Huawei is seeking to gain a bigger share of China's artificial intelligence chip market, dominated by Nvidia, by helping local companies adopt its rival's chips for so-called "inference" tasks.

China's leading artificial intelligence company relies on graphics processing units (GPUs) made by Nvidia to "train" large language models, with the $3.4 trillion U.S. chipmaker's products seen as critical to developing the technology.

Rather than challenge Nvidia in training, Huawei is positioning its latest Ascend AI processor as the hardware of choice for Chinese groups running "inference," the calculations performed by LL.M.s to generate responses to prompts.

The Chinese tech giant is betting that inference will become a bigger source of demand in the future if the pace of model training slows and applications of artificial intelligence such as chatbots become more widespread.

"Training is important, but it only happens a few times," said Georgios Zacharopoulos, a senior artificial intelligence researcher in charge of inference acceleration at Huawei's Zurich lab. "Huawei mainly focuses on inference and will ultimately serve more customers."

The company is focusing on the less technically difficult but potentially lucrative path of adapting artificial intelligence models trained on Nvidia products to run on Ascend chips, according to employees and Ascend customers. Since Nvidia GPUs and Ascend run on different software, Huawei is helping enterprises use another software tool to make the two systems compatible.

Huawei's push has received top-down government support. Chinese officials are urging local tech giants to buy more of Huawei's artificial intelligence chips and ditch Nvidia's chips.

A person familiar with Nvidia's business in China said Huawei is viewed internally as the country's strongest competitor, adding that its chip design capabilities are "advanced."

Washington is trying to curb Beijing's artificial intelligence development through export controls aimed at hindering the development of China's sensitive technology.

Unlike U.S. rivals like OpenAI and Google, the company doesn't have access to China's most advanced GPUs. But while Chinese companies can only buy Nvidia's weaker H20 chips, which are customized to meet export controls, less powerful GPUs are still in high demand because they are considered better than local alternatives.

Analysts and Huawei researchers said Ascend is not ready to replace Nvidia for model training due to technical issues, such as glitches in the way chips interact within a broader "cluster" of AI chips when training larger models.

"Although the single-chip performance of the Ascend chip is good, there is a bottleneck in the inter-chip connection," said Lin Qingyuan, a China semiconductor analyst at Bernstein. "When training a large model, it must be broken down into smaller tasks. If one chip fails, The software needs to find a way to let other chips take over immediately."

Another challenge Huawei faces is convincing developers to abandon Nvidia's Cuda software, which has been hailed as the company's "secret sauce" for being easy for developers to use and capable of dramatically speeding up data processing.

However, Huawei’s upcoming AI chip upgrade Ascend 910C is also expected to solve these problems. "We expect that the new generation of hardware will come with improved software to make it easier for developers to use it," said a Huawei employee who spoke on condition of anonymity.

Huawei and Nvidia face fierce competition. Chinese internet group Baidu and chip design company Cambrian have made strides in developing artificial intelligence chips. Meanwhile, in the United States, Amazon and Microsoft are also betting that they can grab more market share in inference chips as artificial intelligence applications become more widespread.

Estimates from chip consulting firm SemiAnalysis show that Nvidia delivered 1 million H20 chips in China last year, with sales in China reaching $12 billion. Its Ascend 910B AI chip sales were twice that of Huawei.

"Nvidia's China-specific H20 GPU accounts for the majority of AI chips sold in China. But as Huawei increases its manufacturing capabilities, the lead is shrinking rapidly." said Dylan Patel, principal analyst at SemiAnalysis.

Industry insiders warned that Huawei's AI chip rollout was also hampered by a lack of supply, with two potential customers telling the Financial Times they were unable to obtain the chips.

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

Analysts say Huawei's manufacturing could face challenges as U.S. export controls cause Chinese factories to rely on outdated chipmaking equipment.

The focus on inference also points to a different dynamic of evolving AI in China than in the United States. Washington's export controls mean Chinese AI companies cannot compete in the same race as Silicon Valley rivals Meta, Elon Musk's x.AI and OpenAI to build large clusters of Nvidia's most advanced GPUs.

"Chinese companies are playing a different game. They focus more on inference than the United States because even using less powerful chips can greatly improve efficiency, which also means they can achieve commercialization faster." Bernstein analyst Lin explain.

He said Chinese companies are betting they can keep AI competitive by lowering the cost of inference, thereby lowering the cost of running AI applications.

Last month, Hangzhou and Beijing-based startup DeepSeek released its V3 model, which has attracted attention for its lower training and inference costs compared with comparable models in the United States.

The company has proposed a new way to let AI models selectively focus on specific parts of input data as a way to reduce the cost of running the models. It also uses "expert blending" technology popular with other Chinese AI startups, which also helps speed up inference because only part of the model is used to generate responses.

DeepSeek said Huawei has successfully adapted V3 to Ascend, providing developers with detailed instructions on how to use the chip. The Financial Times previously reported that Huawei has sent engineers to help customers migrate from Nvidia to Ascend.

Additional reporting by Wu Zijing from Hong Kong