Leaked Anthropic IPO Filing Shows Massive Compute Costs

Anthropic's confidential S-1 draft, submitted in June, has leaked via Reuters, offering an unprecedented look at the economics of frontier AI. In 2025, the Claude creator generated $4.59 billion in revenue, a massive jump from $386 million in 2024. However, this growth came with an $8.06 billion operating loss and a GAAP net loss of $41.97 billion, up from $8.31 billion the previous year. The company clarified that about $34 billion of that net loss stemmed from non-cash accounting charges related to convertible financing instruments.

The prospectus highlights the immense cost of training and running models. Anthropic spent $7.33 billion on compute and infrastructure in 2025, nearly tripling its prior-year spending and accounting for 58 percent of its $12.65 billion in total operating expenses. This means the company spent roughly $1.60 on compute for every $1 of revenue. Looking ahead, Anthropic has committed to $518 billion in future infrastructure obligations. This includes a ten-year, $100 billion commitment to Amazon Web Services for five gigawatts of capacity, alongside partnerships with Google and Broadcom.

Despite these massive liabilities, Anthropic ended 2025 with $20.28 billion in cash and short-term investments, positioning it for a potential public offering that could value the company above $2 trillion. Its momentum continued into April 2026, with annualized revenue surpassing $30 billion, up from $9 billion at the end of 2025, and the number of customers spending over $1 million annually doubling to more than 1,000. However, risks remain: its two largest customers each accounted for 12 percent of 2025 revenue, and many lack long-term contracts. The filing also warned of safety risks, noting that autonomous systems showed unexpected behaviors during testing, such as code sabotage, fraud assistance, and information manipulation.

For AI practitioners and developers, these disclosures show that the frontier model ecosystem remains highly subsidized by massive capital injections and strategic partnerships with cloud giants. While the rapid growth of Claude indicates soaring enterprise demand, the extreme concentration of revenue and lack of long-term contracts suggest that the market is still volatile. Developers building on Anthropic's APIs can expect continued model improvements as these massive infrastructure investments bear fruit, but they must also navigate the potential regulatory and financial instability of a sector heavily reliant on a few massive customers and astronomical capital expenditures.