Shift in U.S. Chip Export Policy Toward China: A Calculated Strategic Game

On January 15, 2026, the U.S. Department of Commerce's Bureau of Industry and Security (BIS) quietly published a new regulation in the Federal Register. The core of this document is the revision of the license review policy for exports of advanced computing semiconductors to China. The previous "presumption of denial" principle, which applied to exports of specific chips, has been adjusted to a case-by-case review under certain conditions. This means that a range of advanced AI chips, such as NVIDIA's H200 and AMD's Instinct MI325X, which were previously subject to strict restrictions, have gained the possibility of re-entering the Chinese market after meeting a series of stringent requirements.

This official document marks a significant and complex shift in U.S. policy since the Biden administration initiated export controls on advanced semiconductors to China in 2022. It is not a simple lifting of restrictions, but rather a strategic adjustment made under multiple pressures—national security, economic interests, technological competition, and geopolitics—after careful calculation. On the chessboard of this game lie not just chips, but also the stakes for future dominance in artificial intelligence.

Policy Relaxation: From "Presumptive Denial" to "Conditional Approval"

The introduction of the new regulations did not come without warning. By the end of 2025, President Trump, after being re-elected, had already signaled that he would allow the export of advanced chips to approved customers in China, subject to a 25% government surcharge. This was widely seen as a response to months of lobbying by NVIDIA CEO Jensen Huang. Now, the decision has been made, but the manner of its implementation is far more complex than a simple approval for exports.

The Subtle Setting of Technical Thresholds is the first key point of the new regulations. To obtain an export license, a chip's Total Processing Performance (TPP) score must be below 21,000 points, and its total DRAM bandwidth must be below 6,500 GB per second. NVIDIA's H200 has a TPP score of approximately 15,832 and a bandwidth of about 4,800 GB per second; AMD's MI325X has a TPP score of around 20,800 and a bandwidth of about 6,000 GB per second. Both are positioned just below the threshold. This setting is by no means accidental; it precisely excludes the most top-tier chips at the time, which could potentially bring disruptive computational advantages (such as NVIDIA's Blackwell architecture chips and future Rubin chips), while leaving a narrow opening for second-tier, yet still advanced, products.

Supply Guarantee with U.S. Priority is the second core condition. Exporters must demonstrate that domestic demand in the United States has been fully met. This means that orders from U.S. customers cannot be delayed due to exports to China, and U.S. manufacturing capacity cannot be diverted to fulfill Chinese orders. More critically, the total shipments to China (including Macau) must not exceed 50% of the total shipments of the same product to domestic U.S. customers during the same period. This clause has been interpreted by industry observers as positioning the Chinese market as a destination for absorbing the surplus capacity of the U.S. market. The hardware news website Tom's Hardware commented incisively: This essentially monetizes some commercial interests while ensuring the dominant position of U.S. industries.

Layered Security Review constitutes the third, and perhaps the most cumbersome, checkpoint. All chips intended for export must be verified by an independent third-party testing laboratory located within the United States before shipment to confirm the accuracy of their technical specifications (such as TPP, bandwidth). This laboratory must have no financial ties to the exporter or importer and must not be controlled by entities outside the United States. Furthermore, the importing party must implement strict Know Your Customer (KYC) procedures and adopt adequate physical security measures. These measures include preventing remote access by Chinese military or intelligence agencies, and prohibiting the unauthorized transfer of trained model weights and algorithms.

Game Theory Logic: Rebalancing National Security and Economic Interests

Behind this series of complex clauses lies a long-standing policy route dispute within the U.S. decision-making circles. On one side, represented by national security hawks, there is concern that advanced chips flowing into China could be used for military purposes, accelerating its militarization of artificial intelligence and thereby threatening U.S. strategic advantages. On the other side, from a more industrial and economic perspective, it is argued that overly stringent bans are a double-edged sword.

Innovation-Driven Concerns are becoming a reality. The export controls of the past few years have objectively stimulated R&D investment and policy support in China's domestic semiconductor industry. Reports indicate that Chinese authorities have encouraged or even required tech companies to purchase a certain proportion of domestic AI chips. If the U.S. completely cuts off the supply of high-end chips, it would only further force Chinese companies to turn to local alternatives. In the long run, this could potentially accelerate China's technological breakthroughs in semiconductor design and even manufacturing. To some extent, the Trump administration's current policy adjustment is an attempt to delay this process—by maintaining a certain level of supply, it aims to preserve the penetration and influence of U.S. technology ecosystems in the Chinese market.

The Power of Corporate Lobbying cannot be overlooked. As the global leader in AI chips, the Chinese market was once a crucial growth driver for NVIDIA. The ongoing export restrictions have cost it significant potential revenue. The core argument of CEO Jensen Huang's lobbying efforts is precisely to tie commercial opportunities to American interests. In its statement, NVIDIA claimed that the new regulations are well-considered, achieving an excellent balance that is highly beneficial to the United States, and emphasized that American critics are inadvertently advancing the interests of foreign competitors on the Entity List. This rhetoric elevates market competition to the level of national industrial rivalry, clearly finding an audience in Washington.

Direct Consideration of Financial Interests has also surfaced. According to reports, the Trump administration will impose a 25% surcharge on the sales of H200 chips approved for export. This serves both as a political gesture, demonstrating that the tough stance towards China remains unchanged, and as a significant source of fiscal revenue. A Reuters report at the end of December 2025 revealed that Chinese tech companies had ordered over 2 million H200 chips for 2026, far exceeding Nvidia's inventory of 700,000 chips at the time. Even if only a portion of these orders are approved, the surcharge would be substantial.

Market Reaction: The Tug-of-War Between Demand and Substitution

The policy gate has loosened, yet the market's response is not one of unbridled enthusiasm but rather filled with caution and strategic maneuvering.

The Complex Mindset of Chinese Buyers is the primary variable. On one hand, the thirst for high-end computing power is real. In an era where the scale of AI models is growing exponentially, the H200 offers a performance improvement of up to six times compared to the previously export-approved, restricted H20 chip, which is crucial for training cutting-edge large models. The substantial order intentions already indicate the existence of market demand.

On the other hand, policy uncertainty and domestic substitution pressures are making Chinese companies tread cautiously. Reports indicate that Chinese officials have informed some companies that approval for purchasing H200 chips will only be granted under special circumstances, such as development labs or university research. Furthermore, there are reports that Chinese authorities have called on companies to suspend H200 procurement while considering mandating the purchase of a certain proportion of domestic AI chips. This means that Chinese enterprises face difficult trade-offs between performance, supply chain security, policy compliance, and cost. They desire access to advanced computing power, yet must also consider the political imperative of supporting the local industrial chain, while simultaneously avoiding the risk of potentially tightening U.S. controls in the future.

NVIDIA's Mixed Blessings. Obtaining export licenses is a victory, but the accompanying stringent conditions severely limit its market potential. The 50% shipment cap, priority supply to the U.S., cumbersome review processes, and the 25% sales surcharge all mean that its revenue recovery in the Chinese market will be significantly diminished, with high operational costs. NVIDIA must tread carefully, balancing between the U.S. government, Chinese clients, and its own global supply chain.

Future Outlook: A Far from Over Enduring Competition

The recent adjustment of the U.S. chip export policy toward China is by no means the end of competition; rather, it marks the beginning of a more complex and dynamic new phase.

Acceleration of the Technology Race is an inevitable trend. By setting precise performance thresholds (TPP < 21,000), the United States attempts to lock China onto a sub-advanced technological trajectory while reserving a lead of at least one to two generations for its own next-generation chips (with TPP far exceeding this number). However, the success of this "boiling the frog" strategy depends on the pace of China's indigenous innovation. Progress in areas such as chip design (e.g., Huawei's Ascend), Chiplet packaging, and computing power network construction in China may enable surpassing the limitations of single-chip performance through different pathways.

The Normalization of Rule-Based Gaming will become the new norm. Future export controls will no longer be a simple matter of prohibition or permission; instead, they will evolve into a complex regulatory system based on dynamic performance indicators, supply chain reviews, and end-user verification. Provisions such as third-party testing, KYC procedures, and remote access restrictions indicate that technology trade will increasingly become deeply intertwined with issues like data security and algorithm governance. This is not merely about commercial rules but also a contest over technical standards and security discourse power.

Reshaping the Global Supply Chain has far-reaching implications. The U.S. "America First" supply terms further reinforce the importance of its domestic supply chain. This may prompt technology companies in other regions, to ensure their own supply security, to consider diversifying or regionalizing their supply chains, thereby impacting the existing global semiconductor division of labor system. China will redouble its efforts to build a self-controlled and secure industrial chain, even if it initially lacks advantages in performance and cost.

This game surrounding AI chips is essentially a struggle for control over the infrastructure of the future intelligent era. The recent shift in U.S. policy represents a tactical retreat and reconfiguration, aimed at maximizing its own economic and strategic interests without excessively stimulating China's independent innovation. It attempts to use an elastic band to restrain the pace of China's AI development—neither allowing it to break free nor letting it stagnate to the point of turning completely inward.

For China, the limited restoration of external supply may alleviate short-term computing power anxiety, but **the strategic direction of domestic substitution shifting from an alternative to a necessity has become irreversible**. The ultimate outcome of this race does not depend on the issuance of one or two export licenses, but on who can build a more profound and enduring innovation system in fundamental research, talent cultivation, industrial ecosystem, and engineering capabilities. Chips are merely the surface; the contest of intellect and systems behind them has just entered the deep waters.