DeepSeek Orders 160,000 Huawei 950DT Chips for Inference Rollouts, Not Training
teortaxesTex · x · 2026-09-06
Analyst Poe Zhao reports DeepSeek ordered at least 160,000 Huawei 950DT chips with no plan to train on them, keeping training on Nvidia. Commenters note 950DT outperforms 950PR for both training and inference, but DeepSeek needs them for rollouts — training is now mostly inference. Context: Meituan's LongCat-2.0 (1T params) was pre-trained from scratch on a 50,000-chip domestic cluster over 35T tokens with zero rollbacks; DeepSeek-V4-Pro (1.6T MoE) completed full post-training on 1,000+ Ascend 910C with 1,500+ interruption-free iterations. China can run frontier-scale training, but efficiency remains the open question.
More from Companies & People
- Affiniti Pivots Into an AI-Native Operating System for 3,300+ Main Street Businesses — jfiance · 2026-10-21
- CodeRabbit Ships 5,000 PRs a Week With a Tiny Team, Hires by Strategy-Game Hours — omojumiller · 2026-10-21
- AI safety researcher Sören Mindermann moves to Tsinghua, co-founds Beijing AI Safety Hub — dhadfieldmenell · 2026-10-21
- Runway joins Claude Marketplace, letting enterprises spend Anthropic budget on video generation — runwayml · 2026-10-21
- Pichai: everyone is massively short on AI compute — 'everyone wishes you had two times what you had planned' — firstadopter · 2026-10-21
- Open side sweeps all three open-vs-closed AI debates at GitHub HQ; Percy Liang on the enterprise IP bottleneck — stanfordnlp · 2026-10-21