The agreement enables OpenAI to immediately run workloads on AWS infrastructure, utilizing Nvidia GPUs in the U.S., with plans for future expansion. Amazon will build additional infrastructure specifically for OpenAI. This deal follows OpenAI's recent buildout agreements with Nvidia, Broadcom, Oracle, and Google, totaling approximately $1.4 trillion. While OpenAI will still invest significantly in Microsoft Azure, the AWS partnership provides crucial flexibility and capacity.
Amazon's investment in Anthropic, an OpenAI competitor, adds complexity to the cloud landscape. However, AWS's CEO Matt Garman emphasizes the breadth and availability of its compute resources as a key differentiator. The infrastructure will support both AI model training and inference, with potential for incorporating custom chips like Amazon's Trainium in the future.
OpenAI's models are already available on Amazon Bedrock, used by companies like Peloton and Thomson Reuters. This agreement establishes a direct customer relationship, as OpenAI commits to purchasing compute capacity from AWS. The move is seen as a step towards a potential IPO for OpenAI, signaling operational maturity and independence.