In a compute world dominated by Nvidia, hyperscalers are looking to build out their own semiconductor infrastructure capable of training and inference workloads at scale.
We chat to AWS product manager Gadi Hutt about his company's approach, based on its 2016 Annapurna Labs acquisition. We talk about Trainium and Inferentia, how the company balances against its GPU fleet, and what it's cooking with Anthropic and Rainier.