AI progress timing debate: same-node hardware gains deliver a one-time compute windfall

cis_female · x · 2026-09-12

In a debate on what would change if AI had arrived a decade earlier or later, one argument stands out: progress is downstream of compute, which depends on algorithmic/ML progress (incl. better chips) and slower-moving hardware technology. Much of today's compute gains come from adapting hardware at the same node to AI — matmul adaptation, lower precision, better instructions. If AI had boomed ten years earlier, these would have been a one-time huge jump; after the windfall, gains would settle to CPU-like 10-20%/year instead of 50-100%.

Original post →

More from AGI Musings

AGI Musings channel →