Anthropic Needs to Disclose Its Tokenomics Before Its IPO

Anthropic’s preliminary IPO filing offers the first largely audited view of a leading AI company, but leaves investors unable to judge key parts of its economics. Revenue reached $4.6 billion in 2025 and $33.2 billion in the first nine months of 2026. A reported $104 billion net loss over that period mostly reflects noncash accounting charges tied to rising company valuation; operating results are more mixed, shifting from a $974 million second-quarter profit to a $567 million third-quarter loss as compute costs surged. The company commits to computing capacity well in advance, while much of its revenue is usage-based and not guaranteed. Its cloud providers also act as investors, distributors, and compute suppliers: 44% of revenue came through their marketplaces, and Anthropic has made enormous multiyear spending commitments to them.

The authors argue that the SEC should require fuller disclosure before the IPO, both under existing rules and through AI-specific standards. Anthropic should file major compute contracts and report, partner by partner, what it pays cloud firms, earns through their marketplaces, and receives in investment. Investors also need clarity on which of roughly $895 billion in compute commitments are legally binding and when cash payments are due. More fundamentally, the filing omits token volumes, revenue and costs per token, and how much contracted computing capacity is delivered or in use. Standardized token definitions and comparable per-token figures would help assess whether infrastructure spending can generate returns. The case for transparency is heightened by governance arrangements that give founders 50.1% of shareholder votes while a trust selects most directors.