Anthropic Prospectus Details At Least $518B in AI Infrastructure Obligations

Reuters directly reports that a confidential Anthropic IPO prospectus describes at least $518 billion of long-term AI infrastructure commitments, with approximately 80% non-cancelable or payable regardless of usage and specific obligations to Google, Amazon, Microsoft, and Broadcom-related equipment leases. This is a material, forward-looking contractual disclosure, so commercial_agreement is more specific than business_performance or business_strategy. I retain the existing-event match because the update advances the Anthropic compute-commitment storyline, but mark it non-duplicate because it adds materially more specific information on non-cancellability, counterparties, and payment horizons than the earlier aggregate leak. Importance is high due to the scale and multi-year implications, while surprise is moderated because the broader $518 billion figure had already been reported. The September 29 timestamp is the best available estimate for when the disclosure became available to the market; the underlying prospectus was confidential and its exact filing time is not provided. Approved stakeholders are limited to directly named counterparties and the explicitly identified Nvidia-based capacity exposure; indirect semiconductor, utility, financing, and other supply-chain read-throughs are excluded.