The main focus on the deal is on inference workloads, the process of completing tasks and answers from an AI query — which is growing at pace with AI’s general expansion.
— Inference is hard. It’s wickedly hard, Buck told Reuters. — To be the best at inference, it is not a one chip pony. We actually use all seven chips.
Amazon is betting on a broad mix of chips, Reuters reports, and says in their press release that they are buying Blackwell and Vera Rubin chips.
From what Reuters understands, they will also be buying a number of the newly released Groq 3 LPX servers — which are optimized for inference and can do 700 million tokens per second.
Read more: Reuters report, Amazon press release.