Recently, during his visit to Beijing, NVIDIA CEO Jensen Huang announced that the US government has approved the resumption of H20 chip exports to China.

The H20 chip is an AI chip designed by NVIDIA specifically for the Chinese market in accordance with the US technology export control requirements to China. Its performance is far inferior to that of the mainstream GPU chip H100 in the international market. In April this year, the United States banned the sale of this chip to some countries, including China, on the grounds of national security. As soon as the ban was issued, Nvidia faced a $4.5 billion inventory loss due to the unsold H20 and was forced to make an impairment provision of $5.5 billion, with its market value evaporating by $160 billion at one point.

Why has the United States resumed the sale of H20 chips to China? What impact will this policy adjustment have on China's artificial intelligence industry? What should Chinese chip enterprises do in the future? For this, our reporter interviewed Zhang Xin, the director of the Digital Economy and Legal Innovation Research Center of the University of International Business and Economics.

Why lift the ban

Weigh the pros and cons and retreat to advance

Reporter: Against the backdrop of the current global artificial intelligence race and the struggle for technological sovereignty, what strategic considerations does the adjustment of the US export policy on H20 chips and others reflect? Does this mean that the US strategy of containing China's technology has changed?

Zhang Xin: The "relaxation" of the US export policy on H20 chips and others is not a shift in strategic goals, but rather a phased adjustment of strategic tools. Its deep-seated intention has remained unchanged: to gain the upper hand in the long-term game between China and the United States and ensure the United States' global leading position in key technological fields such as artificial intelligence.

Under this framework, the chip sanctions are regarded by Washington as a "trump card" that can be played at any time. Although the marginal utility of this card is decreasing, playing it at this time, retreating to advance, can bring more realistic strategic gains in the short term.

Reporter: Why did the United States choose to lift the ban at this point in time? Is the underlying reason driven by market interests or the combined effect of multiple factors? How is the significance of the Chinese market to American chip companies like NVIDIA reflected in this process?

Zhang Xin: The core of this choice lies in weighing the costs and benefits of continuing the ban: on the one hand, the cost of maintaining the ban has far exceeded its strategic returns; On the other hand, releasing this bargaining chip in specific negotiation scenarios can instead bring the US more attractive phased benefits. Specifically, the United States initially restricted the export of high-end chips to China on the grounds of "national security", aiming to curb the rise of China's technology. However, for "exclusive edition" chips like the H20, the rationality of continuing the blockade is being weakened by several realities.

First, the assumption of military use lacks an empirical basis. Some overseas experts have analyzed that the Chinese military will proactively avoid US technology due to risk considerations, and China has sufficient computing power reserves of its own and does not need to rely on US chips.

Second, the rise of domestically produced chips in China has further weakened the significance of the ban. The performance of several domestic artificial intelligence chips has surpassed that of the H20 in some fields. Huawei's paper shows that the measured efficiency of the new generation 910C has surpassed that of NVIDIA's H100. Continuing to block the H20 can no longer effectively slow down the development of AI in China.

More importantly, maintaining the ban has instead cost American enterprises a heavy price. Nvidia faced a $4.5 billion inventory loss due to the sluggish sales of its H20 products and was forced to make an impairment provision of $5.5 billion, with its market value evaporating by $160 billion at one point. Huang Rengxun warned that if the United States loses China's AI chip market, which is expected to reach 50 billion US dollars in the coming years, it will face a triple loss of "revenue, tax revenue and employment".

In addition, American tech giants such as Oracle and Google have made large-scale purchases and deployments of GB200. The leading position of the United States in AI infrastructure has not been shaken by the export of H20.

After weighing the pros and cons, deregulation is more in line with the current interests of the US side.

Huang Rengxun made three trips between China and the United States within half a year. He not only promised the Trump administration to "strengthen American AI leadership", but also made a high-profile declaration of long-term cooperation in China. His efforts to facilitate policy shifts have been accelerated. The huge market size of China determines that any "decoupling" will backfire on the United States itself. As analysts from research firm Gartner put it, "Apart from China, there is no other market in the world that can absorb such a large inventory of chips."