Under strict control by the United States, how can China obtain Nvidia chips?

(Deutsche Welle Chinese website) NVIDIA CEO Jen-Hsun Huang admitted that he had an extramarital affair and the incident is still unfolding. The alleged target of his affair was a Chinese female executive who worked for Megaspeed, a company headquartered in Singapore, and the United States suspected that this company was involved in smuggling artificial intelligence chips into China.

This scandal has once again focused the attention of the outside world on a question: How did China, one of the world's artificial intelligence powers, obtain the most advanced chips under U.S. export restrictions?

Why are Nvidia chips so important?

NVIDIA designed an advanced component called a graphics processing unit (GPU) originally used to process three-dimensional graphics in video games.

This kind of chip has the ability to process massive amounts of data, so it is also used on a large scale to train and run AI systems.

The launch of ChatGPT in 2022 demonstrated the huge potential of artificial intelligence and triggered an investment gold rush. Governments and technology giants from various countries are competing to promote the development of artificial intelligence technology.

As chip orders continue to pour in, Nvidia's market value has soared to US$4.5 trillion, and it is currently tied with Apple (Apple) as the two most valuable companies in the world.

What chip exports does the United States restrict?

Due to national security concerns, the United States has banned the export of Nvidia's most advanced "Blackwell" and "Rubin" series of GPUs to China because advanced artificial intelligence technology has military applications.

This policy also helps the United States’ top artificial intelligence laboratories maintain their global leadership.

Jen-Hsun Huang has been lobbying the U.S. government to allow the sale of Nvidia chips to China. He warned that the restrictions were driving the development of local alternatives in China.

This year, the United States relaxed restrictions on the "H200" chip, which was previously banned from export and has relatively low performance, and stipulated that the U.S. government can receive a 25% share of sales.

However, the approval conditions set by the United States make it difficult to obtain relevant export licenses. At the same time, with the development of China's local chip industry, Beijing is not active in purchasing H200 chips.

Reuters previously quoted two people familiar with the matter as saying that approximately 600,000 NVIDIA advanced artificial intelligence chips are currently stored in Hong Kong, waiting to be shipped to mainland China. Despite demand from Chinese technology companies, Beijing has yet to approve large-scale imports of these chips.

How does China obtain these chips?

Chinese artificial intelligence start-ups including Moonshot AI (Dark Side of the Moon), Z.ai (Zhipu AI) and DeepSeek (Deep Search) have recently launched high-performance models, further intensifying competition with their American counterparts.

Lian Jye Su, chief analyst at market research firm Omdia, told AFP: "Despite export controls, Chinese companies are still using the most advanced Nvidia chips to train models."

These chips may be obtained by Chinese companies in two ways: one is smuggling. Technology media The Information reported in December that DeepSeek used thousands of chips that were first dismantled overseas and then shipped to China when training its new model.

DeepSeek did not respond to a request for comment at the time. Nvidia told the outlet it had seen no evidence of such smuggling and said the claim "did not appear credible."

The second is to take advantage of the loopholes in the cloud computing system, that is, to rent computing power equipped with NVIDIA chips in data centers in third countries, which are usually located in Southeast Asia.

Stephen Wu, founder of investment fund Carthage Capital, explained: "U.S. export controls restrict where the chip is shipped, not who is using it."

This means that Nvidia's GPUs can be installed in data centers in Johor, Malaysia, for example, and comply with export regulations, while customers in Shenzhen can log in remotely and use the computing power of these chips.

White House adviser Michael Kratsios said this summer that artificial intelligence company Moonshot AI had obtained high-end Nvidia chip resources through Thailand.

The United States also accuses some Chinese artificial intelligence labs of using American models to train their own systems to imitate their capabilities.

What is the current level of China's domestic chips?

Poe Zhao, founder of Hello China Tech, a technology analysis media, said: "Nvidia chips can still make top-level AI research and development work faster and easier to complete, but it is no longer the only option."

The problem, he noted, is cost. Training models using Chinese-made chips usually takes more time and requires more engineering optimization work.

Huawei is currently China’s main Nvidia competitor. Although Chinese-made chips are improving rapidly, there is still a gap compared to the American giant.

According to reports, DeepSeek’s next generation of important artificial intelligence models will use Huawei chips for training.

Huawei announced in September this year that its next-generation artificial intelligence chip will be available in the first quarter of 2027, three quarters earlier than originally planned.

Stephen Wu believes that to improve the overall performance of domestic chips, artificial intelligence companies need to deploy a larger number of chips to make up for the performance gap of a single chip.

"The price they pay is higher power consumption and higher overall costs, but they can afford it because electricity prices are lower in China and the chips receive government subsidies."