Anthropic's IPO prospectus reveals nearly 15x revenue growth and $600bn in compute contracts

Suppliers like Google, Amazon Web Services, Microsoft and Broadcom are tied to deep, largely non-cancellable contracts that run into tens or hundreds of billions, giving them multi-year revenue visibility from Anthropic. The structure supports the view that AI infrastructure demand is locked in rather than speculative, while also placing heavy counterparty risk on a single company that is still loss-making. For investors weighing a potential listing, the central questions are whether revenue expansion can stay ahead of computing costs and the fact that two customers made up around 20% of first-half sales. A valuation of up to $2 trillion offers little room for underperformance on either metric.

The draft prospectus for Anthropic reveals a business spending heavily on infrastructure, with its suppliers, customers and regulators all factoring into the investment thesis despite revenue growth outpacing costs.

Summary:

• First-half 2026 revenue came to approximately $16.3 billion, versus roughly $1.1 billion a year prior, per a draft prospectus seen by the Wall Street Journal
• Net loss reached around $81 billion for the half, with roughly $72 billion stemming from accounting charges on convertible notes and similar instruments; the loss a year earlier was about $13 billion
• Compute and infrastructure spending totalled nearly $9 billion in the half, exceeding the whole of 2025
• Infrastructure commitments surpass $600 billion, with added negotiations that could bring in up to $150 billion
• Two customers accounted for roughly a fifth of first-half revenue
• Risk factors cited include AI safety and liability, potentially uninsurable catastrophic incidents, restricted compute access and coordinated opposition to data centre construction

In the first half of 2026, Anthropic expanded revenue almost fifteenfold while pledging over $600 billion to future computing infrastructure, according to the Wall Street Journal's review of a draft IPO prospectus. Anthropic did not offer comment on the story.

Per the Journal, the AI developer brought in around $16.3 billion in revenue during the six months ending on June 30, up from about $1.1 billion a year before. That already outstrips the roughly $4.6 billion it recorded for the whole of 2025. Infrastructure and computing spending jumped to nearly $9 billion in the period, from roughly $2.3 billion the prior year, overtaking the company's 2025 total of about $7.3 billion.

The net loss figure was substantially larger. Anthropic posted a net loss of roughly $81 billion for the half, compared with about $13 billion a year prior and $42 billion across all of 2025. Of that total, roughly $72 billion represented accounting charges connected to value shifts on convertible notes and other fundraising instruments, not operational expenses.

The most notable element of the filing is the extent of Anthropic's infrastructure commitments. The Journal reports that the company owes roughly $450 billion on cloud agreements spanning the next seven to ten years, with most of it non-cancellable. This covers some $111 billion with Google, $110 billion with Amazon Web Services and $31 billion with Microsoft, in addition to up to $86 billion for Nvidia chip capacity from Nscale and Lambda. An extra obligation exceeding $160 billion to lease Google chips through Broadcom brings the aggregate above $600 billion, with negotiations on further deals that could contribute as much as $150 billion more.

Nevertheless, Anthropic stated that its main cloud partners have failed to supply all the capacity it required and increasingly regard it as a rival. The company is making up the shortfall with leased data centres and external operators, some of which it described as having brief track records.

Revenue concentration is an added factor. Roughly a fifth of first-half sales came from two customers, and Anthropic anticipates continued reliance on a small group of clients for a major portion of its income.

According to the Journal, risk factors occupy about a third of the prospectus. These cover potential harm from misaligned AI models, liability for harm like data theft, and events serious enough to be uninsurable. Anthropic also pointed to growing organised political opposition to data centre building, acknowledging that its own prominence could place a spotlight on related projects.

The submission arrives as Anthropic gears up for a public offering that could be valued at up to $2 trillion, exposing its growth trajectory, cost structure and client diversity to much deeper public scrutiny.