100T tokens/day traced to GLM-5.3-Flash on Chinese chips, validating export control impact
andrew_n_carr · x · 2026-08-27
The massive 100T tokens/day compute usage previously speculated to be Gemini has been identified as GLM-5.3-Flash running on domestic Chinese chips. This development validates Jensen Huang's observation on the Dwarkesh podcast that US export controls may be accelerating China's development of a NVIDIA-independent AI stack.
More from Companies & People
- Ex-OpenAI research lead Barret Zoph joins Google DeepMind — niloofar_mire · 2026-08-27
- HuggingFace Had No Other Bids at $13B; Nvidia Didn't Even Negotiate Down — RyanMcC35236715 · 2026-08-27
- Transformer Interview Fail: Candidates Clueless on Matmul — DrDatta_AIIMS · 2026-08-27
- List features nearly 16 coding agents; how many does the world need? — thisiskp_ · 2026-08-27
- AI empowers top creatives like GarageBand, doesn't limit imagination — rohanjamin · 2026-08-27
- ICML Chair Miro Dudík's multi-agent alignment experiment — misovalko · 2026-08-27