Anthropic's Draft IPO Prospectus Shows $16.3 Billion in Revenue and an $80.7 Billion Mostly Paper Loss

Anthropic lost about $80.7 billion in the first half of 2026, almost entirely on paper, according to a draft IPO prospectus reported on 20 and 21 October by the Wall Street Journal and Reuters. The same document shows first-half revenue of about $16.3 billion, up from about $1.1 billion a year earlier, and compute commitments that run past $600 billion. The figures come from leaks of a draft that Anthropic has not commented on, and they have not been checked against a public filing.

What the draft prospectus shows

As investingLive summarised the WSJ's reading, the first-half revenue of about $16.3 billion already exceeds the roughly $4.6 billion Anthropic booked in all of 2025. Spending on AI infrastructure was nearly $9 billion in the half, against about $2.3 billion a year earlier and about $7.3 billion for the whole of 2025.

The loss is the headline number, but its composition matters more. Roughly $72 billion of the $80.7 billion came from accounting charges tied to convertible notes and other financing instruments, according to Al Jazeera's account of the Reuters and WSJ reporting. In plain terms, as Anthropic's estimated value rises, the fair value of instruments that can convert into its shares rises too, and that increase is booked as a loss even though no cash leaves the company. That leaves roughly $9 billion of loss from other sources, a figure derived here from the reported totals rather than stated in the filing coverage.

For comparison, investingLive reported the first-half 2025 loss at about $13 billion and the full-year 2025 loss at about $42 billion, a number Reuters first reported in September from the same IPO process.

The compute bill

The most consequential section of the draft may be the commitments. According to investingLive's summary of the WSJ report, Anthropic has agreed to roughly $450 billion of cloud spending over the next seven to ten years, mostly non-cancellable:

- about $111 billion with Google; - about $110 billion with Amazon Web Services; - about $31 billion with Microsoft; - up to $86 billion of Nvidia chip capacity through Nscale and Lambda.

On top of that, the filing reportedly includes more than $160 billion to lease Google chips through Broadcom, and talks that could add up to $150 billion more. Al Jazeera, citing the WSJ, said Anthropic indicated it could spend up to $750 billion in coming years on additional computing capacity, including deals under discussion. RuntimeWire repeated the $750 billion figure and the concentration risk, citing the WSJ.

Those obligations help explain the profile of the risk factors. Per investingLive, the risk section makes up about a third of the document and covers AI safety and liability, possible uninsurable catastrophic incidents and organised opposition to data centres. It also says the major cloud partners have not supplied as much capacity as Anthropic wanted and increasingly see it as a competitor, so the company is using leased data centres and third-party operators, some with short track records. Al Jazeera added that the risk factors include preventing products from causing harm, such as sophisticated cyberattacks or blackmail by uncontrolled models.

Who pays the bills

Customer concentration is the other flag. Two customers made up about a fifth of first-half revenue, according to investingLive's summary; RuntimeWire said only that the draft flags risk from a concentrated customer base. The reporting does not identify the two customers. Al Jazeera credited Claude Code, Anthropic's coding assistant, with driving the company's popularity.

The valuation question

The leak arrives as pricing talk for the November share sale is splitting. Reuters spoke to four people who attended Anthropic's investor day in San Francisco last week, and three said investors discussed a valuation of around $1.5 trillion, well below the roughly $2 trillion previously expected, Al Jazeera reported. investingLive framed the listing as one that could still value the company at up to $2 trillion. The two readings are not contradictory so much as a sign that the range is wide and the book is not built.

The Times previously reported that Anthropic told investors its third-quarter revenue reached $16.8 billion. The draft's first-half figure is a different, earlier period, and the two should not be added together or compared without care.

Why it matters for AI finance

Three points stand out for readers trying to size the business rather than the headline.

First, the $80.7 billion figure is mostly a function of Anthropic's own rising valuation. A large paper loss from convertible instruments is a by-product of success in private markets, and it is the sort of item that can reverse or vanish once notes convert at an IPO. It is a poor guide to how much cash the company burns, which the reported coverage does not state.

Second, the compute commitments are fixed costs measured in hundreds of billions of dollars, largely non-cancellable and spread over seven to ten years. Growth in revenue from $1.1 billion to $16.3 billion year on year makes that bet look sized to demand, but the filing's own risk language concedes the partners may not deliver capacity and may compete.

Third, three suppliers, Google, Amazon and Microsoft, are named for a combined $252 billion of the cloud commitments, in addition to the reported Google-chip lease through Broadcom. That concentration ties Anthropic's growth to a small group of companies that also sell competing AI products.

What we don't know yet

The numbers rest on a draft, and details differ between outlets. Al Jazeera's own subhead and body describe the year-earlier comparison inconsistently, so the prior-year loss is taken here from investingLive. The size and structure of the offering, the exchange and the final price range have not been reported in the sources reviewed. A public S-1 filing, when it comes, will settle which figures survive. Anthropic did not comment on the reports.