Today, the Commerce Department released a series of updated export regulations designed to curb China's progress in the field of advanced AI technologies. These revisions come approximately a year after the initial introduction of export controls. CNAS experts weigh in on these new provisions and their implications for chips, artificial intelligence, hypersonic missiles, and more.
All quotes may be used with attribution. To arrange an interview, email Alexa Whaley at firstname.lastname@example.org.
Paul Scharre, Executive Vice President and Director of Studies:
There is a lot to digest in the Biden administration’s new weighty export control regulations on semiconductor technology to China, but the Administration’s goals are clear: to cut off China’s access to leading-edge chip technology without unduly constraining American businesses. The Administration has to manage a difficult balancing act: restricting China’s access to advanced technology where possible without creating inadvertent incentives for global companies to de-Americanize supply chains. These new updates show that the Administration remains determined and will respond flexibly and adaptably to actions by industry, allied governments, and the People’s Republic of China. Export controls will remain a moving target, a necessity if the U.S. is to stay on top of a rapidly evolving technology. Maintaining a U.S. edge in advanced chips, which are used to train the most powerful AI systems, is absolutely essential for the United States to lead an AI-driven future.
Emily Kilcrease, Senior Fellow and Director, Energy, Economics, and Security Program:
The U.S. government has released another round of export control rules intended to slow China’s growth in the key technology areas of chips, supercomputing, and AI. These rules come nearly at the one-year mark of last year’s landmark export controls, which set a new precedent in the U.S.-China competition by using restrictive measures to maintain as large a lead as possible over China. At the time, this fundamental shift in U.S. policy was seen as aggressive and risky, in part because the United States was acting alone. What a difference a year makes. Today’s controls reflect both stronger alignment with key international partners, as well as a more sophisticated understanding of how to prevent legal circumvention of the controls.
The work is not done, however. Difficult issues, such as access to cloud computing infrastructure, remain unresolved. More fundamentally, the controls only address one part of the economic security dilemma with China. The United States needs to compete on a global basis. The controls, while justified on national security grounds, make this harder. The United States will also need to grapple with increased drive by China to dominate legacy chip production, raising a new set of national security risks that the United States is ill equipped to address. Today’s export controls are an important step forward, but the work to assure U.S. leadership is far from over.
Vivek Chilukuri, Senior Fellow and Director, Technology and National Security Program:
The technical updates to last year’s export controls shouldn’t obscure an important truth: America cannot stop China from developing or accessing advanced chips, any more than it could stop an isolated and impoverished North Korea from developing a nuclear weapon. As long as China views cutting-edge chips as foundational to its economic and national security, it will find a way. The question for the U.S. government, then, is how long it can slow China’s development, and ideally, maintain a permanent lead. This will entail a difficult and ongoing balancing act to close new loopholes and keep pace with the breakneck developments in chip design and fabrication without escalating the tech competition with China such that the costs to U.S. consumers, businesses, and allies outweigh the benefits.
The focus on export controls, however, is just one side of the equation. Slowing China’s drive for advanced chips and frontier AI is necessary but insufficient for America to maintain its tech lead. The U.S. must pay equal if not greater attention to promoting a robust, cutting-edge chip and AI ecosystem here at home. That includes accelerating domestic fabrication of cutting-edge chips; passing overdue immigration reform to attract top talent in fields like data science and electrical engineering; striking the right balance between regulation, innovation, and consumer protection; and considering support for advanced chip research and design, an area largely omitted from last year’s CHIPS and Science Act on the assumption that America was comfortably ahead. If we’ve learned anything from our tech competition with China, standing still is a surefire way to fall behind.
Thomas Krueger, Adjunct Senior Fellow, Energy, Economics, and Security Program:
The Biden administration explained the national security basis for the rule, underscoring that its purpose is to restrict China’s military modernization efforts and to “degrade” China’s ability to violate human rights. However, this time, the national security rationale was extended to include threats posed by other countries under an arms embargo. The rules highlighted China’s intent to use of these goods “for the development of Weapons of Mass Destruction, advanced AI, autonomous weapon systems, cyberweapons, hypersonics,” as well as advanced surveillance systems. Probably most significant as a rationale for these controls is that Commerce underscored the threat of China’s goal of “fielding a military by 2027 designed to deter U.S. intervention in a future cross-Strait crisis." The rules that the administration publicized today were not entirely unexpected, however, in an initial reading, there are a few interesting mentions.
First, this rule extended U.S. extraterritorial controls on certain tools that the other countries did not control on their respective lists. Given that the Biden administration prioritizes cooperation with allies, it is safe to assume that the new rule was agreed to by both the Dutch and Japanese. Perhaps the Dutch welcomed this use of U.S. extraterritorial controls in light of the pressure that Dutch would get from the CCP for imposing their own “item-based” controls on these tools. Perhaps the Japanese were pleased to see the U.S. impose these controls on the Dutch tools as well. This new “see-through” control will help level the playing field somewhat. Exporters of U.S. tools and other U.S.-origin items destined to advanced fabrication facilities are still subject to broader export control measures that the very narrow item-based controls of the Japanese and the Dutch.
Diversion, Diversion, Diversion! Commerce has now revised the country wide foreign produced direct product rule to include other countries of concern beside China to address the risk of diversion. However, the licensing policy is a “presumption of approval” for those countries that are not subject to arms embargoes, as opposed to a presumption of denial. These countries are listed in the Commerce regulations as countries of concern. Implementation of the license review will be key if this regulation will meet its objective of stopping the diversion of GPUs for uses that are contrary to U.S. national security or foreign policy. Questions of cloud customers, government end-users, and what kind of activities these end-recipients of the GPUs are engaging in could burden the license review process if not implemented correctly. In addition, these reviews can also lead to unenforceable conditions or conditions that the exporter cannot comply with; that do not serve U.S. national security or U.S. foreign policy well. Then again, it is not clear how many GPUs will be caught by these extensive rules given the new Notified Advanced Computing (NAC) licensing exemption was published that allows exporters of these chips for consumer applications.
Tim Fist, Fellow, Technology and National Security Program:
The updates to the October 7, 2022 export controls aim to prevent workarounds to the previous regulations, as well as address the risk of smuggling. The changes look like a step in the right direction, given these goals.
Expanded licensing requirements to additional countries will likely help prevent the smuggling of controlled items to China through these more indirect pathways. The updates also create a licensing requirement for the export of chips to the overseas subsidiaries/branches of Chinese firms. This addresses another potential smuggling pathway, and could also restrict Chinese cloud providers from acquiring controlled chips and making them available on the cloud to Chinese users.
For chips, the previous technical threshold that was based on a chip’s ability to communicate with other chips has been replaced with a threshold that looks at "performance density,” which means how effective a chip is at calculations relevant to machine learning, per unit of silicon area. The goal of this change is twofold: capture chips specifically relevant to AI, and address a workaround to the previous regulations where a larger number of less powerful chips could be combined to build a similarly powerful system. The methodology seems reasonable, but may still have workarounds. For example, the “silicon area” metric used in the calculation is the area of all logic on the chip, rather than just logic used for machine learning-relevant calculations. This means chips could be designed with excess silicon area to artificially lower the overall performance metric.
Bill Drexel, Associate Fellow, Technology and National Security Program:
Expanding restrictions on the tooling needed for semiconductor manufacturing is exactly the sort of response needed from the Biden administration to curb Chinese attempts to accelerate their own indigenous chip industry at the cutting edge. Updating chip restrictions is also a welcome step in ensuring the U.S. maintains its advantage in AI and supercomputing in the years ahead. But as ever, enforcement will remain a challenge as Chinese companies seek creative ways to circumvent these measures. It also remains to be seen how Washington will respond to Beijing’s widening ambitions in the ongoing chip war on other fronts, especially in terms of legacy chip dependencies and chip design competition. The Biden administration would do well to pair these updated chip restrictions with efforts to diversify our supply chains in legacy chips in critical areas, and measures to stoke innovations in chip design that would continue to give the U.S. chip ecosystem a competitive edge.
Sam Howell, Research Associate, Technology and National Security Program:
The updates to the October 7, 2022 semiconductor, artificial intelligence, and supercomputing-related export controls reflect several lessons learned over the past year. The initial controls marked a significant shift in U.S. policy vis-à-vis China and set a challenging goal—the United States signaled its intent to reign in China’s ability to invent, scale, and adapt advanced AI systems. The new updates—and the administration’s approach to unveiling them—focus on closing loopholes, mitigating the impact of second- and third-order effects, and maximizing the utility of export controls as a foreign policy tool.
The administration’s efforts to get export controls on chips right suggest that similar restrictions on other emerging technologies, such as quantum computing and biotechnology, may soon follow. Though export controls are tools to prevent adversaries’ acquisition of sensitive technology and advance U.S. security and economic interests, they are not a silver bullet solution to U.S.-China technology competition and could even be counterproductive by impeding innovation, handicapping U.S. companies, or inhibiting access to international talent. Moving forward, the administration must carefully weigh the costs and benefits of technology restrictions and work closely with industry partners to prepare for and offset adverse consequences.