Rolling Out What Models Already Do Would Take More Compute Than the World Can Build for Years
stevehou · x · 2026-09-14
A thread on what calls to slow frontier AI mean for the buildout: even if models never improved, deploying what they can already do would take more compute than the world can build for years. The slowdown debate is about future generations of models, not existing demand. Cited: Anthropic CEO Amodei said in May they planned for 10x growth but ran at an 80x pace in Q1; OpenAI president Brockman said in July they face a compute shortage.
Related event: Slowdown Debate Unlikely to Change AI Compute Buildout(2 posts)→
More from AGI Musings
- Should AI models pass NIST-style safety checks before shipping? Industry voices say it's simple — aronchick · 2026-09-15
- Physicist Sabine Hossenfelder on arXiv and AI papers: flag them, don't ban them — skdh · 2026-09-15
- Common AI forecasts ignore algorithmic breakthroughs, argues one skeptical poster — shauseth · 2026-09-15
- Yacine: AI can't get much smarter — only faster from here — yacineMTB · 2026-09-15
- "Name One Thing AI Actually Made Better": Dev's Viral Skeptical Rant Beyond Coding — sebpaquet · 2026-09-15
- Analysis: Anthropic pacing alone buys one month; coordinated US slowdown cuts risk to 21% — soleio · 2026-09-15