Leak Hints at Classified 1M-GPU Training Cluster in China That Skips HBM
teortaxesTex · x · 2026-11-19
teortaxesTex translates a post from infra blogger Zartbot hinting at a classified research project for a 1-million-GPU-scale training cluster that does not use HBM. The math: rumored Gemini 4 is 12T-total/500B-active params; training a 10T-scale model on 300T tokens at 30% MFU needs 300-400K B300s. With export controls and domestic chips at 1/4 Blackwell compute, that means 1 million domestic cards—raising HBM supply, interconnect, and co-design questions. Zartbot says answers are classified but estimates the cluster is worth ¥100-200B in revenue, implying a ¥500B-1T company.
Related event: Rumor: Huawei Secretly Building Million-GPU Training Cluster Without HBM(2 posts)→
More from Infra
- Microsoft's USB4 RDMA lets mini PCs pool memory to run Qwen3-235B — JFPuget · 2026-11-22
- LTX 2.5 keeps crashing on AMD 9070 XT with driver timeouts while MiniMax H3 runs fine — ZestycloseReading504 · 2026-11-22
- FT: AI server suppliers raise prices up to 45% months after deals are signed — nordicinst · 2026-11-22
- China Installed 59% of New Industrial Robots in 2026 Report, ~354,000 Units — ingliguori · 2026-11-22
- MLX over Vulkan hits 124 tok/s on RTX 5060 Ti, beating llama.cpp CUDA by 25% — SumitGup · 2026-11-22
- Intruder swapped Prime Intellect's hosted DeepSeek for an abliterated copy for 53 hours — mawcopolow · 2026-11-22