Latest / Chips, compute and energy
DeepSeek and Huawei narrow AI chip gap with Nvidia, SemiAnalysis says
See how this story connectsSemiAnalysis published a report on 21 October 2026, by Bryan Shan and Dylan Patel, on what it calls the symbiosis between Huawei Technologies and DeepSeek. It says DeepSeek published Ascend support for six of its own software projects on 30 September 2026, TileLang, DeepGEMM, DeepEP, TileKernels, FlashMLA and DeepSelect, and that Huawei supplies the chips and their software stack while DeepSeek supplies the models and the engineering to run them on it. The report benchmarks an Ascend 950PR card with 28 cube cores and 128 GiB of memory, and says DeepSeek is likely to run primarily on Huawei’s stack by 2027 or 2028.
Why it matters
Kernels and compilers DeepSeek wrote for Nvidia chips now run on Huawei’s, the step a Chinese lab needs to train and serve frontier models without Nvidia hardware.
- When
- 21 Oct 2026
- ID
WTA-N-2026-0370- Where
- China, United States
- Who
- DeepSeek, Huawei Technologies, Nvidia, SemiAnalysis; Bryan Shan, Dylan Patel
- Topics
- Chips, Sovereign AI, Frontier models
Summary by wheresthe.ai, from SemiAnalysis and the reporting credited above. Checked against the source on 22 Oct 2026.
Source copy: download the page as saved on 22 Oct 2026 (PDF). Also held by the Internet Archive.
Cite this story
WTA-N-2026-0370: “DeepSeek and Huawei narrow AI chip gap with Nvidia, SemiAnalysis says.” wheresthe.ai, 21 Oct 2026. https://wheresthe.ai/d/cn-deepseek-and-huawei-narrow-ai-chip-gap-with-nvidia-semian-2026-10/ (source: SemiAnalysis).
Spotted an error? Email [email protected].