Anthropic’s leaked IPO prospectus details steep losses, rapid growth, and a fear that AI could end humanity

Anthropic's leaked IPO prospectus shows $4.6 billion of 2025 revenue, up 1,088%, but $42 billion in net losses and an $8.06 billion operating loss, alongside $7.33 billion of annual compute and infrastructure spending. The Claude developer held $20.28 billion in cash at year-end but has committed to $518 billion in future cloud and infrastructure obligations, while two customers represented nearly one-quarter of revenue. Anthropic is reportedly targeting a valuation above $2 trillion in a potential post-U.S.-midterm IPO, despite substantial cash burn, customer concentration and AI safety risks.

Analysis

The key read-through is that frontier-model economics remain a capacity-allocation story rather than a software-margin story. Anthropic's contracted infrastructure pipeline creates unusually visible demand for hyperscaler AI capacity, but the financial benefit to AMZN and GOOG depends on whether commitments are take-or-pay, include cancellation rights, and earn returns above the cost of GPU depreciation and power buildout. Absent those terms, the market should not capitalize the full headline commitment: suppliers may be assuming customer-credit, utilization, and technology-obsolescence risk in exchange for revenue backlog.

For AMZN and GOOG, a successful listing would improve the mark-to-market value of strategic stakes and validate cloud demand, but that is likely a secondary earnings driver versus the potential dilution of cloud returns from bespoke capacity. The more important 6-18 month question is whether rapid model iteration shortens accelerator useful lives faster than depreciation schedules; that would pressure AWS/GCP margins even if reported cloud revenue accelerates. Smaller compute providers and SPCX could gain scarcity premiums, but their economics are most exposed if hyperscaler capacity loosens or the customer concentrates workloads back with AWS/GCP.

The prospective IPO valuation embeds a sharp transition from hypergrowth to durable, high-margin revenue despite meaningful customer concentration and limited contractual lock-in. Consensus may be underweighting the enterprise inference-price war: lower model prices can expand usage, but they also make it difficult for any frontier lab to convert revenue growth into free cash flow while maintaining training spend. A weak IPO aftermarket would matter beyond Anthropic, potentially compressing private AI infrastructure valuations and reducing the strategic-value premium currently assigned to cloud providers' AI investments.

Near term, this is more useful as an earnings-call diligence item than a directional catalyst for AMZN or GOOG. The thesis turns constructive for suppliers only if management quantifies multi-year AI backlog, utilization, and cloud-margin trajectory without a corresponding acceleration in capex; it is falsified if capex rises faster than AI revenue or if management identifies concentrated-customer credit/cancellation exposure. For any Anthropic IPO, pricing, lockups, share count, and the exact obligation schedule are required before establishing a position.

Key Decisions for Investors

• Maintain a 1-3 month relative-value bias long AMZN / short GOOG only if AWS discloses superior AI backlog conversion or margin resilience versus GCP; use a 5-7% stop on the relative spread. Amazon has greater potential strategic-equity upside, while the pair neutralizes much of the broad AI-capex factor.
• Do not treat the reported infrastructure commitment as incremental revenue for AMZN, GOOG, or SPCX until take-or-pay terms, duration, and supplier allocation are disclosed. Add an earnings-call alert for AI revenue, capex, depreciation, and cloud operating-margin guidance; a widening capex-to-revenue gap is a de-risking signal for hyperscaler longs.
• Avoid pre-IPO exposure to Anthropic proxies solely on listing anticipation. Reassess after prospectus pricing: consider a post-lockup short or put structure only if the implied enterprise value/revenue multiple materially exceeds mature software peers while gross-margin, retention, and committed-revenue disclosures remain weak.
• For 6-18 months, monitor AI accelerator depreciation and power-capacity commentary from AMZN and GOOG. Reduce cloud exposure if either signals materially shorter asset lives, rising unused capacity, or customer contract flexibility; these would indicate that AI revenue is being purchased with structurally lower returns on invested capital.