Anthropic orders 500K Trainium2 chips
Anthropic has placed an order for 500,000 Trainium2 chips on AWS, a huge inference/infrastructure commitment reported in industry commentary reported. That scale order illustrates hyperscalers and large AI buyers are shifting real spend toward custom‑silicon models reported.
Anthropic expanded its collaboration with AWS via a $4 billion investment announced on Nov. 22, 2024 (anthropic.com), a deal Anthropic said made AWS its primary cloud and training partner. AWS said it activated Project Rainier — a distributed UltraServer cluster built for Anthropic — and told partners it planned to scale Claude to more than one million Trainium2 chips by the end of 2025 (aboutamazon.com). AWS product documentation states Trn2 EC2 instances run up to 16 Trainium2 chips and Trn2 UltraServers can link up to 64 chips with NeuronLink, and AWS claims Trn2 offers roughly 30–40% better price‑performance versus P5e/P5en GPU instances (aws.amazon.com). CNBC reported Amazon opened a dedicated Project Rainier site in Indiana described as an approximately $11 billion data‑center project on about 1,200 acres aimed at hosting large-scale Trainium deployments for Anthropic (cnbc.com). Anthropic also announced an expansion with Google Cloud to access up to one million TPUs and “well over” 1 gigawatt of capacity coming online in 2026 (anthropic.com), a deal described by business coverage as being worth “tens of billions” of dollars. (cnbc.com) Reporting and company statements show Anthropic’s infrastructure footprint now spans AWS Trainium platforms, up to one million Google TPUs and continued use of Nvidia GPUs, reflecting a deliberate multi‑cloud, multi‑chip provisioning strategy. (datacenterknowledge.com)
