China has taken a giant step forward: it now has an alternative to NVIDIA’s CUDA for AI that allows its code to be recycled.

Huawei, which aims to capture a significant share of this market in China, has CANN (Compute Architecture for Neural Networks), its alternative to CUDA, but for now, the latter software dominates the market. And in the current situation, this is a problem for Chinese companies that want to stop using American hardware and software, as proposed by the government led by Xi Jinping. Having an effective alternative to CUDA is crucial for China right now, and it seems to be ready.

Moore Threads is China’s NVIDIA
Li Guojie, a computer scientist at the Chinese Academy of Sciences who is considered an authority in China, maintains that the most advanced Chinese AI GPUs, such as Huawei’s Ascend 910 chips, are comparable to current NVIDIA solutions in terms of computing power. However, this expert also argues that in the current circumstances, it is crucial for China to break down all the barriers imposed by CUDA.

“China must develop an alternative system to achieve AI self-sufficiency […] DeepSeek has had an impact on the CUDA ecosystem, but it hasn’t completely overcome it because of the persistence of barriers. In the long term, we need to establish a set of controllable AI software tool systems that go beyond CUDA,” Guojie says. Perhaps CANN will gradually break through and eventually gain traction, but as long as CUDA persists, AI development in China will remain somewhat dependent on the US.