Huawei AI Chip Redesign Aims to Break Nvidia’s China Dominance
The Takeaway
• Huawei plans to develop a more flexible, general-purpose AI chip
• The new chip will make it easier for AI developers to transfer from Nvidia’s software
• Huawei has limited experience in general-purpose GPU design
China’s Huawei Technologies is plotting a fundamental redesign of its next artificial intelligence chip in a bid to seize market share from Nvidia.
The Chinese tech giant is working on a new AI chip design that would allow its chips to be used for a wider array of AI development work than its current ones can handle, according to two people with knowledge of the plan. That would bring Huawei’s chips closer to the design used by both Nvidia and Advanced Micro Devices, the people said, making it easier for Chinese AI developers to switch from Nvidia’s chips to Huawei’s.
As a result, Huawei hopes it can make its chips more appealing to Chinese tech companies, which still prefer Nvidia’s chips, even though U.S. export controls have made them increasingly difficult to buy. Most cutting-edge AI models in the country, including those from DeepSeek and Alibaba, are trained and deployed using Nvidia’s chips.
While local tech firms have tested Huawei chips, few have fully adopted them. That’s largely because of the difficulty developers face in adapting to Huawei’s distinct software system, after years of building products using Nvidia’s Cuda software, which runs on Nvidia’s chips.
AI developers using Huawei’s planned new chip design will be able to run the code they write based on Nvidia’s software—although Huawei will still need to develop software that translates the commands from Nvidia’s software into a language its own chips can understand.
Over time, the shortage of Nvidia chips could give Chinese tech firms little alternative but to use locally made chips. Since May, President Donald Trump’s administration has blocked Nvidia from selling even the modified AI chips it had formerly been allowed to export to China.
Nvidia CEO Jensen Huang warned in the latest earnings call that the U.S. export restrictions have made the $50 billion Chinese market inaccessible. “Export restrictions have spurred China’s innovation and scale,” he said on a recent earnings call.
Huawei didn’t have a comment.
Exploring Options
Huawei’s existing AI chips are part of a class of chips known as application-specific integrated circuits, which are designed for very specific purposes, such as certain AI calculation tasks. Nvidia’s chips, by contrast, use more standardized and established designs that can be used for a wider variety of AI tasks. Nvidia’s chips can therefore run all kinds of AI tasks very quickly, while Huawei’s chips are only efficient in certain calculations.
Huawei executives want its next-generation chip, the Ascend 920, to be a general-purpose graphics processing unit, said the two people. Such chips—similar to Nvidia’s—are capable of handling a wide array of AI tasks and accommodating various development methods.
Huawei is still working on the Ascend 920’s design, and the chip isn’t expected to go into mass production until next year at the earliest. Multiple teams are exploring various design options. If the new chip doesn’t perform at the level required, Huawei can switch back to the older design, the two people added.
The redesign won’t be easy for Huawei to pull off. The company lacks the decades of experience in designing traditional GPUs that Nvidia and AMD possess. Furthermore, U.S. export controls continue to restrict Huawei and its chipmaking partners from getting access to the most advanced equipment necessary to produce these more complex GPGPUs.
But if Huawei is successful, the redesign could dramatically improve its standing in the AI race. “Compatibility with Nvidia’s software is important for small developers, who often lack the resources to switch to a new system,” said Qingyuan Lin, a senior analyst at Bernstein covering the semiconductor industry in China.