Intelligence/Brief
Anthropic's two-cloud bet: why Trainium plus TPU changes the math
Custom silicon on two hyperscalers gives Anthropic the lowest unit cost in the frontier — and the biggest single-vendor risk.
Published May 15, 2026
Anthropic is the only frontier AI lab running its training stack across two custom-silicon platforms simultaneously: AWS Trainium2 (Project Rainier, ~400k chips) and Google TPU v7 Ironwood (up to 1M chips announced in late 2025). The combination is unique in the industry.
The economics
Trainium2's published unit economics are roughly 30–40% cheaper per training FLOP than equivalent H100 capacity when amortized over a 3-year contract. That gap is the structural reason Claude 4.6 Sonnet ships at $3 / $15 per million tokens — comparable to GPT-5 mini, not GPT-5. Opus 4.7's $15 / $75 prices in a margin on top of the Trainium cost base; Sonnet prices at near-cost to win seat share.
Ironwood is more expensive than Trainium2 in absolute terms but optimized for inference. The split that's emerging: training continues to run primarily on Rainier, large-scale inference shifts to Google TPU pods where the per-token cost is competitive with Anthropic's own Trainium2 inference deployment.
The risk
The flip side is concentration. If either Trainium2's roadmap slips (Trainium3 is targeted for late 2026 in Anthropic's planning docs) or Google deprioritizes Anthropic's TPU allocation in favor of its own Gemini fleet, Anthropic's compute timeline compresses. Unlike OpenAI (Microsoft + Oracle + CoreWeave) or xAI (its own data center build), Anthropic does not own the underlying capital stack.
What to watch
- Trainium3 tape-out cadence — Anthropic's 2027 training plan depends on it.
- TPU v7 allocation — whether Google ringfences Ironwood capacity for Gemini 4 will be visible in Anthropic's effective inference unit cost.
- Sonnet pricing — if Anthropic raises Sonnet rates, that's the first signal Trainium2 economics aren't holding.
Linked entities
Underlying signals
Amazon's $8B Anthropic investment locks in Trainium as primary training chip
Amazon committed an additional $4B to Anthropic in November 2024 on top of an earlier $4B — total $8B — with the condition that Anthropic adopt AWS Trainium2 as the primary training chip for future Claude models. The deal underwrites Project Rainier and makes Anthropic the anchor tenant for AWS's custom silicon roadmap.
Capex: $8BGoogle TPU v7 "Ironwood": 9,216-chip pods, 42.5 exaflops, dedicated to inference
Announced at Google Cloud Next April 2026, TPU v7 (Ironwood) is the first Google TPU generation purpose-built for inference rather than training. Each pod scales to 9,216 chips delivering 42.5 exaflops of FP8 compute, with 192GB HBM3e per chip. Powers Gemini 3 Pro / 3.5 inference at Google scale. Annual TPU spend ramped to over $40B for FY26.
Accelerators: TPU v7 Ironwood- PartnershipAnthropic + Google press release ↗
Anthropic adds up to 1M TPU v7 Ironwood chips via Google Cloud
Anthropic's November 2025 deal with Google Cloud expands its TPU footprint to up to 1M Ironwood (TPU v7) chips, supplementing the Trainium2 fleet on AWS. This makes Anthropic the rare frontier lab running heterogeneous custom silicon across two clouds. Deal value reportedly tens of billions over multiple years.
Accelerators: 1M · TPU v7 Ironwood - Compute clusterAWS re:Invent / Anthropic announcement ↗
Anthropic's Project Rainier: 400k Trainium2 chips across AWS multi-region
Anthropic's primary training cluster — codenamed Project Rainier — runs on a multi-region Trainium2 fleet AWS built specifically for them. By end of 2025 the configuration was disclosed at roughly 400,000 Trainium2 chips spanning sites in Indiana, Wyoming, and Mississippi. Trainium2's economics are central to Anthropic's ability to sell Sonnet at ~40% the price of equivalent-tier rivals.
Accelerators: 400k · Trainium2