The United States conditionally allows Nvidia chips: A calculated geopolitical technology game.
14/01/2026
Washington's decisions often reveal their hand at the last minute. On January 13, the Trump administration officially gave the green light for NVIDIA to export its second most powerful AI chip, the H200, to China, accompanied by a series of unprecedented strict conditions. This approval document is less of a permit and more of an agreement filled with probing and calculation. On one side, there are pending orders for over 2 million chips, with a total value of tens of billions of dollars, while on the other side, there is fierce opposition from China hawks within Washington and deep concerns about technology leakage. This game of exporting cutting-edge artificial intelligence computing power has suddenly been pushed from behind the scenes to the forefront, reflecting the distortion and reshaping of the global technology supply chain under political divisions.
A meticulously planned "conditional release"
The approval of the Trump administration was by no means an unconditional open door. The new regulations constructed a complex filter, attempting to strike a delicate balance between commercial interests and national security.
Technical Review and Quantity Cap: The Illusion and Reality of Dual Safeguards
According to the new regulations, each batch of H200 chips planned for shipment to China must first undergo technical capability testing at a third-party laboratory. Only after confirming that their AI performance complies with the regulations can an export license be obtained. This is equivalent to adding an external security check after the chips leave the factory. The more critical restriction lies in quantity: the total number of H200 chips obtained by Chinese customers must not exceed 50% of the total number sold by NVIDIA to U.S. customers during the same period. This is a dynamic quota system, intended to permanently anchor China's acquisition volume below half of the U.S.'s own consumption, theoretically ensuring the U.S.'s relative advantage in advanced computing power.
However, analysis suggests that this seemingly stringent quota may have limited effectiveness in reality. Jay Goldberg, a stock analyst at Seaport Research, pointed out incisively that this appears more like a gesture of compromise, with considerable difficulty in enforcement. As we have seen, (Chinese) companies always find ways to acquire these chips, and the U.S. government has shown a highly 'transactional' approach regarding chip exports to China. He bluntly analogized, in other words, this looks like a Band-Aid, attempting to temporarily cover up the significant disagreements among U.S. government export policy makers.
Safety Commitments and Domestic Stockpiles: A Trust Game Difficult to Verify
The new regulations also require NVIDIA to demonstrate that it maintains sufficient H200 inventory within the United States to prioritize meeting domestic demand. Meanwhile, Chinese clients must prove they have adequate security procedures in place and commit to not using the chips for military purposes. These conditions, which had never been explicitly established before, are now being introduced as key leverage by the team led by David Sacks, the White House's Director of Artificial Intelligence Affairs.
The logic of the Trump administration carries a distinct transactional tone. They believe that selling advanced AI chips to China could actually dampen the motivation of heavily sanctioned Chinese competitors like Huawei to double down on efforts to catch up with the most advanced chip designs from NVIDIA and AMD. When announcing in December last year that sales would be permitted, Trump claimed it would be done under conditions that allow for the continued maintenance of strong national security. However, doubts have never dissipated regarding whether the administration would actually enforce these restrictions in practice, or even whether Beijing would allow these chips to be sold domestically.
China's Cautious Response: A Proactive "Cooling Down"
Just as the ink on Washington's approval documents had barely dried, news from Beijing added a new dimension to this game. According to The Information, the Chinese government has informed some tech companies this week that it will only approve their purchase of NVIDIA H200 chips under special circumstances, such as for university research.
This instruction, described as deliberately vague, requires companies to procure only when necessary, but does not specify the exact meaning of necessity. The previous week, reports had already indicated that China requested some companies to suspend H200 orders, aiming to prioritize the development of domestic enterprises in the AI competition. The Chinese government plans to hold meetings with more companies to convey this procurement guidance, although whether more detailed rules will be issued remains unknown.
Beijing's cautious stance sends multiple signals. Firstly, it reflects an inevitable choice under China's strategy of promoting self-reliance and self-strengthening in science and technology. In this critical sector, over-reliance on a single foreign supplier (even a leading performer like NVIDIA) is viewed as a strategic risk. Secondly, it also serves as an indirect response and countermeasure to U.S. restrictions, indicating that the Chinese market is not unconditionally open. NVIDIA finds itself caught in a double squeeze between Washington and Beijing: the U.S. is weighing stricter export controls on its most advanced technologies, while China is intensifying efforts to strengthen domestic capabilities and urging local companies to reduce dependence on foreign technology.
Supply-Demand Imbalance: The Thirst for Computing Power and Strategic Anxiety
The underlying driving force of this game is the intense conflict between massive market demand and geopolitical constraints.
The Computing Power Competition Behind Tens of Thousands of Orders
A Reuters report last month revealed a staggering figure: Chinese tech companies have placed orders for over 10,000 chips. With each chip priced at approximately $10,000, the potential transaction value exceeds $100 million. In contrast, Nvidia's inventory at the time was only 1,000 chips. At the International Consumer Electronics Show in Las Vegas, Nvidia's Jensen Huang admitted that the company is ramping up chip production to meet strong demand from China and other regions worldwide. This demand has even driven up the rental prices of existing chips in cloud data centers.
Such a massive volume of orders underscores China's urgent demand for top-tier computing power in the fields of AI large model training and inference. Saif Khan, former Director of Technology and National Security at the White House National Security Council under the Biden administration, warned that the new regulations would substantially enhance China's AI capabilities. These rules would permit the export of approximately 2 million advanced AI chips like the H200 to China, a quantity equivalent to the total computing power currently possessed by a typical leading-edge AI company in the United States. His concern hits the core issue: computing power is power; in the AI era, the scale of chips directly translates into the competitiveness of nations and enterprises.
Implementation Dilemmas and the Prelude to a "Cat-and-Mouse Game"
Under the new regulations, the Know Your Customer requirements aim to restrict Chinese cloud service providers from supporting malicious purposes. However, Han pointed out that the government will face significant challenges in implementation. How can the final flow of chips be monitored? How can it be ensured that they are not used for military or large-scale surveillance projects? The answers to these questions remain unclear. Historical experience shows that driven by high profits and high demand, underground channels and technical workarounds to circumvent controls will always emerge.
There is also significant internal division within the U.S. government. China hawks, crossing party lines, unanimously criticize this move as a major boost to Beijing's military capabilities and an erosion of America's advantage in the field of artificial intelligence. This internal tension renders the policy inherently fragile and uncertain, with the potential to tighten again at any moment due to shifting political winds.
The Restructuring and Future Direction of Global Industrial Chains
The export controversy surrounding NVIDIA's H200 chip is by no means an isolated incident. It epitomizes the profound fission occurring within the global artificial intelligence industry chain under the impact of technological nationalism.
The United States attempts to maintain its lead in key areas through a "small yard, high fence" strategy. Conditionally granting approval for the H200 can be seen as a flexible adjustment within this "small yard" strategy—not a complete blockade, but rather setting up a valve to try to control the pace and scale of technology diffusion, while allowing American companies (NVIDIA) to profit from the vast Chinese market, thereby sustaining their R&D investment and leading position. This is a precise calculation: aiming both to make money and to fortify its moat.
China is accelerating the advancement of domestically controlled alternative solutions. Beijing's restrictions on purchasing H200 align with its policies to vigorously support local AI chip companies such as Huawei Ascend and Cambricon. The partial retention of market space provides a valuable growth window and time for domestic chips. China's goal is clear: to reduce vulnerability dependence on external supply chains and build an AI computing power foundation primarily based on internal circulation.
Global technology companies caught in the middle, such as NVIDIA, must learn to navigate an increasingly complex compliance maze. They may be forced to develop specially tailored chips with performance slightly below the top tier, designed specifically to meet export control regulations in different regions, leading to fragmentation of global product lines. This increases the costs and uncertainties within the supply chain.
The game surrounding the H200 is far from over. It has ushered in a new phase: the trade of cutting-edge technology will no longer be merely a commercial activity, but a highly politicized process intertwined with continuous scrutiny, dynamic quotas, security checks, and ambiguous political conditions. The conditional clearance by the United States and the restrictive procurement by China together outline a new normal for future global high-tech competition—where cooperation coexists with containment, and dependence intertwines with decoupling.
Ultimately, the flow of computing power will not come to a complete halt, but its path will become more winding and costly. The true winners, perhaps, will be those countries and enterprises that can most swiftly adapt to this new normal and strike the optimal balance between independent innovation and open cooperation. The journey of the chip has only just begun.