Key facts
- Anthropic uses hardware from Amazon Trainium, Google TPUs, Nvidia GPUs, and AMD GPUs.
- The company manages compute as a single, fungible pool from multiple strategic partners.
- This multi-chip strategy could erode Nvidia's dominance and pricing power.
Anthropic is diversifying its AI compute strategy by utilizing hardware from multiple chip providers, including Amazon, Google, and AMD, in addition to Nvidia. The AI lab's S-1 filing, released Monday as it prepares for an IPO, details a sophisticated software system designed to manage compute resources as a fungible pool across these different chip families. This approach, developed over years, aims to optimize productivity and reduce dependence on any single supplier, potentially giving major AI companies more negotiating leverage and impacting Nvidia's market position.