Huawei wants to gain market share in AI chips from China’s Nvidia

Huawei wants to get a bigger share of the Chinese market for artificial intelligence devices dominated by Nvidia, by helping local companies adopt their silicon for so-called “emotional” tasks.

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

Instead of challenging Nvidia in training, Huawei is positioning its latest Ascend AI processors as the tools of choice for Chinese teams that run “inference”, a computer designed by LLMs to produce a quick response.

A Chinese technology expert is betting that innovation will be a major source of future demand if the speed of model training slows and AI applications such as chatbots become more widespread.

“Training is important, but it only happens a few times,” said Georgios Zacharopoulos, a senior AI researcher working on process acceleration at Huawei’s Zurich lab. “Huawei is very focused on feedback, which will ultimately serve more customers.”

It’s focused on a less technically difficult but potentially more lucrative way to replace the AI ​​models trained on Nvidia products using Ascend chips, according to company employees and Ascend customers. Since Nvidia GPUs and Ascend run on different softwareHuawei helps companies use a different software to make the two systems compatible.

Huawei’s push comes with top-down government support. Chinese officials have encouraged local tech giants to buy some of Huawei’s AI chips and from Nvidia.

One person familiar with Nvidia’s operations in China said Huawei was seen internally as the country’s biggest competitor, adding that its chip design capabilities “have evolved first”.

Washington has sought to block Beijing’s AI development with foreign controls aimed at hindering the development of sensitive technologies in China.

Unlike their US rivals such as OpenAI and Google, the companies do not have access to the most advanced GPUs in China. But while Chinese teams are able to get low-end Nvidia H20 chips designed to meet foreign controls, less powerful GPUs are still in high demand as they are considered better than conventional ones. some local.

Critics and Huawei researchers he said Ascend wasn’t ready to replace Nvidia for prototype training because of technical issues, such as the breakdown of how the chips work together within the broader “cluster” of -AI chips when training larger models.

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

Another challenge for Huawei is convincing developers to move away from Nvidia’s Cuda software, known as the company’s “secret sauce” for being easy to use for developers and speeding up the process. too much data.

But Huawei’s recently released and improved version of its AI chip, the Ascend 910C, is also expected to solve these problems. “We expect this new generation of devices to come with improved software that makes developers available,” said a Huawei employee, who declined to be named.

Huawei and Nvidia are facing tough competition. Chinese Internet group Baidu and chip maker Cambricon have advanced the development of AI chips. Meanwhile, in the US, Amazon and Microsoft are also betting that they can gain more market share in chips for productivity as AI applications become more widespread.

Estimates from SemiAnalysis, a chip specialist, suggest that Nvidia made $12bn in sales in China last year by supplying 1mn of its H20 chips in the country, selling twice as many AI chips as Huawei with its Ascend 910B.

“Nvidia’s China-specific H20 GPUs make up the majority of AI chips sold in China. But the lead is shrinking rapidly as Huawei increases production capacity,” said Dylan Patel, chief analyst. in SemiAnalysis.

Industry insiders have warned that Huawei’s AI chip push is also being hampered by insufficient supply, with two prospective buyers telling the Financial Times that they are unable to secure chips.

Huawei did not respond to a request for comment. Nvidia declined to comment.

Analysts have said that Huawei’s production may face problems due to US export controls that have left Chinese manufacturers dependent on outdated chip-making equipment.

The focus also points to an ongoing shift in AI in China that differs from the US. Washington’s import controls mean that China’s AI players are not in the same race as Silicon Valley players Meta, Elon Musk’s X.AI and OpenAI to build big companies. -Nvidia’s most advanced GPUs.

“Chinese companies play a different game. They pay more attention to the idea than in the US because it is possible to get a big profit even with chips that are not very powerful, which means that they can achieve business quickly,” said Bernstein analyst Lin.

Chinese companies are betting that they can stay competitive in AI by lowering production costs, making it cheaper to run AI systems, he said.

Last month, Hangzhou and Beijing-based DeepSeek released its V3 version, which attracted attention due to its lower training and development costs compared to US comparison models.

The company has developed a new way for the AI ​​model to selectively focus on specific parts of the input data as a way to reduce the cost of running the model. It also uses the “Expert Blend” method favored by other Chinese AIs the beginningit also helps speed up feedback when only part of the model is used to generate the answer.

DeepSeek said that Huawei has successfully adapted V3 to Ascend, providing detailed instructions for developers on how to use the chip. The FT was there first report that Huawei has sent engineers to help customers from Nvidia to Ascend.

Additional reporting by Zijing Wu from Hong Kong