Congress has introduced another law tightening export controls on additional segments of the semiconductor industry (the MATCH Act). Like all the others, it is based on the premise that US technological leadership and national security is advanced by looking backwards and throwing obstacles at trailing China rather than running faster and breaking new ground. Like the other laws, it reinforces China’s own commitment to indigenize its semiconductor industry, and so works against its stated goal of not allowing China to catch up. American policy makers and politicians are so fixated on retarding China’s digital development that they have forgotten about what created American leadership in the first place.

Jensen Huang, CEO of Nvidia, debunked the argument that the US can contain Chinese AI development through export controls on computing power. He believes that chip control is futile because China can utilize its other advantages to aggregate and scale up its computing power, and the US will ultimately lose its global dominance on the global market by depriving its own leading companies of markets and profits. Huang’s rare, emotional expression struck at the core question: what is American technological leadership? Different answers to this question lead to divergence on chip export control policies.

The Biden policy of containment

In the Biden Presidency, the US government adopted a “Framework for AI Diffusion”, seeking to contain Chinese development of AI by controlling advanced computing power. This framework was based on an assumption that superintelligence is “imminent”, and computing power is the most important resource enabling it. At that time, the underlying premise of the “AI race” equated it with the race to build the atomic bomb: whoever gets it first would hold supreme power over the world. Hence, the U.S. must use its leading position in semiconductor production to carefully ration access to computing power or it will lose the race. Under this assumption, the Biden administration mandated BIS to develop targeted regulations restricting the export of chips to China, and Floating Point Operations Per Second (FLOPS), a measure of computer performance, became the critical metric for determining restrictions.

Yet Mueller (2025) and many others have debunked the whole theory of super intelligence.  There is no “race” to a defined finish line that brings ultimate power to the winner. There are only incremental, increasingly energy-intensive gains toward faster chips. And as Jeffrey Ding (2024) demonstrated convincingly in his book Technology and the Rise of Great Powers, it is a nation’s ability to put new technologies to work across its economy that matters most in competitions over wealth and power, not being the first to develop the technology.

Even if the rationale behind export controls is taken at face value, Murphy (2026) summarized its three loopholes: diplomatic cleavages created by tiered access categories, China’s own indigenisation efforts on computing, and the substitution of software and algorithmic innovations for compute requirements. These risks all indicate that computing diffusion control is not the right basket into which to put America’s eggs. Tellingly, the debate around export control has already shifted from whether chip control can contain Chinese development to “for how long” it can prolong the catching-up time.

Trump shifts gears, but Congress carries on

The Trump II administration has deviated a little from the “diffusion control” model. The July 2025 executive order on Promoting the Export of the American AI Technology Stack shifted focus from export restrictions to global market dominance. Under this approach, the costs and competitiveness of American companies became more of a priority. As a result, the BIS revised its license review rules in January 2026, shifting from a presumption of denial to case-by-case review. This does not mean that the Trump admin is returning to free trade; the MAGA trade doctrine still applies to chips, but chip exports are still allowed with certain caveats.

Not everyone in Washington agrees with this trend, however. The biggest opposition comes from Capitol Hill. From RASA to MATCH Act, the U.S. Congress has put forward bills to continue clamping down on the ability of American digital companies, which lead the world in size and scope, to do business globally. Five such bills are outlined below.

Why is Congress more aggressively restrictionist than the Trump administration? Three words: special interest lobbying. According to a Reuters article, Micron Technology (MU.O), the largest U.S. memory chipmaker, is a driving force pushing the U.S. Congress to pass the MATCH Act legislation, for example.

Leadership through market domination, or political control?

There are two approaches to sustaining a country’s technology leadership. One argues to keep strengthening America’s domination across world markets selling its advanced chips; the other side wants to build a wall around advanced chip design and manufacturing to restrict Chinese access to new chips.

If one were to steelman Dwarkesh’s argument, export controls are needed for the following reasons: 1) in a direct conflict with China, Chinese AI will provide a decisive advantage in a military conflict, and chip controls can prevent it; 2) The US has substantial leverage in terms of computing power and other layers of the computing stack, so why not use it to gain advantage?

On the other side, Jensen Huang seeks to sustain the American tech leadership by maintaining its central role in the semiconductor supply chain. This approach drives revenue to American companies and fuels the R&D and talent acquisition that drives innovation and better applications. Huang dismissed diffusion control as a “loser” mentality. Comparing chips to nuclear bombs is a “lousy” comparison, a product of fear-mongering. Naturally, Jensen’s judgment aligns with Nvidia’s interest as a private company. Nevertheless, there are real reasons to believe that it is more detrimental for the US to lose its dominance in global markets than to cave in to hyperbolic security risks about China’s AI capabilities.

From a theoretical perspective, this dilemma can be traced back to the popular theory of “weaponized interdependence,” a term political scientists Farrell and Newman popularized and a policy that many countries began to pursue years ago. Interdependence is said to be a source of asymmetric power, and states can weaponize such power to pursue their strategic goals. However, weaponization inevitably diminishes market domination, as the excluded party seeks to reduce their reliance on the original technology diffusion network.

What’s worse, it is evident now that export controls cost American companies more than Chinese firms. Economists from the Federal Reserve estimated that the export control since 2010 creates a total capitalization loss of $18 billion for Chinese targets, but for each affected American company, the average market capitalization loss reaches $1 billion, meaning a total decrease of $158 billion for American companies, along with an average revenue decline of 8.9%, and 7.3% decline in the total number of employees.

The war in Iran is making it clear that a simple technology and power advantage is not always sufficient for a strategic victory in a conflict. The US had faster chips, faster planes, better intelligence, more precision bombs, better missile defenses, but it was not enough to dislodge the Iranian regime or make it surrender, nor was it able to prevent the Strait of Hormuz from being blocked. Likewise, in a real-world US-China conflict, it’s hard to imagine how a 9 or 6 month lead in semiconductor power is going to make a decisive difference – for either side.

How China offsets the chip export controls

Predictably, chip export controls have pushed the Chinese companies away from any dependency on the American computing supply chain. They have undertaken the following efforts to achieve the goal.

• Fostering the domestic chips ecosystem: at many semiconductor production stages, domestic equipment shares have risen in the Chinese market. Per CSIS, overall domestic equipment share surged from 25% to 35% between 2024 and 2025, while domestic producers accounted for less than 15% of the domestic market before the US imposed export control. Noticeable progress came from the etching and thin-film deposition procedures, where local firms now supply about 40% of the domestic market. A recent harder push came from a new procurement policy in December 2025, requiring that Chinese fabs seeking approval for new capacity need to buy at least 50% of their equipment from domestic suppliers.

• Importing from non-US suppliers. Along with these self-sufficiency efforts, China’s semiconductor-manufacturing equipment imports reached a record high of $51.1 billion, almost five times the level of a decade earlier. But this new increase mainly came from increased imports from Japan, Netherland, and Singapore, while the import share from the US has decreased from 23% to 9% in the past 10 years.

• Using cloud and data center capacity in Southeast Asia: the tiered-category of America’s chip export control regime also compels Chinese companies to look for computing capacity outside mainland China, especially in Southeast Asian countries. For example, according to Reuters, by March 2026, ByteDance was working with Aolani Cloud in Malaysia on about 500 Nvidia Blackwell systems, totaling roughly 36,000 B200 chips, in a project costing more than US$2.5 billion. At the national policy level, China-ASEAN FTA 3.0 has positioned digital cooperation as a core pillar, specifically on digital trade, cross-border fiber optics, and data centers.

• Maximizing its advantages in energy efficiency: improvements in model efficiency and energy network can also offset the impact of computing power control. DeepSeek, along with other Chinese open-source models, are the most prominent case following this strategy; their innovations on model design reduces demands for computing power compared to American ones.

However, little attention has been given to a more fundamental advantage that China has been accumulating: the colossal national computing-energy network that reduces the energy price for computation. Since 2022, China has been constructing its “East-Data-West-Computing” (东数西算) project, linking data-abundant Eastern provinces with energy-affluent Western provinces. By the end of 2025, national hub nodes and data-center clusters under this national project took up 80% of China’s total 1.59 million PetaFLOPS computing capacity. All these improvements at the model and energy-computing layers result in comparatively lower token prices and dependencies on American computing devices.

All these efforts demonstrate China’s efforts to reduce reliance on the American computing network, but this does not mean that China has circumvented all the computing bottlenecks. But it is certain that China will continue to reduce its reliance on the US computing devices, undermining the very material foundation of imposing export control.

AI is not just about chips and models, it is a whole digital ecosystem that contains multiple parts. For Mueller, the ecosystem contains software, data, computing devices, and the network; In Jensen’s opinion, the AI ecosystem is a five-layer AI stack that also includes energy. Either way, competition in AI is a multi-layered game driven mostly by the market. Different states possess different advantages and leverage across the layers of the digital ecosystem. What’s clear is that the U.S. will not maintain its advantages by exclusions and restrictions on its own industry. That is a form of unilateral disarmament.

The post Running a Race Looking Backwards: US Digital Export Controls appeared first on Internet Governance Project.