Huawei Technologies has just open-sourced its core AI development toolkit, CANN (Compute Architecture for Neural Networks), designed to work seamlessly with its Ascend AI chips. This initiative promises to accelerate innovation by making the toolkit more accessible, as noted by rotating chairman Eric Xu Zhijun during a developer conference in Beijing.
This step marks a clear effort to reduce reliance on US technology and offers a homegrown alternative to NVIDIA’s well-established CUDA software stack. CANN version 8.0, released last September, plays a key role in Huawei’s broader AI strategy—as developers look for more flexible and user-friendly tools.
NVIDIA’s recent restrictions on using CUDA on non-NVIDIA hardware have spurred a closer look at domestic options. Meanwhile, the Cyberspace Administration of China has opened an investigation into NVIDIA’s H20 GPU for the Chinese market, citing security concerns, even as US lawmakers flag potential vulnerabilities in its chips.
Amid this shifting landscape, a new industry group called the Model-Chips Ecosystem Innovation Alliance was launched at the World Artificial Intelligence Conference by AI start-up StepFun. With backing from organisations like Tencent-supported StepFun, Infinigence AI and Huawei itself, this alliance is set to boost the adoption of local AI chips and support China’s drive towards technological independence.
If you’ve ever struggled with the constraints of closed systems, this move by Huawei might be the breath of fresh air you need to pace innovation in a challenging tech environment.
