In November 2025 Amazon Web Services announced that Project Rainier, one of the world’s largest AI compute clusters, had become fully operational less than a year after it was first revealed. AWS said the system was built collaboratively with Anthropic and featured “nearly half a million Trainium2 chips,” with Anthropic expected to be running more than 1 million Trainium2 chips across training and inference by year end. AWS described it as providing more than five times the compute Anthropic used to train its previous models and being 70 percent larger than any prior AI platform in AWS history.
Notably, Project Rainier runs on Amazon’s own Trainium2 silicon rather than NVIDIA GPUs. The cluster is spread across multiple US data centers - including a site in St. Joseph County, Indiana - connected with high-speed NeuronLinks within servers and Elastic Fabric Adapter networking between them. It is the infrastructure on which Anthropic trains and serves its Claude models.
The project was the centerpiece of a deepening Amazon-Anthropic relationship in which Anthropic committed to spending tens of billions of dollars on AWS compute, and AWS gained a flagship customer for its custom AI chips against NVIDIA’s dominance.
Why business readers should care: Project Rainier shows a hyperscaler using a marquee AI lab to validate its own silicon and break dependence on NVIDIA. It also illustrates the scale of modern training - hundreds of thousands of chips across multiple data centers - and the multibillion-dollar, multi-year compute commitments now underwriting frontier AI.