Anthropic and the Compute Mortgage

Anthropic’s compute commitments create a financing mismatch that could turn rapid revenue growth into severe cash deficits if AI demand or pricing weakens.

• Amazon invested about $18 billion in Anthropic and received a commitment for at least $110 billion of AWS spending over the next decade.
• Google invested $20 billion in 2026, while Anthropic committed to spend at least $114.5 billion on Google Cloud between April 2026 and July 2033; Microsoft and Nvidia invested up to $5 billion and $10 billion respectively, against at least $30 billion of Azure spending.
• Anthropic’s supply contracts include fixed multi-year minimums and take-or-pay clauses, while most enterprise revenue comes from API calls and cloud marketplaces where customers can switch from Claude to GPT or Gemini without exit penalties.
• AI clusters face economic obsolescence within 36 to 48 months, networking can represent 20% to 30% of cluster capital expenditure, and Amazon raised H200 instance prices by 15% in January 2026 and Nvidia GPU Capacity Block prices by about 20% in July.
• Anthropic reported $9.7 billion of research and development costs and $3.8 billion of cloud-platform fees in the first half of 2026, while operating expenses rose to $17 billion and the operating margin reached 8%; additional TPU commitments of $125 billion to $150 billion for 2028 to 2031 remain exposed to capacity and pricing conditions.

Takeaway: Anthropic needs demand growth and durable customer retention to spread its fixed compute costs across enough usage, while a slowdown from 1,000% growth to 30% to 40% could compress margins and create large cash deficits, making its IPO a test for the wider AI laboratory model.