The Memory Wall in the Agentic AI Era
Micron, Wednesday, September 2nd, 2026
Micron's Jeremy Werner argues memory, not compute, is the next binding constraint on agentic AI systems.
In the first of a four-part FMS 2026 series, Micron's Jeremy Werner argues that the next constraint on AI will not be compute alone but memory: how much context systems can hold and how quickly they can move it.
Agentic workloads keep large amounts of state alive across many reasoning steps, which pushes capacity and bandwidth requirements well beyond what traditional inference sizing assumed.
The post frames this as a re-emergence of the classic memory wall in a new form. Micron positions its memory and storage portfolio as the lever for relieving that bottleneck across the AI stack.