Huawei wants to capture the market share of AI chips from Nvidia in China

Huawei is seeking a bigger share of the Chinese market for artificial intelligence chips controlled by Nvidia by helping local companies adopt rival silicon.

Leading AI companies in China rely on graphics processing units (GPUs) made by Nvidia to “train” large-scale language models, with the $3.4tn US chipmaker’s products considered critical to developing the technology.

Instead of challenging Nvidia on training, Huawei is positioning its latest Ascend AI processors as the hardware of choice for Chinese teams to respond to “predictably” running LLMs.

The Chinese tech giant is betting that innovation will be a big source of future demand if model training slows down and AI applications like chatbots become more widespread.

“Training is important, but it happens only rarely,” says Georgios Zacharopoulos, a senior AI researcher working on on-demand acceleration at Huawei’s Zurich lab. “Huawei is mostly focused on data, which ultimately serves more customers.”

According to company employees and Ascend customers, it is focused on the less technically difficult but more profitable way to run AI models trained on Nvidia products on Ascend chips. Because Nvidia GPUs and Ascend work Different softwareHuawei is helping companies use another software tool to make the two systems compatible.

Huawei’s push comes with top-down government support. Chinese officials have urged domestic tech giants to buy more of Huawei’s AI chips and pull out of Nvidia.

A person familiar with Nvidia’s operations in China said Huawei was seen internally as the country’s toughest competitor, saying its chip design capabilities were “advanced.”

Washington has tried to curb Beijing’s AI development by using export controls aimed at hindering the development of sensitive technologies in China.

Unlike their US rivals such as OpenAI and Google, companies cannot afford the best GPUs in China. But even though Chinese teams can only get Nvidia’s smaller H20 chips to meet export controls, the lower-capacity GPUs are in high demand because they’re considered better than local alternatives.

analysts and Huawei researchers Ascend said it was not yet ready to replace Nvidia for model training because of technical issues, such as the way chips in a vast “cluster” of AI chips interact with each other, and when training large models.

“While Ascend’s chips perform well on a per-chip basis, there is a problem with inter-chip communication,” said Bernstein China Semiconductor analyst Lin Qinyuan. “When you train a large model, you have to break it down into smaller tasks. If one chip fails, the software has to find a way for the other chips to take over without delay.”

Another challenge for Huawei is persuading developers to switch from Nvidia’s Cuda software, known as the company’s “secret sauce” to speed up data processing and ease of use for developers.

But Huawei’s soon-to-be-released and updated AI chip, the Ascend 910C, is expected to address these concerns as well. “We are expecting this new generation of hardware to come with improved software that will make it more accessible to developers,” said a Huawei employee who asked not to be named.

Huawei and Nvidia face tough competition. Chinese internet group Baidu and chip designer Cambricon have made progress in AI chip development. Meanwhile, in the US, Amazon and Microsoft are playing to capture more market share in chips as AI applications become more widespread.

Semianalysis estimates that chip consulting firm NVIDIA made $12 billion in sales in China last year, supplying 1mn H20 chips in the country, which is twice as much AI chips as Huawei’s Ascend 910B.

“Most of the AI ​​chips sold in China are Nvidia’s China-specific H20 GPUs. But as Huawei ramps up its manufacturing capacity, its lead is narrowing rapidly,” said Dylan Patel, chief analyst at SemiAnalyst.

Industry insiders have warned that Huawei’s AI chip push is also being limited by insufficient supply, with two prospective customers telling the Financial Times that they could not wait for the chips.

Huawei did not respond to requests for comment. Nvidia declined to comment.

Analysts said Huawei’s production could be challenged by US export controls that have forced Chinese factories to rely on outdated chip-making equipment.

The focus on demand suggests that it is this dynamic that differentiates Chinese AI from that of the US. Washington’s export controls mean China’s AI players are not in the same race as Silicon Valley rivals Meta, Elon Musk’s x.AI and OpenAI, as is large mega-cluster Nvidia to build its most advanced GPUs.

“Chinese companies are playing a different game. They’re paying more attention than the U.S. because they can get big efficiency gains by using less powerful chips, which means they can get to business faster,” Bernstein analyst Lin said.

Chinese companies are betting that they can stay competitive on AI by lowering cognitive costs, which in turn will make AI applications cheaper to run, he said.

Last month, Hangzhou and Beijing-based startup DeepSeek released its V3 model, which drew attention for its low training and reference costs compared to comparable models in the US.

The company proposed a new way to reduce the cost of running an AI model by focusing on specific parts of the input data. It also used the “mix of experts” method popularized by other Chinese AI BeginnersThis helps speed up the estimation as only part of the model is used to generate the response.

DeepSeek Huawei successfully patched the V3 into the Ascend, providing developers with detailed instructions on how to use the chip. FT has in the past. reported Huawei has sent engineers to help customers migrate from Nvidia to Ascend.

Additional reporting by Zhijing Wu in Hong Kong