Amazon's Project Rainier scales toward 5GW Anthropic compute on Trainium
Indiana's $11bn AWS megacampus runs ~500,000 Trainium2 chips for Anthropic; the partners now target up to 5GW, with Amazon guiding to ~$200bn capex in 2026
Summary
Amazon's Project Rainier in New Carlisle, Indiana, an $11bn AWS megacampus, the largest capital investment in state history, runs nearly 500,000 Trainium2 chips training Anthropic's models. The partners have since signed for up to 5GW of new compute, with ~1GW of Trainium2/Trainium3 capacity targeted by end-2026; Trainium3 was co-developed with Anthropic. The full site is set to span ~30 buildings and draw
2.2GW of Electricity. Amazon is guiding to ~$200bn of 2026 capex, the largest of the four hyperscalers. Custom silicon is Amazon's lever to dilute Nvidia dependence while serving one frontier customer at industrial scale; power, not chips, is the named bottleneck.
By the numbers
- $11bn, Project Rainier investment (largest in Indiana history).
- ~500,000, Trainium2 chips in the activated cluster.
- Up to 5GW, Amazon–Anthropic compute agreement; ~1GW Trainium by end-2026.
- ~30 buildings, >2.2GW, eventual campus footprint and power draw.
- ~$200bn, Amazon's 2026 capex guidance.
Why it matters
Rainier is the clearest case of a hyperscaler building a dedicated, custom-silicon city for a single AI lab, vertical integration that, if it works, reroutes hundreds of billions away from Nvidia and ties Anthropic's roadmap to Amazon's grid and balance sheet.
What to watch
- Trainium3 ramp and whether the 5GW agreement converts to sited, powered capacity.
- Indiana/Ohio power sourcing (gas-plant acquisition) and local pushback.
- Whether custom silicon dents Nvidia's data-center share.