Published by The Lawfare Institute
in Cooperation With
On Oct. 7, the U.S.-China tech competition heated up dramatically when the Biden administration imposed wide-ranging semiconductor-related export controls on China. While there has been plenty of discussion of the extraordinary scope and scale of the controls, less attention has been paid to their timing.
Why did the Biden administration choose to impose these controls now? And why did the Biden team act when it did? Did the United States squander future leverage over China’s artificial intelligence (AI) and weapons development efforts? What was the trade-off considered between acting unilaterally now and spending more time to get allies on board? The question of “Why now?” is an important one, because it has direct bearing on whether these policies will be both effective (that is, by preventing the items and services from going to China) and not counterproductive (that is, by not helping foreign competitors of U.S. companies affected).
The Department of Commerce’s Bureau of Industry and Security (BIS) published a rule with five main parts. One part banned any entity or individual from supplying leading-edge graphics processing units (GPUs)—a type of integrated circuit that accelerates the creation of images—and electronics containing them to any other entity or individual in China. These controls were imposed because GPU chips play an important role in the development and use of artificial intelligence applications, particularly the deep learning methods that are the main driver of the current AI boom. This ban applies to all foreign-made GPUs as a result of a modified extraterritorial jurisdictional rule called the foreign direct product rule, or FDPR for short. The new FDPR subjects any such chip made directly from American technology or software, or produced, even in part, from U.S.-made semiconductor production equipment to U.S. jurisdiction.
Because all semiconductor fabrication facilities use at least some U.S.-made equipment, every such GPU on the planet is now subject to U.S. controls. The Biden administration needed to apply this part of the rule extraterritorially because these GPUs are not made in the United States and no other country subjects them to any form of export control given their widespread commercial applications. No supply of these chips to a Chinese entity can take place without a U.S. government license. Unlike the Trump administration’s policy of granting licenses to Huawei and the Semiconductor Manufacturing International Corporation (SMIC) for certain types of older-generation consumer applications, the Biden team’s announced licensing policy is to presumptively deny all such licenses.
The second part of the rule pertains to semiconductor manufacturing equipment that is specific to producing advanced node logic semiconductors. These export controls primarily restrict U.S. companies—specifically Applied Materials and Lam Research—from shipping certain types of production equipment to China. (The restrictions are limited to American-made machines for which there is not now a foreign-made alternative.) The Japanese company Tokyo Electron and Dutch company ASM International (ASMI) are capable, however, of producing substitute tools in a matter of months, which makes the need for the Japanese and Dutch governments to impose similar controls soon all the more significant. If these controls on substitute tools are not introduced, companies in China could get the same types of tools to produce the same advanced node logic devices (that they have been prohibited from obtaining from the U.S.) as early as 2023. The Biden administration did not propose controls on equipment that is specific to producing advanced node memory semiconductors, but will presumably do so in the future.
A third part of the rule tackles the support and tacit knowledge needed to produce and develop advanced node logic and memory chips in China by barring U.S. persons from providing services without a license, such as maintenance and upgrades to equipment already in place in China. This further strains China’s chip-making capacity because the need for after-sales support and servicing of advanced tools is high. The rule, however, does not bar Japanese, Dutch, or citizens from non-U.S. countries working for non-U.S. companies from providing the same services for the repair and maintenance of the same tools in semiconductor fabrication plants (known as fabs) in China.
The primary goal of this part of the rule is to freeze China’s indigenous logic semiconductor fabrication capabilities at above the 16/14 nanometers (nm) or smaller technology nodes and non-planar technologies. The controls are also designed to degrade the Chinese industry’s ability to advance its development of technologies at these levels. The 16/14 nm threshold is the most advanced planar, two-dimensional technology node. To create smaller feature sizes, chips need to transition to three dimensions. The 3-D transistors, called finFETs, are a critical technological advancement because they are necessary to continue to pack more transistors on a chip to boost computing power. In other words, not having the ability to make chips with finFET transistors severely limits a manufacturer’s ability to create more advanced chips and places a ceiling on the computing power of a virtual machine. Limits on computing abilities means degraded capability in training advanced AI models, developing cutting-edge weapons systems, building supercomputers, and developing complex surveillance systems. For the same reasons, this part of the rule prohibits U.S. persons from providing services at fabrication facilities that produce NAND flash memory chips with 128 layers or more or DRAM with a technology node of 18 nm half pitch or less, both key performance thresholds.
A fourth part of the rule aims to prohibit the export of any type of U.S.-origin commodity, software, or technology to China for the development or production of any type of semiconductor production equipment, or related parts or components. The purpose of this part of the rule was to prevent companies in China from producing any type of semiconductor production equipment, regardless of technology node, that contains elements that were manufactured in the United States. The rule, however, also applies to exports to China of any technology that can be used to produce equipment for non-Chinese companies outside of China. Notably, the rule does not prohibit the shipment of foreign-made items for use in producing semiconductor production equipment to China from outside the United States.
A fifth part of the rule prohibits the shipment of foreign-made items made directly from U.S. technology or software or produced with U.S. equipment to China if destined for use in producing supercomputers in China or to any of 28 companies that have provided high-performance computing support for China’s military.
BIS explained that the export controls were imposed to restrict China’s ability “to produce advanced military systems including weapons of mass destruction; improve the speed and accuracy of its military decision making, planning, and logistics, as well as of its autonomous military systems; and commit human rights abuses.” While most of the details of military facets are classified, the Chinese military modernization drive is well known. Recent press reports also hinted at alarming developments pertaining to China’s nuclear arsenal and hypersonic weapons programs. A litany of human rights abuses in China, such as in Xinjiang, have also been extensively documented. A range of facts, factors, and trends pertaining to semiconductor manufacturing equipment, the looming availability of new advanced chips, and China’s efforts to boost domestic chip manufacturing capabilities with knowledge gained from U.S.-products further support the timing of the administration’s actions.
To be clear, the rule did not impose any new controls on any semiconductors or other items, regardless of sophistication, that have been in any way designed or modified for use in or with weapons of mass destruction, conventional military items, or space- or satellite-related applications. This is because for several decades, the United States has completely prohibited such items from being exported to China, even if they are incorporated into foreign-made items. The rule also did not prohibit the export of any items when there is U.S. government knowledge they would be used to develop or produce weapons of mass destruction or other military items because such exports have also been prohibited for years. The United States and its allies have also long prohibited exports of this technology and other items needed to produce supercomputers at 29 weighted TeraFLOPs, systolic array computers, neural computers, and optical computers to China.
Rather, this rule focused on unmodified items for commercial applications that are nonetheless important to China’s development or production of items needed to develop and produce modern weapons of mass destruction as well as conventional weapons. Another way to think of the controls is that the administration decided that previously uncontrolled supercomputers, semiconductors, and semiconductor production equipment are critical to China’s ability to develop modern weapons. Therefore, the computers and the advanced node chips needed for them to function should be controlled, as should the tools needed to make the chips and the services by U.S. persons needed to keep the tools running. The very existence of indigenous capabilities to develop or produce advanced node semiconductors and supercomputers in China is a national security risk, regardless of any particular application, end user, or end use.
Chokepoint: Semiconductor Manufacturing Equipment
China is entirely dependent on foreign sources for the types of semiconductor manufacturing equipment necessary to produce advanced node semiconductors. Companies based in the Netherlands, Japan, and the United States are the major producers of such semiconductor manufacturing equipment and account for 90 percent of global supply of these essential machines. These firms comprise a major technology chokepoint in Beijing’s efforts to develop a world-class domestic chip industry, providing the United States and allies with tremendous leverage.
White House officials have been discussing restrictions on capital equipment with counterparts in the Hague and Tokyo since Biden’s inauguration. The rationale for controls on this equipment is that using such chokepoints is the most effective way to affect an actor’s advanced node technology indigenization efforts. So far, these talks have not resulted in agreement and Washington proceeded with unilateral action.
The Netherlands apparently does not yet see eye-to-eye with U.S. assessments on the need to set the threshold at 16/14 nm or smaller for logic chips. While the Dutch government has been willing to block shipments to China of advanced extreme ultraviolet (EUV) lithography equipment—sophisticated machines made solely by the Dutch firm ASML Holdings that are required to make the most advanced semiconductors—further restrictions are elusive. Also, EUV tools have been subject to strict multilateral export controls for many years. The types of tools the U.S. proposes for controls are those that are not now subject to any multilateral export controls. A multilateral control is one agreed to in one of the four international organizations that were created near the end of the Cold War to identify the bespoke and dual-use commodities, software, and technologies necessary for the development, production, or use of weapons of mass destruction and conventional weapons.
Dutch disagreement with the White House position on where to set the technology threshold is likely rooted in a divergent threat perception and economic factors. The Dutch government is indeed willing and able to stop the export of items destined for military end uses or end users from the Netherlands. The Biden administration’s approach toward semiconductor production equipment, however, appears for the moment to be too indirect of a threat for the Netherlands to consider imposing unilateral, Dutch-only controls on otherwise uncontrolled tools for making the advanced semiconductors that are overwhelmingly for use in commercial products.
Biden administration officials will need to continue to make the case that it is not in the strategic interest of the tech-leading democracies that China develop a world-class domestic semiconductor industry, and that the best way to ensure that is to freeze China’s capabilities in place. This is a departure from an unwritten strategy used by previous administrations of keeping China a few generations behind with respect to items that the multilateral regimes had identified as dual-use items. The old strategy has an increasingly tenuous objective, as Beijing continues to commit outsized resources to technology indigenization efforts as part of its civil-military fusion policies.
The other objection is financial. As of late October 2022, ASMI expected its China sales to drop 40 percent due to U.S. export controls. China accounted for approximately 16 percent of ASMI’s approximately $1.7 billion in total revenue. There is also the potential that Washington pushes for further restrictions on other equipment by asking the Hague to curtail exports of direct ultraviolet lithography systems. ASML is a key developer of such equipment and derives nearly 15 percent of its total revenue from sales to customers in China, its third biggest market after Taiwan and South Korea. Net sales in China in 2021 were double those in 2019, making China an important growth market for the company. ASMI and ASML executives would likely argue that sales to China are needed to generate the income needed to fund research and development efforts to out-innovate and out-compete its U.S. competitors.
Tokyo, conversely, appears more sympathetic to the U.S. perspective, but has no incentive to move forward unless the Netherlands is on board. The challenge here for the Biden administration is to convince their Dutch counterparts that assessment of the national security risks posed by Chinese capabilities in areas of AI and supercomputing are of such a nature that they cannot be otherwise mitigated. The White House’s efforts to show that it had “skin in the game” by restricting semiconductor manufacturing equipment sales by U.S. companies underscores this point dramatically. More transparency on U.S. intelligence assessments would help to close the gap in risk assessment. That is, the Biden administration needs to more directly explain to the Netherlands that the national security threat is posed by the availability of tools that make the chips that make the computers that are then used to design advanced weapons, when all such items are now in widespread commercial applications. It needs to explain better for public and allied government acceptance why Chinese motives, intentions, and plans make such items a threat to our collective security. Another strategy to achieve this could be to assure that ASMI and ASML are more than able to make up the revenue lost in sales to China with capitalization of newly built fabs in the United States.
The Biden administration nevertheless appears bullish that the issue will be resolved satisfactorily and that the Netherlands and Japan will follow suit. During an event at the Center for a New American Security in late October, Undersecretary of Commerce Alan Estevez, head of BIS, said he expected to strike a deal with allies in the near term. Secretary of Commerce Gina Raimondo later said that up to nine months would be needed to come to such an agreement. Moreover, the incoming chairman of the House Foreign Affairs Committee, Rep. Michael McCaul (R-Texas), called upon the administration to soon “conclude comprehensive agreements with the governments of Japan and the Netherlands to harmonize export control policies.” Reporting by Bloomberg in early December indicated that a deal could be struck as early as January 2023.
Convincing allies to follow suit on manufacturing equipment restrictions is imperative, as controls will be broader and have greater impact when executed in tandem with the Netherlands and Japan. Partnership with the Netherlands and Japan also eliminates the potential that firms—such as ASMI or Tokyo Electron—would fill the gap left by the controls on American-made machines by designing and selling equipment with those specifications.
What is apparently not being discussed with Japanese or Dutch allies is the imposition of controls over services provided by their or any other country’s nationals that would be of use to the development or production of advanced node semiconductors in China. The United States and its allies have “catch-all” controls, which authorize a government to prohibit one of its citizens from providing support to the development or production of weapons of mass destruction. The United States used this authority to say that U.S. support of advanced node semiconductors in China is essentially support for the development of weapons of mass destruction, given China’s military-civil fusion policies. These policies require otherwise commercial items—such as telecommunications and aerospace technologies—to be used in support of China’s military modernization efforts. Getting the Netherlands, Japan, and other countries to use their own catch-all authorities in this way appears to be a bridge too far. Again, the long-term effectiveness of the Biden administration’s new controls, however, will depend on finding a way to convince the allies to control support by their citizens of advanced node development activities in China that cannot now be provided by U.S. persons.
The chip export controls are aimed squarely at China’s military modernization and weapons development efforts, and monitoring, tracking, and surveilling of Chinese citizens. The near-term availability of new highly capable chips was certainly a key factor in timing. The rule specifically targets GPUs, a type of semiconductor that has been optimized for deep learning by speeding up computational processes. The American firm NVIDIA has been one of the main providers of such chips, producing the A100 chip, which was the most sophisticated chip available until the company released a significantly improved GPU, the H100, in October.
Those opposing the controls on specific chips may argue that they prioritize a short-term effect in setting China back over long-term U.S. leverage over AI developments by Chinese researchers. There are two main thrusts to this argument. The first is the well-documented trend that the amount of compute required for the final training run of a machine learning model doubles around every 10 months. Should that trend hold, that could mean that the United States would have more leverage in the future.
There are a couple of reasons, however, that cast doubt on the extent of potential future leverage. One is the general challenge of technology forecasting. Projecting technological advancements by another country is not easy to begin with, even when there are ongoing joint research efforts and scientific exchanges. Technology forecasting becomes very difficult when another country actively obfuscates its progress and collaborative research is dwindling, as is the case with the United States and China. Anticipating the best time to act becomes increasingly uncertain the farther out one projects.
Another reason for the uncertainty over future leverage is specific to AI research efforts. There is plenty of incentive to break the trend of the exponential growth in compute needed, given the cost involved. Researchers are working to develop less compute-intensive deep learning algorithms in transfer learning, which is a “small data” approach to AI. There is also the potential for breakthroughs in AI that address some of the shortcomings of deep learning—such as not understanding causality and limitations in reasoning—by pursuing so-called hybrid AI, which marries aspects of deep learning, neural networks, and symbolic AI. Here too, one likely advantage is that less data and less compute would be required to train the models. Also, to the extent that advanced computing capabilities are needed, one could outsource that to computing systems outside of China. Neither the U.S. export controls nor those of any allies prohibit the provision of advanced computing software as a service to China unless there is knowledge that the service is specific to the development of weapons of mass destruction.
The second thrust of the argument against the chip-specific restrictions is that it will accelerate China’s semiconductor industry indigenization efforts and encourage the development of chips that are not produced with U.S. technology, software, or equipment and thus would not be subject to the FDPR. There is little doubt that the new export controls will increase the urge in Beijing to reduce reliance on non-U.S. technology in this critical field. But China cannot will its way to autarky. It has fallen well short of domestic capability targets set in 2015, and corruption in the chip sector has been a major headwind. Beijing faces considerable dependencies and shortfalls in tech, components, talent, skills, and knowledge—all of which will be bigger issues because of the new export controls and U.S. persons restrictions. These are not problems that can be solved simply by throwing more money at them.
That the export controls would incentivize other companies to create chips, components, and equipment free of U.S. content is a realistic concern and a potential unintended consequence. The Biden administration took a calculated risk in this regard. U.S. officials will need to persuade counterparts in allied countries that dominate the global semiconductor industry—especially Japan, South Korea, Taiwan, Germany, and the Netherlands—that it would be unwise to do so. Given the hardening attitudes toward China in these countries and mounting concerns over China’s technological capabilities, there is room for optimism that other countries will follow the U.S. lead and will not attempt to exploit the situation. These attitudes will have to overcome a desire by companies in their countries to fill behind in China the types of items that U.S. companies are now no longer able to sell.
Quiet Masterstroke: U.S. Persons Control
The addition of restrictions on U.S. persons activities, such as by U.S. citizens and corporations, adds a lot of bite to the rule. These novel controls apply to persons who maintain and repair items used in the development or production of advanced node semiconductors (such as manufacturing equipment, notoriously finicky machines) or even authorizing deliveries of noncontrolled foreign-made items used in the development or production of semiconductors to facilities in China that produce or develop advanced node semiconductors.
These measures are likely to make an outsized contribution to an eventual degradation of Chinese semiconductor capabilities because it directly impacts production lines currently in operation. Much of the progress China has made in technological development is due to the contributions of foreign-born scientists, engineers, and many key executives that fall in the U.S. persons category. The impact was immediate: Firms such as ASML and Naura Technology, a Beijing-based semiconductor company, pulled U.S. persons off servicing and research development efforts shortly after the rule was announced.
There are, however, many clever engineers, scientists, and technicians, particularly at Tokyo Electron and ASMI who do not have U.S. passports or green cards. Unless and until the Biden administration can convince allies to impose controls over the activities of their citizens in support of advanced node production equipment in China, there will be no prohibitions on their doing so.
What to Look For
The prevailing view is that the new export controls will have a major impact on China’s semiconductor sector and its AI and weapons development over the next few years. Longer term projections are more difficult, but there is ample reason to assess that the Biden administration will largely achieve its stated objectives. Two factors have outsized importance on the overall effectiveness: implementation and engagement by allies. Expect scrutiny by Congress of how BIS conducts enforcement and compliance and a full court press by U.S. officials on their colleagues in key countries to enact their own controls.
From a technical perspective, eyes will be on whether the Biden administration remains confident in where it drew the line on chip performance parameters. In early November, NVIDIA announced it would offer the A800, an advanced GPU with a similar computational performance of the A100 but with a capacity to send and receive data from other chips that is below the new export control thresholds. Similarly, Chinese firms Alibaba and Biren Technology announced design changes to reduce processing speeds for chips under development. The question of whether chips with reduced bandwidth can still train cutting-edge models will be closely scrutinized. Breakthroughs in the aforementioned research on less compute-intensive algorithms would be another factor that could prompt U.S. officials to expand controls.
Decision-makers in Beijing have a say in the outcome as well, of course. Response options by China’s leaders range from harassment of U.S. businesspeople, to stepped-up industrial espionage, to curtailing exports of rare earths, although the latter is unlikely because it would be highly escalatory. Beijing will surely try to peel off allies, Germany and South Korea perhaps, in an attempt to circumvent or weaken U.S. measures to constrain China’s technological development.
On semiconductors specifically, Beijing may choose to double down on cornering the market for chips made with 28 nm process technology by undercutting competitors such as TSMC, UMC, and GlobalFoundries on price. These so-called legacy chips are widely used in consumer electronics, automobiles, and certain defense systems. China’s dominance in this area could create a huge supply chain vulnerability, which would amount to a major unintended consequence of the export controls.
One measure to stave off the possibility of Chinese firms dominating legacy chipmaking, however, is a provision in the National Defense Authorization Act for Fiscal Year 2023 (NDAA). The amendment would prohibit government contractors from providing the federal government with products that contain chips from the Chinese firms SMIC, YMTC, and CXMT. Should this amendment be in the version of the NDAA that becomes law, it would ensure some purchasing power by the U.S. government for non-Chinese firms to maintain production lines for these semiconductors. The new provisions would not, however, prohibit U.S. companies, foreign companies, or foreign governments from purchasing such chips for use in products that are outside the U.S. government’s procurement network.
These developments are as complex as they are consequential. There is no crystal ball that can divine the outcome, given how unprecedented and wide ranging these actions are. What is clear, however, is that the near-term impact will be significant and the ability to decipher when to act in the future is too tenuous and risky. The Biden administration made the right call by acting now, particularly if it is successful at getting allied cooperation on the essence of the rules soon.
Martijn Rasser is director of the Technology and National Security Program at the Center for a New American Security. Kevin Wolf is a nonresident senior fellow at Georgetown’s Center for Security and Emerging Technology, a partner at Akin Gump, and served as assistant secretary of commerce for export administration from 2010 to 2017.