 billion, compared with about $13 billion a year earlier and $42 billion across 2025. Roughly $72 billion of it is non-cash, tied to the accounting for convertible notes and similar instruments, so the operating picture is less grim than the headline number. Spending on compute and infrastructure came to nearly $9 billion in the half, already more than the $7.3 billion of all 2025.

The larger worry in the filing is what Anthropic has promised to pay for computing power. It owes about $450 billion under cloud agreements running seven to ten years, mostly non-cancellable and due whatever the usage: roughly $111 billion to Google, $110 billion to AWS and $31 billion to Microsoft, plus up to $86 billion of Nvidia capacity through Nscale and Lambda. A further obligation of more than $160 billion to lease Google chips via Broadcom pushes the total past $600 billion, and deals still being negotiated could add as much as $150 billion. Dealroom reads the combined exposure as up to $750 billion.

Anthropic tells investors that its big cloud partners would not supply all the compute it asked for and increasingly regard it as a competitor. It is therefore leasing data centres and turning to third-party operators with limited track records. The risk factors, about a third of the document, also cover models showing self-preserving behaviour, liability for harm, a catastrophic AI incident that outruns its insurance, and organised opposition to new data centres. Investor Chamath Palihapitiya singled out the dependence on compute as the standout risk, and critic Gary Marcus doubts the valuation.

The filing follows a Blackstone-led debt facility that closed this month and Dario Amodei's warning in February that a lab buying compute far ahead of its revenue could go bankrupt. Separately, Anthropic suspended access to two models in June under export controls, restoring it on 1 July. Reuters reported in September that Amazon and Google held 60% of its outstanding customer bills at the end of 2025, and OpenAI is reportedly rushing a funding round that could value it above $1.4 trillion. A public version of the Anthropic filing is expected within weeks.