TL;DR — Key Takeaways
– DeepSeek and Huawei are collaborating on open-source programming infrastructure for Huawei’s Ascend AI processors.
– The effort includes compute and communications libraries and uses TileLang, an open-source programming language designed for AI chips.
– The partnership is intended to make Huawei’s AI hardware easier for developers to program and could help expand the Ascend software ecosystem.
DeepSeek has partnered with Huawei to develop open-source programming for Huawei’s Ascend AI processors, strengthening China’s effort to build an alternative to NVIDIA’s chip dominance.
DeepSeek is making programming infrastructure for Huawei’s Ascend platform open source, including compute and communications libraries. The project also uses TileLang, an existing open-source programming language designed for AI chips.
Making the Ascend programming infrastructure open source gives other developers access to tools for building on Huawei’s AI hardware. DeepSeek’s use of the already open source TileLang language could make the Ascend platform more accessible to AI developers.
The DeepSeek initiative helps Huawei and other companies attempting to compete with NVIDIA by providing the extensive software layer that allows developers to get high performance from the hardware.
NVIDIA, to be sure, has a formidable lead in this area. Its CUDA platform has become a standard development environment for AI computing, supported by an estimated four million developers worldwide. This large software ecosystem helps NVIDIA retain customers even as competing chipmakers develop alternative AI accelerators.
DeepSeek and Huawei are attempting to build more of that software infrastructure around Huawei’s Ascend processors. The companies have worked together to optimize computing and communications for a system containing 128 Ascend 950 chips.
DeepSeek said TileLang provides developers with a higher-level programming model intended to make coding AI processors easier while still delivering strong hardware performance. The company has also tested TileLang on older NVIDIA processors and uses the language as a core development tool in its work on AI systems.
Extended Partnership
Beyond software development, DeepSeek plans to deploy at least 160,000 of Huawei’s most powerful accelerators in a large data center under construction in Inner Mongolia. DeepSeek is also completing a funding round that could value the company at about $74 billion, ahead of a potential IPO.
Huawei recently announced new generations of Ascend processors and larger computing systems as it works to improve its ability to support AI model training.
The company plans additional generations of its flagship Ascend processors for 2028 and 2029. Huawei has also acknowledged that demand for its AI hardware inside China exceeds its current supply, so it is prioritizing domestic customers over overseas sales.
China’s AI software sector has moved faster than its domestic chip industry. Companies including DeepSeek, Z.ai and Moonshot AI have developed competitive models, while Chinese chipmakers continue working to narrow the significant hardware gap with leading U.S. technology.
U.S. export restrictions have added pressure to that effort by limiting Chinese access to NVIDIA’s most advanced AI processors and to the leading semiconductor manufacturing equipment.
