Anthropic’s New TPU Deal, Computing Crunch, The Anthropic-Google Alliance

• Source: Stratechery (Ben Thompson)
• Date: 2026-04-07
• Type: daily-update
• RDCO Relevance: Very High (Anthropic economics, compute constraints, Google-Anthropic dynamics)

Anthropic announced a 3.5GW TPU compute deal with Google/Broadcom, starting 2027. Anthropic’s run-rate revenue now exceeds $30B (up from $9B at end of 2025), with 1,000+ customers spending $1M+ annually (doubled in two months). Thompson notes the revenue curve is steeper than exponential, and the inflection coincides with the U.S. government standoff, which functioned as massive marketing.

Thompson quotes Amodei’s earlier caution about over-investing: even at 10x annual growth, buying too much compute risks bankruptcy if growth is “only” 5x. But Anthropic’s conservative approach means compute is now the binding constraint on revenue growth. Thompson estimates this deal covers roughly $100B in incremental Anthropic revenue and predicts more deals soon.

The Google alliance logic: Google has the most AI compute (mostly TPUs), is spending $175-185B in capex this year, and cannot use all of it for its own Gemini products. For Google, Anthropic winning enterprise over OpenAI is ideal: it keeps compute revenue flowing through Google infrastructure while weakening Google’s main consumer AI competitor.

RDCO note: Anthropic’s revenue trajectory ($9B to $30B+ in months) and compute constraints directly affect our access to Claude. The Google-Anthropic alliance means TPU-based inference will increasingly matter. Cross-ref: Jaya Gupta moat thesis (06-reference/2026-04-10-jaya-gupta-anthropic-moat.md).