October 2, 2026 — Chinese AI firm DeepSeek has partnered with Huawei to develop and open-source programming tools for Huawei’s Ascend chips, in the clearest sign yet that China is building a full software stack to replace Nvidia’s CUDA ecosystem.
What happened
In a post on its official WeChat account this week, DeepSeek said it was open-sourcing programming infrastructure for Huawei’s Ascend platform, including a high-level programming language called TileLang plus related compute and communication libraries. DeepSeek says TileLang offers “a simpler programming model” than Nvidia’s CUDA, improving development efficiency and simplifying code logic.
Huawei “provided full support” in developing the infrastructure, according to DeepSeek. The two companies also jointly advanced a “supernode” solution linking 128 Ascend 950 chips, optimizing both computation and communication — a configuration meant to prove the platform can handle serious training workloads, not just inference.
The bigger picture
The announcement came about two weeks after Huawei unveiled its next generation of AI processors and supernode computing systems, saying it expects its AI systems to be widely used for model training next year. DeepSeek’s move addresses the real bottleneck for Chinese chips: not silicon, but software. CUDA’s two-decade head start is the moat that matters, and an open-source, developer-friendly alternative is exactly how challengers chip away at it.
Why it matters
If you’re watching where AI infrastructure money flows, this is the decoupling story in action. A credible CUDA alternative backed by China’s two most prominent AI companies — and released as open source — lowers the barrier for every lab and cloud outside the Nvidia ecosystem to migrate. For Western AI companies, the risk isn’t tomorrow’s hardware; it’s waking up to a parallel software stack they can’t easily compete with or port to.
Quick answers
What is TileLang?
A high-level open-source programming language for AI chips, developed by DeepSeek for Huawei’s Ascend platform. DeepSeek describes it as a simpler alternative to Nvidia’s CUDA for writing high-performance AI code.
What does the partnership cover?
Open-sourced compute and communication libraries for Ascend chips, plus a jointly developed “supernode” configuration of 128 Ascend 950 chips for combined compute and communication performance.
Why is this a challenge to Nvidia?
Nvidia’s real moat is CUDA’s software ecosystem. A credible open-source alternative from DeepSeek and Huawei could let developers move AI workloads off Nvidia hardware more easily.
Sources: Reuters, Open Source For You, nogentech.org.