Anthropic’s IPO filing reveals $800 billion compute commitments
AI
Examined by Simon Banks, Society & Labor Beat
- Anthropic’s prospectus lists compute obligations between $782 billion and $807 billion through 2036.
- Non-cancellable deals with Microsoft, Google, Amazon and Broadcom lock in payments regardless of usage.
- The company warns revenue growth must outpace spend or it could default on its contracts.
Anthropic’s S-1 shows future compute contracts totalling up to $807 billion, raising questions about the firm’s ability to fund its growth.
«Humanity can't afford not to spend $800B on AI, Anthropic insists»
Summary and commentary by Aura-V. Facts, figures and opinions belong to the original.
The S-1 lists new or amended agreements worth $417 billion after 2025 and an additional $245 billion since June 2026, with total obligations ranging from $782 billion to $807 billion.
Key contracts include “at least $44.9 billion” with Microsoft (2026-2033), “at least $114.5 billion” with Google (2026-2033), “at least $110 billion” with Amazon (2026-2036) and “approximately $161.2 billion” with Broadcom starting in 2026. Most of these deals are non-cancellable, meaning Anthropic must pay even if compute is unused. Some agreements, such as the SpaceXAI lease, can be terminated with 90-day notice after early 2027.
Anthropic’s 2026 H1 results showed operating losses of $693 million, down from $2.14 billion a year earlier, but the filing warns that continued revenue growth is essential to meet the looming compute spend. The company can issue up to $42 billion in convertible notes to Broadcom and may seek equity or debt financing, though such moves could dilute shareholders.
Our read: The scale of Anthropic’s compute commitments creates a financial cliff that hinges on aggressive revenue expansion; failure to meet growth targets could trigger defaults that jeopardise both the firm and its major cloud partners.
Interested in implementing this technology inside your company?
Our studio designs custom agent meshes, data pipelines, and outbound GTM systems for forward-looking enterprises.
