Huawei is trying to grab a bigger share of China’s artificial intelligence chip market, dominated by Nvidia, by encouraging local companies to adopt rival processors for “inference” tasks.
China’s leading AI companies rely on Nvidia-made graphics processing units (GPUs) to “train” large-scale language models, making the $3.4 trillion US chipmaker’s products essential to developing the technology. It is considered that
Rather than challenge Nvidia in training, Huawei is using its latest Ascend AI processors as the hardware of choice for Chinese groups to perform “inference,” the calculations that LLMs make to generate responses to prompts. It is positioned.
Chinese tech giants are betting that inference will become a bigger source of demand in the future as the pace of model training slows and AI applications such as chatbots become more popular.
“Training is important, but you can only do it a few times,” said Georgios Zacharopoulos, a senior AI researcher who works on accelerating inference at Huawei’s Zurich research lab. “Huawei is mainly focused on inference, which will ultimately allow us to serve more customers.”
According to company employees and Ascend customers, this is a technically challenging but potentially lucrative opportunity to take AI models trained on Nvidia products and refine them to run on Ascend chips. It is said that the focus is on the path of sexuality. Because Nvidia GPUs and Ascend run on different software, Huawei is helping companies use different software tools to make the two systems compatible.
Huawei’s promotion comes with top-down support from the government. Chinese authorities are urging local tech giants to buy more Huawei AI chips and divest from Nvidia.
A person familiar with Nvidia’s operations in China said Huawei is considered its most serious competitor in the country, adding that the company’s chip design capabilities are “advanced.”
The U.S. government is trying to curb Beijing’s AI development through export controls aimed at preventing the development of sensitive technologies in China.
Unlike U.S. rivals like OpenAI and Google, companies don’t have access to cutting-edge GPUs in China. But even though the Chinese group only has access to Nvidia’s less powerful H20 chip, which is tailored to meet export regulations, the GPU is considered superior to local alternatives. As such, it remains in high demand.
Analysts and Huawei researchers say technical issues such as breakdowns in the way chips interact with each other within broader AI chip “clusters” are a challenge in training ever-larger models. Because of the issues, Ascend said it is not yet ready to replace Nvidia in model training.
“Ascend’s chips have good per-chip performance, but there are bottlenecks in the connections between chips,” said Lin Qingyuan, China semiconductor analyst at Bernstein. “When you train a large model, you have to break it down into smaller tasks. If one chip fails, the software has to figure out how the other chips can take over without delay.”
Another challenge for Huawei is convincing developers to switch from Nvidia’s Cuda software, known as the company’s “secret sauce” for its ease of use and significantly faster data processing.
However, Huawei’s soon-to-be-released updated version of its AI chip, the Ascend 910C, is also expected to address these concerns. “We expect this new generation of hardware to come with improved software that will be more accessible to developers,” a Huawei employee said on condition of anonymity.
Huawei and Nvidia face fierce competition. Chinese internet group Baidu and chip designer Cambricon have made progress in developing AI chips. Meanwhile, in the U.S., Amazon and Microsoft are also betting that they can gain more market share for inference chips as AI applications become more widespread.
Last year, Nvidia delivered 1 million H20 chips in China and sold twice as many AI chips with Huawei’s Ascend 910B, generating $12 billion in sales in China, according to estimates from chip consulting firm Semianalysis. He says he earned it.
“Nvidia’s China-specific H20 GPUs account for the majority of AI chips sold in China, but the gap is rapidly closing as Huawei ramps up manufacturing capacity,” said lead semi-analyst. Analyst Dylan Patel said.
Industry insiders warned that Huawei’s AI chip push is also being constrained by supply shortages, with two prospective customers telling the Financial Times they were unable to secure chips.
Huawei did not respond to requests for comment. Nvidia declined to comment.
Analysts say Huawei’s manufacturing is likely to face difficulties as U.S. export restrictions leave Chinese factories relying on outdated chip-making equipment.
The focus on inference also illustrates the evolving dynamics of AI in China, which is different from the US. Washington’s export controls mean Chinese AI companies won’t be in the same race as Silicon Valley rival Meta, Elon Musk’s x.AI and OpenAI to build massive megaclusters of Nvidia’s cutting-edge GPUs. means.
Recommended
“Chinese companies are taking a different strategy,” said Lin, the Bernstein analyst. “They’re paying much more attention to inference than the U.S., because they can significantly improve efficiency even with less powerful chips. It also means that commercialization can be achieved faster.”
He said Chinese companies are betting they can remain competitive in AI by lowering the cost of inference, which in turn will make AI applications cheaper to run.
Last month, Hangzhou and Beijing-based startup DeepSeek released its V3 model, which attracted attention for its lower training and inference costs compared to comparable US models.
The company proposed a new way for AI models to selectively focus on specific parts of input data as a way to reduce model execution costs. We also used an “expert mixing” approach popular with other Chinese AI startups. This also helps speed up inference since only part of the model is used to generate the response.
DeepSeek said Huawei successfully adapted V3 to Ascend and provided developers with detailed instructions on how to use the chip. The FT previously reported that Huawei sent engineers to help customers migrate from Nvidia to Ascend.
Additional reporting by Zijing Wu in Hong Kong
