Huawei is trying to capture a larger share of China’s artificial intelligence chip market, dominated by Nvidia, by helping local companies take on their rival chips for so-called “inference” tasks.
Leading AI companies in China are relying on Nvidia graphics processing units (GPUs) to “train” large language models, with the $3.4 trillion US chipmaker’s products seen as crucial to the technology’s development.
Rather than challenging Nvidia in training, Huawei is positioning its latest Ascend AI processors as the hardware of choice for Chinese groups running “inference,” the computation performed by LLMs to generate an answer to a prompt.
The Chinese tech giant expects inference to be a larger source of demand in the future as the pace 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 inference acceleration at Huawei’s Zurich lab. “Huawei mainly focuses on inference, which will ultimately serve more customers.”
The focus is on the less technically demanding but potentially lucrative route of retrofitting AI models trained on Nvidia products to run on Ascend chips, according to company officials and Ascend customers. Nvidia GPUs and Ascend continue to run different softwareHuawei helps companies use another software tool to make the two systems compatible.
Huawei’s move is supported by the government from above. Chinese officials have urged local tech giants to buy more AI chips from Huawei and move away from Nvidia.
A person familiar with Nvidia’s operations in China said Huawei is viewed internally as the country’s most serious competitor, adding that its chip design capabilities are “advanced.”
Washington has sought to curb Beijing’s development of AI through export controls aimed at hampering the development of sensitive technologies in China.
Unlike their US competitors such as OpenAI and Google, companies in China do not have access to the most advanced GPUs. But even though Chinese companies can only purchase Nvidia’s less powerful H20 chips, which are specifically tailored to export controls, the less powerful GPUs remain in high demand because they are considered better than local alternatives.
analysts and Huawei researchers said Ascend was not yet ready to replace Nvidia for model training due to technical issues, such as a disruption in the way the chips within a larger “cluster” of AI chips interact with each other when training ever larger models.
“While the Ascend chips perform well on a per-chip basis, there is a bottleneck in connectivity between chips,” said Lin Qingyuan, Bernstein’s semiconductor analyst for China. “When you train a large model, you need 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.”
The other challenge for Huawei is convincing developers to switch from Nvidia’s Cuda software, which is known as the company’s “secret sauce” because it is easy for developers to use and can significantly speed up data processing.
But Huawei’s soon-to-be-released updated version of its AI chip, the Ascend 910C, also aims to address these concerns. “We expect this new generation of hardware to come with improved software that will make it more accessible to developers,” said a Huawei official who declined to be named.
Huawei and Nvidia face tough competition. The Chinese internet company Baidu and the chip designer Cambricon have made progress in the development of AI chips. Meanwhile, Amazon and Microsoft in the US are also betting that as AI applications become more widespread, they can capture more market share in chips for inference purposes.
Estimates from SemiAnalysis, a chip consulting firm, suggest Nvidia made $12 billion from sales in China last year by shipping 1 million of its H20 chips to the country, selling twice as many AI chips like Huawei with its Ascend 910B.
“Nvidia’s China-specific H20 GPUs account for the majority of AI chips sold in China. But the lead is shrinking quickly as Huawei increases production capacity,” said Dylan Patel, principal analyst at SemiAnalysis.
Industry insiders warned that Huawei’s AI chip push was also limited by insufficient supply. Two potential customers told the Financial Times that they were unable to secure the chips.
Huawei did not respond to a request for comment. Nvidia declined to comment.
Analysts say Huawei’s manufacturing is likely to face challenges because of U.S. export controls that have left Chinese factories relying on aging chip-making equipment.
The focus on inference also points to an evolving dynamic in Chinese AI that differs from that in the United States. Because of Washington’s export controls, Chinese AI players are not in the same race as Silicon Valley rivals Meta, Elon Musk’s x.AI and OpenAI to build large mega-clusters of Nvidia’s most advanced GPUs.
“Chinese companies play a different game. “They put a lot more emphasis on conclusions than the US because it is possible to achieve large efficiency gains even with less powerful chips, which also means they can reach commercialization more quickly,” said Bernstein analyst Lin.
Chinese companies are betting that they can remain competitive in AI by reducing inference costs, which in turn makes running AI applications cheaper, he said.
Last month, Hangzhou and Beijing-based startup DeepSeek released its V3 model, which attracted attention for its low training and inference costs compared to comparable models in the United States.
The company proposed a new way for an AI model to selectively focus on specific parts of the input data to reduce the cost of running the model. It also used the “expert blend” technique popular with other Chinese AIs Startupswhich also helps speed up inference since only part of the model is used to generate an answer.
DeepSeek said Huawei has successfully adapted V3 to Ascend and provided detailed instructions for developers on how to use the chip. The FT has previously reported that Huawei had sent engineers to help customers migrate from Nvidia to Ascend.
Additional reporting by Zijing Wu in Hong Kong
