Last week, U.S. President Joe Biden signed an executive order that began the process of enacting restrictions on U.S. investment in three technology sectors in China: semiconductors, quantum information technologies, and artificial intelligence. The executive order was accompanied by proposed rulemaking from the U.S. Treasury Department that would impose prohibitions and notification requirements on some investments in Chinese technologies. Although hawks in Congress pushed the administration to adopt broader controls on such investments, cooler heads prevailed, limiting the scope of the draft regulations to these three areas and prioritizing curbs on military applications of these technologies.
At the same time, even moderate restrictions on China’s technology sector will increase the temperature in the U.S.-China tech war. Biden’s order is in line with the administrations strategy of slowing the growth of China’s technology sector. This includes blacklisting of hundreds of companies and blocking the export of key technologies. Investment restrictions will likely expand, just as other U.S. sanctions programs have undergone mission creep.
Biden tasked the Treasury Department with implementing his executive order, leading it to issue an advance notice of proposed rulemaking that requests comments on how the government should limit U.S. companies’ investment in chips, quantum, and AI in China and establish notification requirements for related investments. Though headlines portrayed the executive order as enacting broad bans on investment in key technologies, Treasury’s advance notice is actually relatively narrow in scope.
With respect to semiconductors, Treasury is considering prohibiting U.S. investment in the same areas covered by the Commerce Department’s October 2022 semiconductor export controls, closing another avenue for companies to assist China in developing these sanctioned technologies. In particular, Treasury is planning to bar U.S. entities from investing in Chinese technology relating to the design, fabrication, and packaging of advanced logic and memory chips; the installation of supercomputers powered by advanced chips; and the software and machine tools required to manufacture chips.
Treasury’s proposed limitations on investment are more circumscribed when it comes to quantum information technologies and AI systems in China–two areas where the Biden administration has yet to enact sweeping export controls. The broadest of these restrictions is a potential ban on investment in the production of quantum computers and their components. Other investment prohibitions related to quantum would apply only to certain end uses: investment in Chinese quantum sensing platforms would be prohibited if they are designed for use in military, intelligence, or surveillance applications, while investment in Chinese quantum communications systems would be prohibited if they are “designed to be exclusively used for secure communications.”
Similarly, investment in the development of Chinese software that incorporates AI systems would be prohibited only if such systems are designed to be used for “military, government intelligence, or mass-surveillance end uses.” These end-use controls are important because they limit the scope of the restrictions and would not block U.S. investment in beneficial civilian technologies in China such as using machine learning to better diagnose diseases or predict the impact of the climate crisis.
In addition to these limited prohibitions on investment, Treasury has proposed notification requirements for some types of investment in chips and AI in China. This notification regime has led observers to refer to such rules as “outbound CFIUS” because they create an external complement to the Committee on Foreign Investment in the United States, which screens inbound investments. Treasury is considering requiring firms to notify the government within 30 days of making investments in the design, fabrication, and packaging of legacy chips in China, as well as for investments in AI systems that are primarily designed for cybersecurity, facial recognition, robotics, and intelligence gathering.
On the whole, Treasury’s advance notice contains a number of measures that moderate the impact of outbound investment controls. The controls would not target intracompany transfers from U.S. parent companies to existing Chinese subsidiaries, exchange-traded fund investments, or passive limited partner investments below an as-yet-undetermined threshold. Investment in non-Chinese entities would be restricted only if at least half of their business is composed of Chinese subsidiaries engaged in prohibited activities. And unlike CFIUS, the advance notice indicates that Treasury will not review transactions on a case-by-case basis, reducing the potential for every investment notification to become a veto point.
Part of the reason that Biden’s proposed outbound investment controls are limited in scope is that they have been subject to intense private-sector lobbying. The companies have been against the introduction of far-reaching regulations, as it could result in substantial legal fees and limit existing business activities. The Treasury advance notice will be a hit with lobbyists, who are expected to submit hundreds of comments before the deadline in late September.
Although previous drafts of the executive order reportedly included restrictions on investment in Chinese biotechnology and green tech, Treasury’s advance notice targets only chips, quantum, and AI. On the other hand, the advance notice includes requests for comment on additional technologies where investment should be restricted as well as an annual review to determine if outbound investment controls should be tweaked or expanded. Treasury will need to provide more details, but Biden’s order could be a model for future restrictions. It is important that Treasury does not rush into making decisions, consider other factors than national security, or get ahead of the allies.
This measured approach is a welcome departure from the type of process by which the Commerce Department enacted its October 2022 semiconductor export controls on China. Unlike those restrictions, the executive order was telegraphed ahead of time, the proposed rules stemming from it will be subject to the typical public notice and comment period before taking effect, and they focus on specific end uses of emerging technologies in China.
Members of the Biden administration have argued that the element of surprise was necessary to properly execute last year’s semiconductor export controls, but the ultimate effect was to unnecessarily alienate U.S. allies and partners. As Singapore’s foreign minister said at the time, the export controls were “all but a declaration of a technology war.”
Predictably, hawks have reacted negatively to Biden’s executive order, claiming that even unprecedented restrictions on outbound investment somehow demonstrate that Biden is soft on China. Rep. Mike Gallagher, chair of the House Select Committee on Strategic Competition Between the United States and the Chinese Communist Party (CCP), said, “The loopholes are wide enough to sail the PLA [People’s Liberation Army] Navy fleet through. … Congress needs to step up now and ensure we stop funding the CCP’s military buildup.”
Though Treasury’s advance notice proposes somewhat stronger restrictions than those contained in a measure that passed the Senate, which does not contain outright prohibitions on investment, there are indeed many loopholes in the proposal. The proposal does not cover U.S. investments in many technologies in China. It also does not require prior notification of any investment or retroactively apply to those made in advance.
But closing every loophole in U.S. curbs on China’s tech sector is a less urgent priority than many lawmakers believe. The U.S. already has extensive restrictions on various companies in China through blacklists such as the Entity List, the Unverified List, and the Non-Specially Designated Nationals Chinese Military-Industrial Complex Companies List, as well as export controls on thousands of technologies with military applications. Framing every gap in this sprawling array of constraints as a gaping loophole is misleading and counterproductive.
National security risks should not be the sole consideration when formulating U.S. international technology policy–policymakers must also weigh the economic, diplomatic, and technological fallout from restrictions on China’s tech sector. A myopic focus on ensuring than not a single U.S. dollar or chip flows to the People’s Liberation Army may be outweighed by the negative consequences of slowing global economic growth, antagonizing other countries, and increasing the risk of conflict with China.
In formulating additional outbound investment controls, the U.S. should carefully assess such costs, which may be serious. Casting a wide net could affect a large range of technology companies in allied countries, making U.S. allies less likely to adopt their own outbound investment controls. Preventing U.S. investors from acquiring equity in Chinese technology companies–which allows investors to learn about Chinese business practices–could slow U.S. technological development in areas where China is ahead of the U.S., such as batteries and telecommunications. At the extreme end, the International Monetary Fund projected this year that severe decoupling in foreign direct investment resulting in two blocs centered on the U.S. and China would reduce long-term global economic output by 2 percent, or $2 trillion annually using 2022 as a baseline.
Outbound investment controls on quantum information technologies may be much less impactful than restrictions on investment in semiconductors and AI in China. There is no evidence of any U.S. investment in top Chinese quantum startups such as QuantumCTek, CIQTEK, or Origin Quantum, and the majority of China’s work on quantum is done by government-funded labs. Additionally, U.S. providers of cryogenic technologies for quantum computing do not appear to partner with Chinese entities to sell their products into China, unlike European competitors such as Bluefors and Leiden Cryogenics.
Outbound investment controls on quantum are also less significant because there are relatively few applications of this early-stage technology. Despite years of effort, there are no practical applications of quantum computers. Biden’s executive order states that quantum information technologies could aid in the “breaking of cryptographic codes,” but there is scant evidence that this is imminent. The National Security Agency has said that it “does not know when or even if a quantum computer of sufficient size and power to exploit public key cryptography … will exist,” and the National Academy of Sciences estimates this will not occur before 2030. Moreover, though the Pentagon is keenly interested in the potential for quantum sensors to identify stealth aircraft, even the Office of the Undersecretary of Defense for Research and Engineering has admitted that “quantum radar” is impractical.
The primary impact of restrictions on U.S. investment in Chinese quantum information technologies may be that they foreshadow complementary quantum export controls on China. One could say that U.S. restrictions on investment are prudent, as they would help advance Chinese quantum technologies with military applications. However, it is also possible to argue that these controls are premature due to the fact that U.S. companies have not invested in Chinese firms developing this nascent technology.
One specific way in which the Biden administration may alter outbound investment controls is by limiting investment in large AI models. Many national security analysts have criticized the proposed restrictions on AI systems as too narrow because they apply only to specific end uses, potentially exempting many large AI models with more general capabilities. Some analysts argue that large language models such as OpenAI’s GPT-4 or Baidu’s Ernie are worthy of scrutiny as they are “dual-use” technologies that could be used in China for both civilian and military purposes.
As in the U.S., AI hype has intensified in China in recent months as Chinese companies are developing dozens of large language models. A handful of Chinese large language models perform moderately well on some industry benchmarks, but there are substantial barriers preventing China from catching up to the U.S. in the near term. Major hurdles include China having fewer top-tier engineers of large AI models and associated compute clusters, more heavy-handed regulation, slower rollouts of user-facing large language models, and having to contend with U.S. export controls on advanced graphics processing units and semiconductor manufacturing equipment.
Efforts to block investment in the development of large AI models in China–or, for that matter, the export of related technologies and services to China–are fraught. Large AI models are a general-purpose technology that have the potential to significantly alter economies by, for instance, automating a large number of basic job-related tasks. Stalling China’s ability to develop such models could seriously impact its economy and should be approached with caution.
The scope of U.S. investment in Chinese AI companies is not completely clear. According to the Center for Security and Emerging Technology, U.S. investors participated in one-third of investment deals involving Chinese AI companies from 2015 to 2021, though the percentage of those investments that came from U.S. investors is unknown.
The Biden administration’s restrictions on investment in AI systems in China are significant not because they will starve Chinese AI startups of U.S. funding, but because they set the stage for U.S. export controls on AI. As both Commerce Secretary Gina Raimondo and Under Secretary of Commerce for Industry and Security Alan Estevez have stated, the U.S. intends to enact AI export controls on China; the comments that Treasury receives on its advance notice will help the Commerce Department determine the scope of those controls.
Some AI researchers have argued that the U.S. could reasonably use export controls and other tools to block Chinese companies’ ability to release AI models above a certain capability threshold, thereby reducing the severity of such measures. For instance, the amount of compute used to train an AI model, unlike the size of a model or its anticipated capabilities, is a sufficiently good predictor of a model’s performance that the U.S. could adopt restrictions on this basis. Beyond efforts to slow China’s development of home grown large language models, Washington could attempt to prevent Chinese cloud providers from deploying compute-intensive AI models built in the U.S., as Alibaba and Baidu have done with Meta’s LLaMa 2, an open-source model.
The fundamental issue with this proposal is that large language models are civilian technologies that have yet to be applied in military settings. It is likely that these models will eventually be used for weapons design and strategic decision-making, but such applications are still in the R&D phase, and there is currently no candidate for a “killer app” for military uses. In any case, the export of large language models for many military applications, such as the production of defense articles or biological weapons, is already prohibited under the Commerce Department’s existing export controls and the State Department’s International Traffic in Arms Regulations.
Restrictions on China’s development of large AI models would also be thorny to implement. They would likely require a licensing regime for such models, which would necessitate mass surveillance to enforce and would reduce competition in the AI industry by raising barriers to entry. A poorly crafted threshold would either be unsuccessful in preventing China from developing powerful AI models–especially given the prevalence of open-source models–or be overbroad and slow China’s development of less capable AI systems. Though expansive restrictions on China’s technology development may be appealing at first glance, they are likely to provoke substantial blowback from China.
The limited nature of Treasury’s advance notice reflects restraint on the part of the Biden administration, though China is unlikely to interpret it as such. In recent months, Beijing has blocked Intel’s acquisition of the Israeli chipmaker Tower; limited procurement of memory chips from the U.S. chipmaker Micron; and restricted exports of gallium- and germanium-based compounds used in semiconductors. These moves have helped further escalate the tech war, and Washington is now considering restricting U.S. companies’ ability to provide cloud services to Chinese firms that use U.S. cloud infrastructure to train AI models.
Additional AI or quantum export controls reminiscent of semiconductor restrictions enacted by the U.S. in October 2022 are likely to incite a strong response from Beijing. Perhaps the most likely form of retaliation by China would be an enlargement of existing countermeasures, such as export controls on additional inputs for semiconductor manufacturing.
These tit-for-tat measures make it more difficult to diffuse tensions and decrease the likelihood of conflict between the U.S. and China. Wang Yi, China’s top diplomat told U.S. secretary of state Antony Blinken, during his visit to Beijing, that U.S. sanctions on China’s tech sector were a major impediment in improving U.S. – China relations. Wang demanded that the U.S. “lift illegal unilateral sanctions against China, stop suppressing China’s scientific and technological advances, and not wantonly interfere in China’s internal affairs.” While the Biden administration has crafted a measured approach to outbound investment controls, it remains unclear if it will be able to avoid a dangerous escalatory spiral in the tech war.
Outbound investment controls are unlikely to seriously damage China’s economy when they ultimately take effect. As Rhodium Group has found, U.S. investment in Chinese tech companies has dwindled since the beginning of the Trump administration’s trade war, meaning relatively few transactions related to semiconductors, quantum, or AI will be blocked by the Treasury Department. But if other countries get on board and U.S. restrictions expand over time, these new controls could have a significant impact on China.
The EU is on track to follow Washington’s lead and propose restrictions on outbound investment later this year, though the breadth of such measures remains unclear. The European Commission’s Economic Security Strategy states that Europe should adopt outbound investment controls because it has “a common interest in preventing the narrow set of technological advances that are assessed to be core to enhancing military and intelligence capabilities of actors who may use them to undermine international peace and security from being fuelled by our companies’ capital, expertise and knowledge.”
Germany, which accounts for roughly half of EU foreign direct investment in China, voiced support for such restrictions in its new China strategy, and the United Kingdom is also considering outbound investment controls. But business interests will likely prevent most U.S. allies from restricting outbound investment, as multinationals in allied countries are eager to profit from investments in China. In May, the U.S. was able to persuade the G-7 to a make only a mild statement recognizing the potential importance of future outbound investment controls. It is possible, however, that other countries will enact similar restrictions in response to U.S. arm-twisting. Many analysts anticipated that the U.S. would struggle to multilateralize semiconductor exports controls, but the Biden administration succeeded through high-level talks and substantial economic pressure.
New restrictions on capital flows into China are the last thing Beijing needs as it strains to boost consumer and investor confidence while its post-COVID recovery sputters. By one metric, foreign direct investment into China has fallen to its lowest level in 25 years, meaning that even a slight decline in investment due to new restrictions could have a disproportionate impact.
The Biden administration has done well to craft a measured approach to outbound investment controls on China. Yet even modest restrictions risk escalating the tech war and stymying diplomacy between Washington and Beijing. This delicate balance is unlikely to hold for long.
The post Biden Takes Measured Approach on China Investment Controls appeared first on Foreign Policy.