AMP founder says poor scheduling and chip stalls waste most AI compute
Kyrannio · x · 2026-07-24
AMP PBC founder Anjney Midha argues that a huge share of AI compute is being wasted in the U.S. because of poor scheduling and low chip utilization.
- He says long-term leases can lose 30–40% of FLOPs simply from bad scheduling.
- Inside the chip, he estimates only about 15% utilization because work stalls on storage, memory, networking, or handoffs.
- Combined, he frames this as roughly 85% of FLOPs not doing useful work, and calls it a frontier and even national-security problem.
- His broader point: the industry needs both more supply and better use of existing capacity.
More from AGI Musings
- Ilya’s TED talk says AI could push people into unprecedented self-interested cooperation — AdrienLE · 2026-07-24
- The career skill that compounds for decades is learning fast, not Python or Kubernetes — _jaydeepkarale · 2026-07-24
- APSR study says rebel organization predicts which authoritarian regimes endure — RobbWiller · 2026-07-24
- A case for one human and one sovereign AI counterpart, built for continuity — Lesterpaintstheworld · 2026-07-24
- Biosecurity debate says bioweapon risk stays urgent as technology keeps improving — sebkrier · 2026-07-24
- DeepSeek call argues continual learning is still the main unsolved AI problem — hrishioa · 2026-07-24