AI’s $518B compute bet

AI compute is becoming a decade-long commitment.

Reuters reported on September 29, citing Anthropic’s confidential IPO prospectus, plans to spend at least $518 billion across six infrastructure partners over the next decade.

The reported commitments include:

• About 80% is non-cancelable or payable regardless of usage
• $250 billion in minimum spending across Google, Amazon, and Microsoft
• $161.2 billion in largely non-cancelable Broadcom-related equipment leases
• Additional agreements with xAI and AMD

Anthropic also described compute availability as the primary constraint on future AI development.

The company is moving beyond a cloud-only model toward dedicated data centers and

directly leased chips.

This creates three strategic challenges:

• Demand forecasting over seven to ten years
• Dependency on providers who also serve as investors, distributors, and competitors
• Exposure to changes in pricing, technology, power availability, and model economics

My takeawayAI infrastructure strategy now requires the same discipline applied to capital allocation and supply-chain resilience.

Leaders need a portfolio approach across providers, silicon, power, networking, geography, and contract flexibility.

Is your AI infrastructure strategy prepared for demand growth, or committed ahead of demand?