Chinese AI company DeepSeek announced on Wednesday that it’s teaming up with Huawei Technologies to build programming tools tailored for Huawei’s Ascend chips.
It’s the latest example of Chinese tech firms pulling closer together as they look for a way around their dependence on Nvidia.
DeepSeek shared the news in a post on its official WeChat account.
The company said it’s making the underlying programming infrastructure for Huawei’s Ascend platform freely available, including tools for computing and communication between chips.
The headline release is TileLang, a programming language DeepSeek says makes it easier to build for AI chips.
Alongside TileLang, DeepSeek released a set of underlying libraries built specifically for high-throughput training and inference on Ascend chips.
These include DeepGEMM Ascend, a matrix multiplication kernel library, and DeepEP Ascend, which handles efficient, large-scale communication across devices.
It could eventually reduce how much the industry depends on Nvidia’s CUDA software, the platform that has quietly powered AI development for more than a decade.
A joint supernode effort
The announcement came two weeks after Huawei unveiled its next generation of AI processors and supernode computing systems, saying it expected its AI systems to be widely used for model training next year.
DeepSeek said Huawei provided full support in developing the programming infrastructure, and the two companies jointly advanced a “supernode” solution based on 128 Ascend 950 chips, optimizing both computation and communication.
“To build a new generation of independent, self-controlled GPU software ecosystems, the first priority is establishing a high-level language that is universal, easy to program, and still capable of reaching the hardware’s full performance potential,” DeepSeek said.
“TileLang was created precisely to meet this need,” offering “a simpler programming model” than Nvidia’s CUDA, the company said.
DeepSeek’s push toward domestic hardware has been building for months.
The company previously relied heavily on Nvidia’s H800 GPUs for training, but has increasingly adapted its models and software to Huawei’s Ascend platform in 2026.
Its V4-Flash AI model was optimized specifically for Huawei chips.
DeepSeek is also planning to install more than 160,000 Huawei Ascend chips at a new data center in Inner Mongolia.
Nvidia’s China headache gets more complicated
The news adds to a complicated week for Nvidia’s China exposure.
Beijing is separately considering allowing domestic firms, including Alibaba and ByteDance, to buy Nvidia’s RTX Pro 5500, according to The Information.
Nvidia sales staff have told customers to place orders by September 30 to secure supply, the report said.
Nvidia CEO Jensen Huang has said US export restrictions pushed the company’s share of China’s advanced AI chip market from roughly 95% to effectively zero, and H200 sales to Chinese customers made up less than 1% of Nvidia’s data center revenue in the quarter ended July 26.
Chinese chipmakers have used that vacuum to advance domestically.
Homegrown players including CXMT and Moore Threads have completed stock market debuts over the past year as Beijing pushes to reduce dependence on foreign chip technology.
Nvidia closed Tuesday at $227.21, down 0.72%, and was trading higher in premarket action Wednesday at $228.22, up 0.44%, as of about 6:07 a.m. ET.
The post DeepSeek partners with Huawei on chip tools to cut reliance on Nvidia appeared first on Invezz
