SAN FRANCISCO — October 2, 2026
Artificial-intelligence company Anthropic has given investors an unusually detailed look at the economics behind frontier AI, revealing a business growing at extraordinary speed while simultaneously committing enormous sums to computing infrastructure.
The company’s IPO prospectus, reviewed by Reuters, showed that Anthropic generated nearly $4.6 billion in revenue in 2025, roughly twelve times its revenue a year earlier. But the company also reported a $42 billion net loss, with approximately $34 billion of that figure stemming from accounting charges related to financing instruments that could ultimately convert into equity.
The more revealing figure for operating economics was an $8.06 billion operating loss. Anthropic spent $12.65 billion on operating expenses, with computing and infrastructure accounting for more than half of that amount.
The disclosures illustrate one of the central challenges facing the AI industry: developing and operating increasingly powerful models requires vast amounts of computing capacity, while companies must simultaneously turn that capacity into recurring commercial revenue.
Broadcom enters the financing equation
The scale of the infrastructure build-out became even clearer this week when Anthropic’s filing revealed an agreement under which semiconductor company Broadcom could lend the AI company up to $42 billion to finance infrastructure spending.
The financing is linked to Anthropic’s plans to lease computing capacity based on Broadcom-designed chips. Reuters reported that Anthropic expects to become Broadcom’s largest customer in its chip-design business in 2027.
That arrangement illustrates how the AI boom is creating increasingly complicated relationships between model developers, cloud providers, chip companies and financiers.
Anthropic already relies heavily on technology giants such as Amazon and Google for computing infrastructure and distribution. Reuters reported that 47% of Anthropic’s sales in 2025 were routed through Amazon and Google’s cloud platforms, highlighting the company’s dependence on businesses that are simultaneously partners, investors or competitors in different parts of the AI ecosystem.
The prospectus has reportedly outlined a path toward a valuation potentially exceeding $2 trillion, although an eventual valuation and timing would depend on market conditions and the company’s eventual offering.
The IPO therefore represents more than a conventional technology listing. It could provide public-market investors with one of the clearest tests yet of how financial markets value an AI company whose revenue is growing rapidly but whose infrastructure requirements are also extraordinary.
Anthropic’s filing also contains extensive risk disclosures concerning advanced AI, including warnings about potential severe risks from increasingly capable systems.
For the wider technology industry, the financial question is becoming increasingly important: how much computing infrastructure can AI companies economically absorb, and how quickly can revenue growth catch up with the cost of building it?
Business takeaway: Anthropic’s disclosures demonstrate that the next phase of the AI race is not simply about software models. It is also about financing chips, data centres and computing capacity on an unprecedented scale.
Leave a Reply