China Approves Import of NVIDIA Chips: Balancing Technological Needs and Independent Strategy
29/01/2026
On January 23, Beijing approved the first batch of import licenses for NVIDIA's H200 artificial intelligence chips. This decision was made during NVIDIA CEO Jensen Huang's visit to Shanghai, Beijing, and Shenzhen this week. The procurement rights for the initial batch of several hundred thousand chips were primarily granted to leading internet companies such as Alibaba, Tencent, and ByteDance. This move signifies a critical tactical adjustment in China's regulatory stance within the complex chess game of Sino-U.S. technological competition—between adhering to the long-term national policy of semiconductor self-reliance and meeting the immediate computing power demands of the AI industry, Beijing has chosen a more pragmatic middle path.
A meticulously arranged "clearance" and Jensen Huang's low-profile itinerary.
The approval did not come out of the blue. According to multiple sources from Reuters and The Wall Street Journal, the entire decision-making process was accompanied by subtle signals and cautious interactions. Jensen Huang arrived in Shanghai on January 19, ostensibly to attend the annual celebration of NVIDIA's Chinese subsidiary. Unlike previous years, the trip of this chip giant's leader was exceptionally low-key. Media captured images of him strolling through a vegetable market and distributing auspicious kumquats to employees, without meeting high-level officials such as Chinese Vice Premier He Lihua as he did last year. This posture of focusing solely on business rather than politics created a relatively straightforward environment for the negotiation of technical licenses.
The approval occurred during Jensen Huang's visit to China, which is by no means a coincidence. Analysts point out that this is both a response to the previous loosening of U.S. policies and a clear declaration that China holds the ultimate authority in the approval process. In October last year, after then-U.S. President Trump met with Chinese leaders in South Korea, Washington cleared the way for NVIDIA to sell H200 chips to China. However, the U.S. set conditions: ensuring sufficient AI chip supply in the United States and requiring Chinese customers to demonstrate adequate security procedures. In the following months, the ball was in Beijing's court. Chinese customs once informed agents that H200 chips were not allowed to enter the country, indicating internal hesitation and deliberation. It was not until this week that the deadlock was broken.
The initial release involves approximately hundreds of thousands of H200 chips, with a market valuation of around $10 billion. Behind this figure lies the long-suppressed demand from Chinese technology companies. Data shows that orders for H200 chips placed by Chinese tech firms for delivery in 2026 have exceeded 2 million units, far surpassing NVIDIA's current available inventory. Both Alibaba and ByteDance have individual demands exceeding 200,000 units each. Even with the additional 25% tariff imposed during the Trump administration, companies are willing to pay.
Practical Considerations Under the Performance Gap: Why Is It?
The shift in Beijing's policy is fundamentally driven by the unavoidable performance generation gap. The H200 is NVIDIA's second most powerful AI chip, and although it is no longer the top-tier in its product lineup (with the more advanced Blackwell architecture already available), for the Chinese market, it represents a significant leap in performance.
Key data reveals the severity of the gap: the computational performance of the H200 is approximately six times that of the H20 chip previously designed specifically for the Chinese market. Meanwhile, domestic chips such as Huawei's Ascend currently only match or, in certain scenarios, surpass the H20, leaving a full generation of technological disparity between them and the H200. This gap directly translates into differences in AI R&D efficiency. Chinese tech giants are investing tens of billions of dollars in building data centers, aiming to develop AI large models and services capable of competing with American rivals like OpenAI. AI startups such as DeepSeek are also rapidly iterating models, work that heavily relies on the most advanced hardware for training and inference.
The failure of the H20 chip has made China more vigilant. This castrated version of the chip, which was allowed for export, was criticized by Chinese state media for its insufficient performance and alleged environmental issues, and its import was eventually halted due to security concerns. This served as a lesson for both Beijing and tech companies: chips with significantly degraded performance cannot meet the demands of cutting-edge competition—either accept technological backwardness or allow the import of more powerful hardware. Clearly, in the midst of the intensifying AI arms race, China has chosen the latter.
The "Dual-Track System" of Independent Strategy: Import Quotas Coupled with Domestic Substitution
Approving imports does not mean that China has abandoned its ambition for semiconductor self-sufficiency. On the contrary, this approval is cleverly designed as a conditional access, reflecting typical Chinese regulatory wisdom—seeking autonomy through openness.
Multiple sources confirm that while approving imports, Beijing has discussed a key condition with relevant companies: they must commit to purchasing a certain proportion of domestic chips in order to obtain permission to import the H200. The specific quota ratio has not yet been finalized, but the policy direction is exceptionally clear. It aims to achieve a dual objective: on one hand, enabling leading technology companies to acquire the advanced computing power necessary to maintain international competitiveness; on the other hand, creating valuable market demand and iteration opportunities for domestic chip manufacturers such as Huawei Ascend and Hygon Information, using the market to feed back into research and development.
This dual-track strategy is not uncommon in China's industrial policy. The third-generation National Integrated Circuit Industry Investment Fund has committed over $47 billion to support semiconductor development. Huawei's Hubble Investment operates more than 60 semiconductor-related enterprises and is building a parallel AI software ecosystem independent of U.S. suppliers. The reality is that China's AI R&D teams have been working in heterogeneous computing environments for many years. Unable to rely on a unified, top-tier chip infrastructure like their American counterparts, they have been forced to develop software systems that can run efficiently on different types of chips (domestic and imported). This experience with hybrid architectures, born out of necessity, could potentially transform into a unique advantage in system optimization once the performance of domestic chips catches up.
The New Normal and Future Landscape of Sino-US Technological Competition
The release of NVIDIA's H200 serves as a small yet significant pressure relief valve in U.S.-China technology relations. It temporarily alleviates revenue pressure for the American chip giant—NVIDIA lost $2.5 billion in revenue last fiscal quarter due to disruptions in its China business, and is projected to lose an additional $8 billion this fiscal quarter. Its market share in China's high-end AI accelerator market has plummeted from 95% to nearly zero. These revenues were originally the lifeblood supporting NVIDIA's efforts to maintain its technological leadership through research and development.
From a broader perspective, this event defines a new normal: in the tug-of-war between decoupling and globalization, complete separation is difficult to achieve, and it will be replaced by controlled, conditional, and limited flows based on performance differentials. The United States delineates technological red lines through export control lists, while China regulates dependency levels through import approvals and domestic production quotas.
This is giving rise to two parallel AI ecosystems globally. One track is dominated by NVIDIA, with top-tier GPUs and a global developer network; the other track is built by China, emphasizing self-reliance, supply chain resilience, and a willingness to accept near-term performance compromises for long-term strategic autonomy. The outcome of this race will depend on how long it takes for Chinese chip manufacturers to close the performance gap with global leaders, and whether Chinese AI companies can leverage limited advanced computing resources to remain competitive in the global market during this transition period.
After concluding his trip to China, Jensen Huang proceeded to Taiwan as scheduled to discuss increasing production of H200 with supply chain partners to meet Chinese demand. This detail alone outlines the reality of the global semiconductor industry—interdependent yet filled with tension. Beijing's approval this time is a pragmatic calculation based on strength, as well as a strategic layout racing against time. The flow of chips has never been as much a commercial activity as it is today, and even more so, a precise extension of national will.