The Limits of the China Chip Ban

Washington’s Export Controls Could End Up Helping Beijing

By Hanna Dohmen, Jacob Feldgoise, and Charles Kupchan
July 24, 2024

In 2022, amid rising U.S.-Chinese tensions, the Biden administration rolled out export controls to prevent Beijing from obtaining advanced semiconductors and the equipment to produce them domestically. The stated objective of these restrictions was to deny China the cutting-edge AI capabilities it could use to modernize its nuclear and conventional weapons. U.S. Secretary of Commerce Gina Raimondo insisted that the controls were “laser-focused” on impeding Beijing’s military development. But these measures may also protect the United States’ technological and economic edge over China. Although leadership in AI is not officially stipulated as an aim of the restrictions, U.S. officials, including Raimondo and U.S. National Security Adviser Jake Sullivan, have regularly asserted that it is central to the country’s competitive economic advantage, which in turn advances its national security.

But the chip controls will probably fall short of achieving either outcome. They are unlikely to substantially slow Beijing’s military modernization, much of which can be accomplished using older legacy chips. Where cutting-edge AI chips are needed, the Chinese military can use previously imported chips, smuggled chips, and domestically designed and produced chips. The controls will likely be more consequential when it comes to enabling the United States to maintain its technological edge. By impeding China’s ability to develop and deploy AI throughout its economy, the export restrictions could slow China’s growth and curb its competitiveness, thereby helping the United States stay ahead.

The benefits, however, will be only temporary and the costs high. There are strong indications that the controls are expediting China’s indigenous development of its own semiconductor supply chain. As a result, U.S. actions may impede Chinese innovation and growth only in the short term, and thereafter actually speed up its technological advance. Meanwhile, chip equipment companies in the United States and in allied countries are already seeing declines in revenue as they are forced to leave the Chinese market, denying them funds to fuel research and development. China’s semiconductor industry may soon be able to catch up, potentially leaving the United States and its partners with diminished leverage over China ​​at the same time as export controls increase the risks of economic decoupling and geopolitical fracture.​​

The current U.S. strategy is flawed. Washington should place less emphasis on slowing down China and focus more on improving its own innovative prowess. Moving forward, the U.S. government must capitalize on the temporary Chinese slowdown afforded by export controls to establish a decisive lead in the most important technologies of tomorrow.

STAYING ONE STEP AHEAD

China’s semiconductor industry has made significant progress over the past ten years in AI chip design, tooling, and advanced fabrication. Two U.S. design companies, Nvidia and AMD, lead the AI chip market, but Chinese design firms Huawei and Biren have been making progress. Both companies unveiled advanced chips—Huawei in 2019 and Biren in 2022—with performance specifications similar to those of U.S. firms. But even if Chinese firms are close behind U.S. market leaders in chip design, they lag behind in manufacturing advanced-node semiconductors—the lower nanometer chips that power AI. Currently, the most advanced manufacturing process developed by China’s leading fabrication company—Semiconductor Manufacturing International Corporation (SMIC)—is producing chips that are about five to six years behind the state of the art. Furthermore, SMIC’s best fabrication facilities still rely heavily on Dutch, Japanese, and U.S. equipment, as Chinese equipment​​​​​​​​ remains incapable of producing the most advanced chips.

The U.S. government wants to protect the country’s advantage, seeking to “maintain as large of a lead as possible” in technologies that power both military modernization and economic growth, as Sullivan put it in September 2022. The Biden administration is committed to ensuring that the United States stays in front on AI, primarily by preserving its technological edge over the chips crucial for developing AI systems. Raimondo bluntly summarized the plan in December 2023: “America leads the world in artificial intelligence. America leads the world in advanced semiconductor design, period. . . . We’re a couple years ahead of China. No way are we going to let them catch up.” Hence the export controls.

Washington’s stated objective is narrow: to block China’s access to chips that the U.S. Bureau of Industry and Security states can be used to “train frontier AI models that have the most significant potential for advanced warfare applications.” Yet as Sullivan has explained, maintaining the United States’ lead in computing hardware and AI advances its economic competitiveness, not just its military superiority. Although the Biden administration denies that the export controls are intended to limit China’s technological or economic development, the restrictions do offer a potential means of doing just that. Denying China access to the most advanced chips should limit the computational power available to Chinese firms and engineers, thereby holding back the country’s ability to develop sophisticated AI systems and reap their productivity benefits.

The Biden administration has characterized its policy as a “small yard, high fence” approach, meaning that stringent restrictions apply only to a limited range of chips and fabrication tools. In 2022, U.S. regulators banned the export to China of the most advanced chips, which are used to build AI models, and the equipment used to make them. After gaps in the restrictions became apparent in 2023, the administration tightened the controls and coordinated the restrictions with its allies. The regime will likely continue to evolve as Chinese companies find workarounds and the U.S. government identifies additional shortcomings in its effort to slow China’s technological advance.

Washington is right that the country’s national security goes hand in hand with its economic competitiveness. It is therefore in the interests of the United States to hinder China’s military modernization, as well as to seek to ensure that U.S. economic and technological competitiveness remain several steps ahead of Beijing's. The problem is that the export controls are unlikely to significantly hamper China’s military modernization and will only temporarily impair its economic competitiveness.

WHERE THERE’S A WILL

Washington’s export controls will do little to arrest China’s military modernization, as most current weapons systems do not rely on the advanced chips that are subject to the restrictions. U.S. regulators have even acknowledged as much. Instead, military systems prioritize reliable, well-tested chips, which are typically fabricated on older equipment. The processors used in most weapons of war—tanks, missile systems, and even drones—are not covered by the 2023 controls, meaning that the restrictions will have a limited effect on these weapons’ capabilities. Nonetheless, advanced chips do have some military applications; in particular, they can expedite the design of advanced missile systems and weapons of mass destruction.

But to the extent that China’s military does need advanced computing capabilities to develop weapons and train AI models, it will likely be able to satisfy that demand with smuggled chips and larger quantities of lower-performing domestic alternatives. The technology controls that the West imposed on Russia following its invasion of Ukraine in 2022 have made clear the challenges of preventing evasion. Despite the unprecedented scope of coordinated sanctions, existing export controls on Russia have not stopped it from importing significant quantities of strategic goods, including semiconductors designed by U.S. companies. If China’s military and its commercial affiliates are able to smuggle just 11,000 of Nvidia’s cutting-edge H100 AI chips—0.73 percent of the 1.5 million that the company is expected to ship in 2024—China’s military would likely be able to train a relatively advanced large language model in about a week. The controls will have even less of an effect on China’s ability to develop other types of AI models that can be trained on fewer or older chips. China can also make use of domestically produced chips, the best of which is roughly as fast as one of Nvidia’s better chips.

Export controls may drive up the costs of China’s military modernization efforts, but the Chinese government has a proven track record of expending the resources necessary to pursue its strategic objectives. Indeed, the progress made by China’s domestic semiconductor industry is due in part to Beijing’s steady investments in the sector. Given the global proliferation of controlled AI chips, the difficulty of preventing smuggling, and the availability of uncontrolled alternatives, U.S. export restrictions hold little prospect of slowing China’s military modernization efforts.

A SHRINKING WINDOW OF OPPORTUNITY

U.S. export controls hold more promise for maintaining the United States’ competitive edge by slowing China’s development and deployment of AI models. In the short term, Chinese stockpiles of AI chips are likely large enough to meet demand, but these stockpiles will soon be used up, and U.S. controls will start to bite. Even with smuggled and domestically fabricated alternatives, China will likely lack access to the large supply of advanced chips it needs to scale up AI across its economy, potentially slowing its economic development relative to that of the United States. Until Beijing develops domestic alternatives for advanced semiconductor tooling and chip fabrication, Washington should have an edge. This advantage, however, is likely only to be temporary.

The main problem for Washington is that the export controls may inadvertently accelerate the development of China’s domestic semiconductor industry. By limiting China’s access to foreign-made chips and manufacturing tools, the controls are creating new demand for indigenous Chinese equipment, fabrication capacity, and AI chips. This domestic demand is putting pressure on Chinese companies to invest in and collaborate up and down the semiconductor supply chain. Firms in Japan, the Netherlands, and the United States still have a virtual lock on the tools needed to make the most advanced chips. But by preventing these firms from selling their equipment to China, U.S. export controls are creating market conditions that will drive revenue toward China’s own equipment makers, allowing these companies to invest in the research and development activities needed to produce more sophisticated tools.

The export controls will probably spur China’s semiconductor industry to catch up with the market leaders. Given the current rate of progress, it is likely that China will, before long, be able to manufacture the equipment needed to fabricate advanced semiconductors. When that happens, China’s capable chip design firms will be able to use domestic fabrication capacity to produce AI chips at scale and to diffuse AI throughout its economy. Exactly when China will reach this threshold is unknown, but the United States has only a temporary window of opportunity to capitalize on the disparity.

This window of opportunity may well be narrowed by industry trends. AI chip design companies are connecting multiple less powerful chips—known as chiplets—to form a single higher-performing package that is more capable of training and using AI models. Many semiconductor companies are using this strategy to reduce design and manufacturing costs, and Chinese firms are following suit with government support. In addition, AI researchers in China and other countries are developing smaller and less sophisticated AI models that, although still quite capable, require less computational capacity. These two innovations could help China keep pace in AI as its firms work to expedite the development of advanced chip-making equipment.

In the meantime, U.S. controls will cause chip and tool manufacturers in the United States and allied nations to lose considerable business in China, leaving these companies with less money to invest in research and development. In the short term, the effect will likely be modest. The current AI boom has created demand for high-performing chips that dramatically exceeds supply; chips that Nvidia would have sold in China are instead being sold elsewhere. But over the longer term, pulling out of China’s huge market is expected to take a significant toll on Western firms’ revenue.

The export controls could also expedite the fragmentation of the global economy and deepen the geopolitical divide between China and the United States. A bigger yard and higher fence would reduce economic interdependence and intensify geopolitical rivalry, potentially putting the United States and China on a collision course. Indeed, the chip controls have already ramped up U.S.-Chinese tensions, contributing to Chinese leader Xi Jinping’s assessment that the United States is seeking to implement “all-round containment, encirclement, and suppression against us.”

ACCELERATING INNOVATION

The key question moving forward is whether Washington can capitalize on the window of opportunity afforded by the export controls to make paradigm-shifting breakthroughs in computational science. To do this, the United States’ best bet in the long run is to promote its own technological advances and innovative capacities.

The U.S. government has made good progress on this front with the 2022 CHIPS and Science Act, which included a $52 billion investment in advanced semiconductor production, applied research, prototyping, and commercialization. Advanced domestic fabrication plants should make it easier for U.S.-based semiconductor researchers to develop and scale new and leading-edge technologies. Furthermore, the National Semiconductor Technology Center, which the CHIPS Act financed, is funding state-of-the-art facilities to foster collaboration between U.S. semiconductor companies and corporate, government, and university researchers.

These are important steps in the right direction, but they will likely be insufficient. They focus on accelerating incremental innovation rather than investing in the next big thing. By the time Chinese domestic chip production has caught up, the United States needs to have already moved on to tackling the next technological frontier in advanced computing. Light-based computing (using photons to process data) and neuromorphic computing (employing operations that mimic the human brain) are two promising candidates for the next-generation computing paradigm. In addition, quantum computing (using subatomic particles to process information) should exponentially accelerate calculations for certain applications. These emerging technologies hold the potential to compute at unprecedented speeds and with much less energy use—facilitating scientific discovery, transforming industries, and modernizing militaries. Washington must now advance these moonshot technologies by funding basic research, expanding workforce development programs, and investing in the domestic manufacturing ecosystem.

Driving U.S. innovation should be Washington’s top priority when competing with Beijing. Given China’s size, economic heft, and scientific sophistication, the U.S. government has only limited capacity to obstruct its rival’s technological development. The United States should, instead, focus its energies on its own innovative prowess and ability to stay at the head of the pack on AI and the next generation of critical technologies.

https://www.foreignaffairs.com/china/limits-china-chip-ban