Anthropic's IPO prospectus discloses a $42 billion net loss — the largest annual loss ever recorded by a company ahead of a public listing — alongside $518 billion in committed cloud, compute and infrastructure obligations that will flow, directly or indirectly, to the world's leading foundries, accelerator vendors and datacenter builders.

The scale of the spend is already visible in the operating numbers. Anthropic spent $7.33 billion on compute and infrastructure last year, more than half of its $12.65 billion total operating expenditure. Revenue for 2025 came in at $4.59 billion against an operating loss of $8.06 billion. That gap means the company is effectively buying silicon, power and datacenter capacity at roughly 2.5 times what it earns — a burn rate sustained by private capital and now, prospectively, public markets.

The prior record for a pre-IPO loss belonged to SpaceX, at $4.9 billion. Anthropic's figure is nearly an order of magnitude larger, and it resets expectations for what infrastructure-heavy AI companies will need to raise to stay on their current compute trajectories. For the semiconductor supply chain, the $518 billion in contractual obligations is the number that matters most: it represents locked-in demand for GPUs, custom accelerators, HBM, advanced packaging capacity and the foundry wafers behind them, stretching years into the future regardless of Anthropic's own profitability.

Customer concentration is a second structural risk. The prospectus discloses that two customers — widely believed to be Microsoft and Google — account for 24% of Anthropic's revenues. The filing also warns that many of its largest customers are not under contract and could stop spending at any time. That cuts both ways for hardware suppliers: the same hyperscalers that anchor Anthropic's revenue are the ones building their own accelerator fleets and in-house models, and they control the datacenter capacity Anthropic rents.

The prospectus also flags a novel liability category. It states that if Anthropic's agents went rogue and caused damage, the company could be liable for unforeseeably large losses — an acknowledgment that agentic AI products, which increasingly act autonomously on customer systems, carry tail risks with no established actuarial baseline. How insurers and underwriters price that exposure during the IPO roadshow remains an open question.

On valuation, reports suggest Anthropic is targeting as much as $2 trillion at the IPO. If achieved, that would be the biggest valuation at IPO in history, exceeding anything the semiconductor or software sectors have produced. The company is thought to have pushed its listing back to November, giving underwriters time to digest the loss figures and the obligations stack.

For chip vendors and foundries, the filing crystallizes the commercial reality of the AI buildout: demand commitments now run to half a trillion dollars from a single model developer, but the counterparty economics rest on revenue that is small, concentrated and largely uncontracted. The competitive dynamics of the next cycle — accelerator allocation, packaging capacity and cloud pricing power — will hinge on whether Anthropic and its peers can close the gap between compute obligations and recurring revenue before the capital markets reconsider the bet.