DeepSeek and Huawei have released open-source programming tools for Huawei’s Ascend AI chips, including a language called TileLang. The move gives developers a new route to build for Ascend hardware, part of a broader effort to make China’s AI software stack less dependent on Nvidia’s CUDA.
DeepSeek Watch analysis
What happened
Businesskorea reports that DeepSeek announced TileLang through its official WeChat account and released it as open source. The package includes computation and communication libraries, and DeepSeek says TileLang offers a simpler programming approach than CUDA while making full use of hardware performance.
The report says DeepSeek has tested TileLang on older Nvidia chips as well as Huawei’s Ascend hardware. It also describes joint work to optimise a 128-chip Ascend 950 supernode. DeepSeek reportedly plans to deploy at least 160,000 Ascend 950DT chips at a data centre under construction in Inner Mongolia. Those plans and performance claims are reported by Businesskorea, not independently established results.
Why it matters
AI chips are only as useful as the software developers can use to make them do work. CUDA’s reach has helped make Nvidia’s platform a difficult act to follow; an open-source language and supporting libraries could make Ascend hardware easier to program and give developers more reason to try it.
The report also draws an important distinction: DeepSeek is still understood to use Nvidia chips for model training, while Huawei chips are mainly used for inference. The new tools point towards a more self-reliant stack, but do not show that the hardware or software has displaced Nvidia across the board.
Our read
This is more than a chip announcement: it is a bid to make the software layer travel with the hardware. Open source may lower the barrier to experimenting with Ascend, but the real test is whether developers can get strong performance without wrestling their projects into shape. A CUDA rival needs more than a name and a launch; it needs a useful ecosystem.
What to watch
- Whether TileLang and its libraries attract developers beyond DeepSeek and Huawei.
- How the reported 128-chip Ascend system performs on published, reproducible tests.
- Whether the planned Inner Mongolia deployment proceeds at the reported scale.
Discussion spark: For an alternative AI chip platform to challenge CUDA, what matters more: easier-to-use software, or proven performance on widely used workloads?
Sources and evidence
- DeepSeek, Huawei Join Forces to Challenge Nvidia’s CUDA Dominance – Businesskorea (2 October 2026, 05:29 UTC)
not affiliated with or endorsed by DeepSeek