China’s artificial intelligence developers are expected to welcome Washington’s go-ahead for Nvidia to ship its H200 AI chips to approved customers on the mainland, according to analysts, even as Beijing continues to drive the country’s tech self-sufficiency efforts.
In a post on his Truth Social platform on Monday, US President Donald Trump said he had informed his Chinese counterpart, Xi Jinping, of that decision “under conditions that allow for continued strong national security”. Xi had “responded positively,” Trump added.
A strong demand exists for the H200 from China’s AI developers because the Nvidia graphics processing unit (GPU) still outperformed domestic AI chips, especially in terms of computing power and memory bandwidth on a single-card basis, according to semiconductor industry analyst Zhang Haijun.
For many Chinese developers, Zhang said a large part of their code was based on Nvidia’s Hopper GPU microarchitecture, which was purpose built for AI and high-performance computing.
That means the H200 chips could be used immediately by Chinese developers without rewriting their code, “which is what these Big Tech companies like the most”, Zhang said. He added that the impact of H200 sales would be “minimal” on domestic GPUs, which are mostly used in inference tasks as opposed to the more demanding training tasks.
The H200 was introduced by Nvidia in 2023 after the Biden administration’s roll-out of rigid tech export controls in 2022, which restricted the company’s sale of advanced H100 GPUs to China for national security reasons.
While approval of the H200 shipments to China marked a successful lobbying effort in Washington by Nvidia CEO Jensen Huang, the US chip designer now faced stronger domestic competition from the likes of Huawei Technologies, Cambricon Technologies, Moore Threads Technology and MetaX Integrated Circuits.
The Shanghai-listed shares of Moore Threads, dubbed as China’s little Nvidia, gained 5.7 per cent to close at 628.31 yuan on Tuesday, while those of Cambricon were up 1.28 per cent to 1,436 yuan.
“US chip exports are unlikely to hinder the long-term goal of localisation,” analysts at mainland brokerage firm Guotai Haitong said in a research note on Tuesday. “The H200 access is more about increasing China’s total computing power supply than about replacing domestic products.”
The US relaxation would not deter China’s efforts to push forward the greater adoption of locally developed chips, the analysts said.
Still, the H200 chip’s availability in China was expected to boost capital expenditure by the country’s cloud computing services providers, including Alibaba Cloud, Tencent Holdings and ByteDance-run Volcano Engine, according to the Guotai Haitong note. Alibaba Cloud is the AI and cloud computing unit of Alibaba Group Holding, owner of the South China Morning Post.
Nvidia’s supply of H200 chips to China was expected to boost the speed and scale of China’s AI development, according to Nigel Green, CEO of global financial advisory and asset management company deVere Group.
“Access to chips of this class directly affects how quickly firms can build and refine advanced AI systems,” Green said.
China, however, remained concerned over the stability of US export control policies, according to Liu Jie, an engineer at a Shanghai-based GPU developer. For those who do not need top chip performance, these developers are expected to be reluctant to switch to US GPUs, he said.
Alibaba Cloud recently introduced a computing pooling solution that it said led to an 82 per cent reduction in the number of Nvidia GPUs needed to serve the firm’s AI models.