2,200 MW. AWS Project Rainier — 500,000 Chips, Zero Nvidia
In October 2025, Amazon Web Services quietly switched on the world's largest concentration of non-Nvidia AI chips. Project Rainier — a 30-building, 1,200-acre campus in New Carlisle, Indiana — came online with roughly 500,000 AWS Trainium 2 accelerators drawing 420 megawatts in Phase 1 alone. There is not a single Nvidia H100, H200, or GB200 in the cluster. Instead, Amazon bet billions on its own silicon — and used the result to train Anthropic's Claude, the AI assistant that millions of people interact with every day.
What is AWS Project Rainier?
Project Rainier is Amazon Web Services' flagship owned AI training campus, purpose-built to run AWS's proprietary Trainium 2 AI accelerator chips at hyperscale. The site sits in New Carlisle, St. Joseph County, Indiana — chosen for its access to Great Lakes transmission infrastructure, available land, and proximity to Amazon's existing Midwest logistics and cloud operations.
The campus is designed for 30 buildings at full build, covering 1,200 acres. Phase 1 activated 7 buildings in October 2025 with ~420 MW of load. A $15 billion Phase 2 expansion was announced in November 2025, aiming for a final installed capacity of over 2,200 MW and more than 1 million Trainium chips total.
Anthropic — the AI safety company behind the Claude model family — has a deep partnership with AWS, and Project Rainier is the physical expression of that partnership: Claude is trained here.
The Numbers
The scale of Project Rainier's silicon bet is staggering. 500,000 Trainium 2 chips in Phase 1 — before Phase 2 doubles or triples the count — makes Rainier the single largest deployment of custom AI silicon outside of Nvidia's own validation labs. Each Trainium 2 chip is purpose-built for transformer model training, optimized for the memory bandwidth and compute density that large language models demand. The target of 1 million-plus chips by final build would make Project Rainier's total compute capacity, measured in AI-training FLOPS, competitive with the largest GPU clusters in the world.
How It's Powered and Cooled
Indiana's grid is managed by MISO (Midcontinent Independent System Operator), a region with a historically stable wholesale electricity market and significant coal and natural gas baseload — though increasingly supplemented by Midwest wind. Project Rainier's 2,200 MW final target is large enough to represent a material fraction of regional grid capacity, and Amazon has entered long-term power purchase agreements with regional utilities to secure clean power certificates alongside raw capacity.
The Trainium 2 chip's thermal design power is not publicly disclosed, but custom AI silicon typically runs at lower thermal envelopes than equivalent Nvidia hardware — one of the hidden advantages of vertically integrated silicon. Even so, 420 MW of Phase 1 compute produces immense waste heat. Indiana's climate — cold winters and moderate summers — gives Rainier a natural advantage in cooling efficiency relative to Sun Belt sites, with economizer cooling viable for much of the year, reducing chiller energy consumption significantly.
The Stromfee Connection
Project Rainier is the clearest case study in the energy intensity of AI model training. Training a frontier AI model like Claude requires sustained, uninterrupted compute — which means sustained, uninterrupted power. Every interruption costs millions in lost training runs. Energy reliability and price predictability are existential concerns for Amazon at this scale.
Stromfee's energy intelligence layer — BESS-Optimizer, day-ahead price forecasting, and HVAC pre-cooling scheduling — is designed for exactly this class of problem. For a 2,200 MW campus in a MISO market where day-ahead prices fluctuate significantly, intelligent energy dispatch and storage management can represent hundreds of millions of dollars in annual cost reduction while maintaining the uptime that billion-dollar training runs demand.
Sources: AI Datacenter Index — AWS Project Rainier Indiana · Data Center Dynamics — AWS activates Project Rainier · CNBC — Amazon opens $11B AI Data Center Project Rainier in Indiana