Atlas Research Note
What a Conversation With a Senior AWS Executive Taught Us About Serving Larger AI Models
Ben Bajarin · September 17, 2026
We recently spoke on background with a senior AWS executive about the future of AI infrastructure, compute, and model development. The discussion clarified the relationship between growing model size and the larger, tightly connected systems being built to serve those models. Much of that relationship comes down to keeping the model in accelerator memory, moving data quickly between its different parts, and handling enough simultaneous requests to make the infrastructure economical.