A CUDA alternative goes open source
DeepSeek and Huawei released Ascend compute and communication libraries plus TileLang support — the software moat around Nvidia is being attacked from China.
China competes with open weights and its own chips — 12 points behind the frontier, far ahead on reach
Chinese labs hold the top of the open-weights table (Xiaomi MiMo-V2.6-Pro 46, Z AI GLM-5.3 45, Alibaba Qwen3.8 Max 45, Kimi K3 44) at a fraction of US prices, and Qwen counts 151,448 derivatives worldwide. Under US export controls, DeepSeek and Huawei are building a parallel stack: V4 runs on Ascend chips and the Ascend tooling is now open source. Rules are clear but state-directed, including mandatory labelling of AI-generated content.
- Compute & energy3/5
- Capital4/5
- Talent & research5/5
- Business adoption4/5
- Rulebook clarity4/5
This week
This month
Winning the Global South
Free, adaptable Chinese models are becoming the default in emerging markets — including parts of Latin America.
What to watch next
- Restrictions — from Beijing or Washington — on cross-border use of Chinese open models
- Huawei's next Ascend generation and its supply volumes
- Licence changes on new Qwen releases (revenue-sharing for commercial users)
What to do
Chinese open models can cut AI costs sharply; if you use them, self-host or use an EU provider, check the licence and keep a non-Chinese fallback for regulated data.
Key news
Verified signals for China
Ranked by relevance. EU-wide signals are included for EU members.
Infrastructure
DeepSeek and Huawei open-source Ascend programming tools to cut reliance on Nvidia's CUDA
The release includes compute and chip-to-chip communication libraries and Ascend support for TileLang. China's AI stack is closing ranks around domestic chips, which matters for anyone pricing Chinese open models.
Oct 1, 2026 · CNTechNode ↗
Models
Chinese open-weight models are winning the Global South; Qwen counts 151,448 derivatives on Hugging Face
Foreign Policy argues that free, adaptable Chinese models are becoming the default in emerging markets. For companies, open weights mean cheaper customisation — with licence and export-control checks.
Sep 21, 2026 · CNForeign Policy ↗
Models
DeepSeek V4 lands as open weights and runs on Huawei Ascend chips
Two versions (Pro and Flash) with 1M context. China's open models keep closing the gap while reducing dependence on Nvidia — a structural shift in the model race.
Apr 24, 2026 · CNThe Register ↗
Contact us
Let's find your edge.
Tell us what slows your business down. In 30 minutes we'll show you where AI pays back first — and what it costs.
Book a free diagnosis info@alfageme.euReply within 1 business day · No phone calls unless you ask