Chinese AI chips gaining market traction

In the wake of the United States' export controls on advanced chips, China's push for self-sufficiency in artificial intelligence processors is gaining remarkable momentum, with at least nine domestic chip companies having surpassed a significant threshold of 10,000 units in terms of order shipments, according to experts and executives.

This 10,000-chip club signals that domestic AI chips are gaining tangible market traction based on their performance, stability and cost. It heralds a shift from mere scale competition to a more comprehensive battle encompassing software ecosystems, commercial services and sustained reliability, experts said.

The cohort includes the in-house chip divisions of tech heavyweights such as Huawei's Ascend, Baidu's Kunlun, and Alibaba's T-head, alongside listed or soon-to-be-listed AI chip specialists such as Cambricon, Moore Threads, Enflame, and Iluvatar CoreX. Even some startups, including Sunrise and Tsingmicro, have crossed this volume threshold.

Companies with the largest scale of shipments, such as Huawei, Baidu and Alibaba, have reached several hundred thousand.

Sources close to Alibaba told China Daily that "several hundred thousand" units of T-head's Zhenwu PPU chips have been cumulatively shipped. These have powered multiple clusters of 10,000 AI chips within Alibaba Cloud, serving more than 400 clients, including the State Grid Corp of China, the Chinese Academy of Sciences, XPeng Motors, and Sina Weibo.

He Hui, semiconductor research director at the United Kingdom-based tech research firm Omdia, said, "Alibaba's T-head is at the forefront in terms of AI chip shipments, benefiting from its early start and massive internal demand."

According to US market research company International Data Corp, in the first half of 2025, the size of China's sever market powered by AI chips reached $16 billion, with over 1.9 million units shipped. US chipmaker Nvidia held about 62 percent of the market share, while Chinese chips captured about 35 percent.

IDC ranked Huawei's Ascend series first in market share among domestic chips in the first half of 2025. Backed by its parent company's vast ecosystem, Ascend enjoys stable demand. Ascend chips are used in large-scale clusters by telecom operators and tech firms.

The comments came amid China's compliance and security concerns over Nvidia's AI chips. Although some media organizations, including Reuters, reported on Friday that China's regulators have approved several Chinese companies, such as DeepSeek and ByteDance, to buy Nvidia's H200 chips, the decision has not yet been officially announced.

He, from Omdia, said that Chinese internet companies are adopting a two-way approach. On the one hand, they are making efforts to buy Nvidia's advanced chips within regulatory frameworks. On the other hand, they are supporting the development of domestic AI processors as much as possible. For instance, ByteDance is testing and using chips from Baidu's Kunlun and Cambricon, while Tencent is using processors from Enflame.

"Nvidia's H200 chips are believed to accelerate the development of China's AI large language models, but we don't know whether the US will restrict exports in the future," He said, adding that it makes sense to have other alternatives.

The growing demand for Chinese AI chips is also reflected in the documents filed by listed companies. Prospectuses submitted between late 2025 and early 2026 by companies such as Moore Threads, Iluvatar CoreX, and Enflame show accumulated shipments surpassing 10,000 chips. Iluvatar CoreX, for example, reported delivering 52,000 AI chips to 290 clients across finance, healthcare and transportation sectors by June 30.

Industry insiders said they view the emergence of multiple 10,000-chip vendors as a critical transition. Roger Sheng, vice-president of research at US market research company Gartner, said it marks the entry into a "large-scale verification" phase for this round of industrial trial and error, as Chinese AI chip companies make progress.

But challenges exist. Sheng outlined critical areas for improvement, including adopting advanced packaging technologies and optimizing computer algorithms through cooperation with Chinese large language model providers.

Xiang Ligang, director-general of the Zhongguancun Modern Information Consumer Application Industry Technology Alliance, a telecom industry association, said that Nvidia's dominance in the global AI chip market relies not only on chip performance, but also on its entrenched CUDA ecosystem — a 20-year-old framework for AI compatibility.

While Chinese chips, such as Huawei's Ascend series, can technically rival certain Nvidia chips, they lack CUDA's universal adaptability for AI large language models, he said.

"However, geopolitical uncertainty is driving users toward Chinese chips, and developers are now shifting their focus to compatibility layers for domestic alternatives, eroding Nvidia's ecosystem advantage," Xiang added.

He, from Omdia, said that with domestic manufacturing capacity for AI chips expected to be ramped up in 2026, the industry anticipates "another wave of explosive growth".

Zhang Wen, founder, chairman and CEO of Chinese AI chipmaker Beijing Biren Technology Development Co, said, "The production capacity of advanced domestic AI chips will be crucial for the future of China's AI industry."