Anthropic filed a Form S-1 with the US Securities and Exchange Commission on October 26, disclosing $16.3 billion in revenue for the six months to June 30, 2026 and at least $265.6 billion in guaranteed compute commitments as it prepares to list on the Nasdaq under the ticker "ANT." The filing is the clearest public view yet of what a frontier AI lab earns and what it owes, and the two figures sit in sharp tension: revenue grew more than thirteenfold from $1.2 billion a year earlier, while the compute obligations run to roughly sixteen times that half-year revenue.
The revenue ramp is steep by any measure. Anthropic reported $41 million in 2023, $386 million in 2024 and $4.6 billion in 2025, then $11.6 billion in the second quarter of 2026 alone, up from $787 million a year before. It posted second-quarter operating income of $1.0 billion at an 8 percent margin, and narrowed its first-half operating loss to $693 million from $2.1 billion. A separate disclosure this week pointed to the engine behind that: the company said its coding tool Claude Code showed 900 percent net revenue retention, meaning the same customers spent nine times more in the third quarter of 2026 than a year earlier.
The compute number is where the risk concentrates. The $265.6 billion covers fixed or minimum guaranteed commitments with terms from one to 20 years, and the company has further deals with no fixed dollar figure where payments scale with capacity delivered. Press coverage has cited totals from $417 billion to over $600 billion once estimates and leases are added, but $265.6 billion is the contractual minimum in the filing. The S-1 describes a "multi-chip compute architecture" spanning AWS Trainium, Google TPUs, Nvidia GPUs and AMD GPUs, including a July framework with AMD for Instinct MI450 systems that could exceed $20 billion if Anthropic buys the full first tranche, against which AMD agreed to buy up to $5 billion of Anthropic stock.
Spreading work across four chip families lets a lab shift workloads to the cheapest available capacity, part of why mid-tier and fast-mode prices keep falling across the industry. It also makes the lock-in run both ways. Signing for compute up to 20 years out means Anthropic needs steady, growing token demand for two decades to pay for commitments it has already made. A supply-side report reinforced the squeeze the same week: SK hynix said on October 28 that memory deals already cover more than half its total capacity for both 2027 and 2028, and that it expects demand to outrun supply through 2030.
The filing also lands against Anthropic's unusual public posture on its own product's dangers. Its prospectus warns of catastrophic and existential risk, and the company has told California's governor it backs giving courts the power to suspend high-risk models. The S-1 separately notes Anthropic delayed pre-training of a new advanced model by more than a month to strengthen protections against theft of its model weights, a rare instance of a lab tying a schedule slip in a securities filing to security rather than capability.
Key terms remain open. Anthropic has not set a share count or price range, and Goldman Sachs and Morgan Stanley are named on the preliminary prospectus; those details should arrive in later amendments. Earlier reporting put the contemplated valuation near $2 trillion, a figure the filing does not itself confirm. The question an IPO forces into public view is whether revenue growing this fast can keep pace with obligations this large, and whether investors will price Anthropic as a software business at an 8 percent margin or as the counterparty on a quarter-trillion dollars of long-dated compute bets.