Reddit theory: the AI 'slowdown' talk masks a coming compute, power and datacenter crunch
maxpayne07 · reddit · 2026-09-13
A viral Reddit post argues the recent chorus of 'responsible slowdown' talk from Anthropic, OpenAI and Musk coincides with labs hitting hard physical compute limits.
Key numbers cited:
- Anthropic: 1M+ AWS Trainium2 chips, an up-to-5 GW Amazon deal worth $100B+ over 10 years, next-gen Google/Broadcom TPU capacity from 2027, plus all of Colossus 1 (300+ MW, 220k+ Nvidia GPUs).
- OpenAI: behind the Ohio AI campus with 4.25 GW initial IT load from Nvidia/SB Energy.
- Musk's Colossus: 200k H100s, roadmap toward 1M GPUs.
The author also notes OpenAI's own admission that coding agents have pushed experiments-per-researcher to an all-time high, with compute increasingly the binding bottleneck. Conclusion: safety concerns and competitive slowing are real, but so is the physical crunch — 'you can't pip install 5 gigawatts' — and behind the speeches, every lab's engineers are working around the clock on cheaper, less compute-hungry training.
Related event: Amodei, Altman and Musk unite on slowing AI, sparking debate over motives(62 posts)→
More from AGI Musings
- DeepMind's Economics of Intelligence: AI reshapes economy via agent coordination, not solo model smarts — Scobleizer · 2026-10-31
- Claude can build pro software for ~$20k in tokens — and TAM shrinks to match — bigblueboo · 2026-10-31
- Will AI kill the 9-5? A long read traces the workday back to Josiah Wedgwood's 1769 pottery factory — Konstantine · 2026-10-31
- AI memo: China to restrict exports of 23 AI technologies, plus Kimi K3.1 Pro and Genie 4 — dralandthompson · 2026-10-31
- Chinese firms offer AI 'ghost bots' that resurrect deceased loved ones from 30 seconds of footage — ArtificialOther · 2026-10-31
- SCOPE workshop proposed for ICLR 2027 asks what counts as an AI scientific discovery — mariyaivasileva · 2026-10-31