Chinese artificial intelligence firm DeepSeek has released a suite of open-source software tools designed specifically for Huawei Technologies’ Ascend AI chips, marking a significant step in China’s efforts to reduce dependence on American technology, particularly chips and software from Nvidia. The announcement was made by DeepSeek on its official WeChat account, highlighting the collaboration as part of a broader national push for technological self-reliance amid ongoing U.S. export restrictions.
DeepSeek, headquartered in Hangzhou, introduced an Ascend-compatible version of its custom programming language, TileLang, which is intended to simplify the development of high-performance GPU and CPU kernels essential for AI processing. TileLang, originally optimized for Nvidia chips, now supports Huawei’s Ascend 950 accelerators, offering capabilities such as native code generation, automatic scheduling, and synchronization. DeepSeek described TileLang as a "universal, easy-to-program language" targeting the full utilization of chip hardware performance, with the Ascend version made freely available as an open-source project to encourage broader adoption and development within China’s AI ecosystem.
In addition to TileLang, DeepSeek has released several supporting libraries optimized for Ascend chips. These include DeepGEMM Ascend, a matrix multiplication kernel library, and DeepEP Ascend, which facilitates efficient large-scale cross-device communication. Other tools are designed to improve the efficiency of processing long data sequences and data filtering on Huawei’s hardware. Early tests reportedly show these software modules closely approaching the hardware’s maximum performance capabilities.
This collaboration reflects an effort to address one of the key challenges in developing domestic AI capability: not only producing advanced semiconductors but also creating the software infrastructure necessary to maximize chip performance. Though China has made significant progress in AI software development, led by companies such as DeepSeek, Zai, and Moonshot AI, its semiconductor manufacturing remains behind the cutting edge. Huawei has sought to bridge this gap with plans for more powerful AI computing systems and advanced Ascend chips targeted for release in 2028 and 2029.
Huawei has publicly emphasized the urgency of technological independence in light of U.S. export controls that restrict Chinese firms’ access to the latest Nvidia chips and advanced semiconductor manufacturing equipment. At a conference in Shanghai last September, Huawei Chairman Eric Xu underscored the company’s determination to avoid reliance on foreign suppliers, stating that China "cannot accept a destiny" defined by others’ willingness to sell critical technology.
The partnership between DeepSeek and Huawei has included close technical cooperation, such as optimizing AI supernode architectures powered by 128 Ascend 950 processors. Huawei has also pledged to prioritize domestic demand over exports in its AI chip production. These efforts align with Beijing’s broader strategy to achieve an “independent and controllable” AI hardware and software ecosystem, reducing reliance on U.S. technology while fostering domestic innovation.