Deepseek and Huawei release open-source language TileLang for Ascend chips
In collaboration with Huawei, Deepseek has released as open source the TileLang programming language and computation libraries for Huawei Ascend chips. The goal is to create an alternative to the CUDA platform for Chinese AI hardware.
The company Deepseek announced on its official WeChat channel the release of open-source programming tools for Ascend chips from Huawei, Reuters reported. The released software includes libraries for computation and for data transfer between chips. According to Deepseek, Huawei "fully supported" the project, and the two companies jointly optimized a so-called supernode, i.e. a cluster of 128 Ascend 950 chips.
The core of the release is the TileLang language, originally developed by researchers at Peking University, which Deepseek says it has been using for about a year. The company presents it as a simpler programming model than Nvidia's CUDA, arguing that an independent software ecosystem for AI chips requires a universal language that is easy to use while still enabling full use of the hardware's performance. According to the NYT, TileLang was first tested on older Nvidia chips and is now Deepseek's main tool for work on artificial general intelligence (AGI).
The move responds to a key problem of the Chinese AI industry, where domestic chips need software that can extract their full performance. According to sources, Nvidia's dominance rests not only on chip design but also on an estimated four million developers worldwide working with CUDA, an ecosystem that competitors such as AMD have not yet matched. According to the NYT, Chinese model developers such as Z.ai and Moonshot AI have so far moved faster than domestic chip manufacturers. Two weeks before Deepseek's announcement, Huawei unveiled new AI processors and supernode systems, which it says it plans to deploy at scale for model training next year; at the same time, it admitted it cannot keep up with domestic demand and therefore plans to sell fewer chips abroad. Huawei's rotating chairman Eric Xu commented in this context that the company cannot accept a future dependent on others' willingness to sell chips to China, referring to US export restrictions.
Why it matters
For developers working with AI hardware, a potential alternative to CUDA is emerging that could lower the barrier to entry for programming chips outside the Nvidia ecosystem. For companies tracking AI hardware supply chains, this is a signal that China is building its own software layer as a response to US export restrictions, which may in the longer term affect the availability and competitiveness of alternative AI chips.
Two audiences, two different impacts
What this means
For individuals
Developers working with AI infrastructure are getting another open-source tool worth following as a potential alternative to CUDA, though for now it is tied to Chinese Ascend hardware.
For a business
Companies tracking supply risks and the competitive landscape in AI hardware should note that China is building an independent software ecosystem for Ascend chips as a response to export restrictions, which may over time affect the global supply of AI computing power.
StrategyCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.