AWS treats DRAM as the scarce resource in agent infra with new AgentCore runtime

bookwormengr · x · 2026-09-19

AWS had the biggest hyperscaler presence at AI Infra Summit, pitching software control over inference at every layer, from custom agents to disaggregated serving.

Key points: AgentCore harness is now the on-ramp to Bedrock, and agent traffic is shaping how AWS configures Rubin and Trainium capacity. The new AgentCore runtime treats DRAM as the scarce resource — holding only actively used session memory and billing accordingly — while oversubscription packs more sessions per host and passes savings to customers. Commenters note AWS can also offer the "safest" place to run agents, isolating guest sandboxes from the host OS.

Original post →

More from coding & agent

coding & agent channel →