Amazon invests up to $4 billion in Anthropic

Amazon announced it would invest up to $4 billion in Anthropic in exchange for a minority stake, with an initial tranche committed immediately and the remainder to follow. The deal made Amazon Web Services Anthropic’s primary cloud provider for training and hosting its models, and committed Anthropic to using AWS’s own Trainium and Inferentia chips alongside its existing hardware, with the companies also agreeing to collaborate on future chip development.

The investment came with governance conditions attached. Anthropic said its existing structure would be unchanged: the Long Term Benefit Trust, a body empowered to elect a portion of the company’s board independent of investor interests, would continue to govern the company, and Anthropic’s own Responsible Scaling Policy, published six days earlier, would continue to direct its development decisions regardless of Amazon’s stake. Claude access through Amazon Bedrock, AWS’s managed model service, was central to the commercial rationale, with Anthropic naming early enterprise customers including LexisNexis and Bridgewater Associates.

The deal gave Anthropic a second major hyperscaler backer, following Google’s earlier investment, and reflected a pattern that recurred across the industry that year: cloud providers extending capital to frontier labs in forms that were as much compute and chip commitments as financing, tying a lab’s future training runs to a specific provider’s infrastructure in exchange for cash the lab needed to keep training ever-larger models. For Amazon, lacking a frontier model of its own to rival GPT-4 or Gemini, the investment secured a stake in a leading lab’s commercial success and a flagship customer for Trainium, the chip line it was trying to establish as an alternative to Nvidia’s GPUs.