Napkin math on the AI bubble: 100x cheaper models still mean 1000x more GPUs
gabriel1 · x · 2026-10-10
gabriel1's napkin math on the "AI bubble": if all data eventually passes through an LLM, that's 100,000x more data than today — so even if model costs drop 100x, we'd still need 1,000x more GPUs. A counterintuitive take suggesting compute demand may be far from peak.
More from Fun
- Chris Albon jokes: my only request is that super intelligence not be named SuperIntelligence — chrisalbon · 2026-10-10
- Musk shows off Grok simulating the Battle of Cannae, claims it can sim anything — elonmusk · 2026-10-10
- AI artist: critics deny my work is mine just because I took creative liberties — eschadiol · 2026-10-10
- "It's 2026 and people still excuse AI hallucination because humans lie too," says ML author Burkov — burkov · 2026-10-10
- Gary Marcus fires back at Noah Smith: AI's math leap vindicated my neurosymbolic advocacy — GaryMarcus · 2026-10-10
- CMU's Keenan Crane wraps AI-assisted 3D modeling thread, releases all model files under CC0 — keenanisalive · 2026-10-10