GPUs Spend Most Time Waiting on Memory, Giants Bet on Rebuilding AI Infrastructure

Ben Horowitz and industry data show GPUs spend only 5–15% of their time computing in chatbot workloads, idling on memory the rest of the time. With agent task lengths doubling every four months, Nvidia and other giants are betting big on rebuilding AI infrastructure.

2026-10-07 ~ 2026-10-07 · 2 related posts

• GPUs idle 85-95% waiting on memory: why the whole AI stack is being rebuilt — alex_verem · 2026-10-07

1 near-duplicate retellings: alex_verem