[US-China Economic War and Korea's Choice Series] ⑥ Military Artificial Intelligence in the Context of US-China Strategic Competition

I. Opportunities and Risks Presented by Artificial Intelligence

Among the technologies of the Fourth Industrial Revolution, Artificial Intelligence (AI) is a meta-technology that underpins all technologies across a wide range of fields and will undoubtedly bring about tremendous changes in the future. AI, which has already brought about revolutionary changes in both military and commercial technologies, is rapidly developing at an unprecedented pace, and the direction of technological development is unpredictable. When the variable of geopolitical competition between great powers is added to the unpredictable extent of AI technology development, it is difficult to gauge whether AI technology will be a blessing or a curse for humanity.

AI is bringing about tremendous changes across all domains of the military. It is a key technological area, along with quantum computing, biotechnology, space, and cyber technologies, that will determine the future landscape of warfare. The United States and China are engaged in fierce strategic competition and are striving to lead the AI technology innovation race to steer future warfare in a favorable direction. While the US is currently assessed to be ahead of China in existing military domains and various AI areas, China has rapidly advanced its military technology in a short period and is making efforts to surpass the US in the future by investing heavily in intelligent warfare utilizing AI (Luo 2022). The issue of AI technology is closely linked to the economic relationship between the US and China and will inevitably be one of the variables that determine the future landscape of US-China strategic competition, based on the military and security domain.

Meanwhile, if international cooperation is not achieved regarding the future development direction of AI, AI could bring great harm to humanity (Bremmer and Suleyman 2023, 102-105). Unlike other fields, AI technology cannot be unilaterally led by governments and militaries; many companies and economic actors are at the forefront of AI development. Rapidly developing AI technology is difficult to control solely through agreements between nations, and knowledge about AI technology itself is still insufficient. For example, there is still insufficient knowledge about whether AI will reach artificial general intelligence (AGI) and gain self-controlling power beyond human control, or whether its subjectivity and consciousness will require a fundamental shift in our perspective on AI.

Regulation of AI also requires different resources and agreements compared to past efforts to develop regulations and controls in areas such as nuclear weapons or climate change, due to differing national interests and the distinct positions of governments and corporations. Although rudimentary efforts are underway to create the necessary regulatory regimes in this new domain, if great power geopolitical competition, such as the US-China AI competition, operates as the driving logic, AI could pose a significant challenge to all of humanity (Kissinger and Allison 2023).

South Korea, caught in the midst of US-China geopolitical competition, is experiencing significant changes not only in military competition between the two countries but also in economic areas such as export controls and investment restrictions to gain an edge in the AI competition. For South Korea, which holds a certain importance in semiconductor production, how to predict and respond to future situations is a crucial issue.

II. Artificial Intelligence and Future Warfare

Future warfare will be characterized by the use of AI-enhanced technologies, particularly fully autonomous weapon systems, and the side that preempts this will gain an advantageous position not only in military security but also in overall strategic competition. The development of military AI is progressing at a very rapid pace; capabilities such as the US Air Force's 'Loyal Wingman' unmanned aerial vehicles or drones can now identify, track, and attack targets without human intervention. The recent use of lethal autonomous weapon systems in conflict zones such as Gaza, Libya, Nagorno-Karabakh, and Ukraine has also raised significant legal, ethical, and moral issues. It remains unclear how AI-enhanced military technologies will alter the nature and dynamics of warfare. Some of the most concerned individuals about the military use of AI even discuss a dystopian future where machines develop to the point of dominating the world, an 'AI apocalypse'.

The extent to which AI will shape future warfare, and for what purposes, will significantly influence future international politics. The adoption of AI-enhanced military technologies can vary depending on the level of decision-making (tactical or strategic) and the type of human involvement (human or machine). Each country will optimize algorithms to perform tactical operations on the battlefield or to carry out strategic decision-making processes to support overall war objectives.

Tactically, these technologies can enhance the success rate of field commanders by rapidly analyzing vast amounts of data collected from sensors dispersed on the battlefield to generate targets faster than the enemy. This can be achieved by significantly reducing the 'sensor-to-shooter time,' the time interval between target acquisition and attack. The US Department of Defense's Task Force Lima and Project Maven are examples of such AI applications.

At the strategic level, AI-enhanced military technologies can assist political and military leaders in setting major goals and objectives. That is, it helps to synchronize the approach to warfare with the combination and synchronization of limited resources, including materiel and personnel. New capabilities may even emerge that can replace humans in future military operations, such as formulating strategic direction and national-level strategies.

Classified by the type of human involvement and oversight, nations can adjust the type of delegated oversight or control for AI-enhanced military technologies. This involves designing systems to enhance human oversight of technology, thereby strengthening the authority over decision-making. Such systems are often referred to as semi-autonomous, meaning they are under human control. Most US weapon systems, such as the General Atomics MQ-9 Reaper drone, operate in this manner with AI-enhanced weapon systems.

Nations may also design AI-enhanced military technologies with reduced human oversight. These systems are non-human-in-the-loop, with killer robots being a prime example. In these applications, humans exercise only limited oversight over target selection.

Summarizing the above based on the application of AI in military security, warfare can be categorized into four types depending on the decision-making level at which AI operates and the type of human oversight. First, nations can use AI-enhanced military technologies for tactical decision-making under human supervision. Second, countries can use AI-enhanced military technologies for tactical decision-making with minimal human involvement and machine supervision. Third, this is a type that maximizes machine control at the strategic decision-making level, granting significant authority to AI-enhanced military technologies to shape the trajectory of warfare between nations. Warfare conducted using AI will enable nations to gain and maintain an advantage over the enemy in time and space, thereby determining the overall outcome of the war. Finally, this involves maintaining human oversight of AI-enhanced military technologies while optimizing strategic decision-making using algorithms to attack and exploit the vulnerabilities of the adversary. The objective of this warfare model, called algorithmic decision support systems, can be achieved through purposes such as predicting possible enemy courses of action through real-time threat prediction, identifying the most feasible, acceptable, and appropriate strategies, and customizing key warfighting functions like logistics to ensure that forces maintain the initiative in a competitive operational environment (Lushenko 2023).

AI will be used at almost every level of military security. It is already being used to a significant extent and will continue to be used in the future, from administrative tasks such as logistics and procurement to strategic decisions in warfare. The political system, whether democracy or authoritarianism, is also an important variable. The operational environment of military AI will change depending on the degree of human intervention allowed and the role of civil society in AI operations. Since military AI is a matter of war, and the outcome of war determines the fate of a nation, the efficiency of warfare may be prioritized over ethical and moral responsibilities in AI operations. In the absence of international agreements or norms on AI operations to date, the operation of AI by individual countries is highly likely to be competitive.

III. US-China Competition over Military Artificial Intelligence

1. China's Strategy for Military Artificial Intelligence Development

The strategic competition between the US and China has expanded beyond the economic sphere into military security. In military competition, the US leads China in conventional weapons, nuclear power, and AI capabilities. Based on this superiority, the US's strategic goal towards China is to maintain overwhelming military superiority over China, prevent strategic competition from escalating into military confrontation, maintain areas of cooperation in the economic sector, solidify the ideological foundation based on the superiority of liberal democracy, and strengthen relationships with allies and strategic partners (Krepinevich 2024, 103-111). On the other hand, China's pursuit in conventional power is progressing very rapidly, and particularly in AI capabilities, China is making all-out efforts to create an opportunity to overwhelm the US (Flournoy 2023, 102-106).

In 2017, China announced its goal to develop the People's Liberation Army (PLA) into a world-class military by the mid-21st century, thereby overcoming its military capability deficiencies and firmly establishing China among the world's major military powers. This goal reflects the view of Chinese Communist Party (CCP) leaders that China is approaching center stage in the world and represents the military component of a multifaceted objective to establish China's leading global position in all critical elements of national power. China sees a world-class PLA as capable of surpassing other militaries, particularly the U.S. military, in power and prestige and deterring other countries from resisting China's pursuit of its national objectives.

Given the advantages of being a latecomer, China may be in a more favorable position than the United States to adopt new technologies, such as military AI. The current U.S. advantages in areas like stealth aircraft, aircraft carriers, and precision munitions may prove to be a disadvantage in the long run, as the entrenched business and political interests that underpin today's military superiority could hinder the U.S. transition to an AI-driven military technology paradigm. China has tripled its military spending from 2007 to 2017, prioritizes technology, and generally recognizes that many of its current platforms and approaches are obsolete and must be replaced. While the U.S. still outspends China on defense, a significant portion of that spending is tied up in legacy programs. The existence of these legacy programs works to its disadvantage compared to investing in new approaches built from the ground up (Huang and Drexel 2023).

China recognizes the technological advancement of its military capabilities as an essential component of becoming a world-class military. It has been noted that to succeed in future conflicts, the PLA must enhance its ability to win "informatized and intelligentized warfare" by paying attention to technological changes alongside its transformation into a world-class military. China views AI as the foundation for the future of its economic and military power and, in July 2017, the State Council released the Next Generation Artificial Intelligence Development Plan (AIDP). Chinese military leaders believe that the advent of intelligentized warfare utilizing AI signifies a military technological revolution comparable to the mechanization and information revolutions of the 20th century. The use of AI in weapons systems has been at the center of Chinese military reforms in recent years and is certain to be a major aspect of building a world-class military.

The CCP leadership views AI as a breakthrough technology with the potential to rapidly enhance various warfighting capabilities—such as reconnaissance, data processing, and targeting—beyond human capacity. Both Chinese military leaders and AI engineers recognize the inevitability of AI's application in warfare and believe that early adoption of AI for military applications offers an opportunity for the PLA to "leapfrog" U.S. military capabilities.

Over the past few decades, the Chinese military has honed asymmetric capabilities and strategies for use in warfare leveraging modern information technologies. It is currently researching methods to gain decisive advantages in wartime decision-making and processes using AI and other technologies. Military leaders have reportedly been developing a concept of "hybrid intelligence," which blends human and machine intelligence through technologies such as the use of brain-computer interfaces. This concept can be seen as being realized through new programs, including projects to promote human performance enhancement, such as the use of "intelligent autonomy" in weapons that exercise command authority through brain-machine integration. In addition to using AI in autonomous military robotics, China also appears to be interested in AI capabilities for military command decision-making. These are grounds to suggest that China's strategy is an ambitious plan moving toward increasingly autonomous AI-driven warfare beyond human oversight on the battlefield. It is a direction that reduces the degree of human intervention to maximize the military application of AI in the future of warfare, as examined above.

In line with its strong interest in AI, the CCP is also increasing its investment in AI. To become an AI leader, China is reportedly increasing its total government spending on AI development by 27% annually, with plans to reach $27 billion by 2026. While China's private AI companies are not state-owned enterprises in name, the CCP operates influential CCP committees within many of them.

Although China was subject to Western imperialism during its modernization process, it is demonstrating its will to take the lead in the future global order by gaining an advantage in the Fourth Industrial Revolution. In this process, it is dedicating its efforts to solidifying the political and economic foundations for military AI, as well as AI in general.

2. China's Military-Civil Fusion Strategy and Its Strengths and Weaknesses

It is well known that artificial intelligence technology develops within an ecosystem that combines knowledge from government, military, and commercial sectors, as well as universities and research institutes. The United States has leveraged its government efforts, along with the innovative technological environment of Silicon Valley and AI research from major universities, to create the world's strongest AI innovation ecosystem. China is also accelerating AI development tailored to its authoritarian system (Scharre 2023).

China's defense technology innovation increasingly relies on contributions from civilian companies and universities under its Military-Civil Fusion (MCF) strategy. The MCF strategy is a crucial element of China's agenda to catch up with and surpass the United States, particularly in domains such as space, cyberspace, and the deep sea, as well as in strategic technologies like artificial intelligence and quantum information science (Muhammed and Vieira 2022, 85-102).

Recent PLA procurement contracts indicate that most AI equipment suppliers are non-governmental Chinese technology companies established after 2010. This includes Anwise Global Technologies (北京安怀信科技股份有限公司), founded in 2016, which primarily serves the military aerospace and electronics industries and has grown into China's largest intelligent equipment manufacturer. Realis (Shenzhen Realis Multimedia Technology Co., Ltd.), an AI company founded in 2015, developed an AI-powered virtual reality training room for PLA personnel capable of multi-person training. The PLA Strategic Support Force (PLASSF) is particularly well-positioned to seek AI partnerships as it strives to perform a portfolio of missions with high AI applicability, including algorithm development, satellite constellation management, and electronic warfare for potential attacks.

State-backed investors are also eager to provide funding. China's forward-looking surveillance systems provide vast datasets for early AI companies collaborating with the government to experiment with and develop technologies. This supports China's growth as a global leader in AI applications. This has allowed the Chinese government to gain experience in managing AI development and has accelerated innovation in specific AI fields, such as computer vision, which enables the collection and analysis of information from image and video data. The Chinese government considers computer vision crucial for both surveillance and military applications. China strongly supports computer vision research, with researchers from Chinese institutions publishing over one-third of the papers on computer vision and visual surveillance research, making China the leading country in terms of research output on computer vision and its government applications (Fedasiuk et al. 2022).

Commercial AI companies are also collaborating with the Chinese military, sometimes leveraging their influence in the global market. DJI, a drone manufacturer, has applied machine learning tools to object detection and navigation to achieve a 76% global market share for commercial drones.

As Chinese military and non-governmental AI companies collaborate on AI development, evading international sanctions is also a significant aspect. Many Chinese companies continue to operate as civilian non-governmental technology firms, avoiding the scrutiny and sanctions associated with supporting adversarial militaries. Of the 273 Chinese PLA AI equipment suppliers identified in a study by the Center for Security and Emerging Technology, only 8% (22 companies) were designated under U.S. export control and sanctions regimes as of 2021. Many of these companies have relied on U.S. technological advancements and some U.S.-based funding in their development processes.

AI technology requires semiconductors, and many components of the semiconductor ecosystem are controlled by the United States and its partners. Of the 97 AI chips identified by CSET in recent PLA public procurement records in 2020, almost all were designed by U.S.-based chip companies such as NVIDIA, Xilinx (now part of AMD), Intel, and Microsemi. Almost all AI models are trained on Graphics Processing Units (GPUs), which are highly capable chips that can train sophisticated AI models. As of September 2022, the two U.S. GPU suppliers, NVIDIA and AMD, accounted for 95% of the Chinese domestic GPU market and are known to supply chips essential for China's AI development, including for military use.

As will be discussed later, China's AI development has shown a tendency to slow down following the implementation of U.S. export controls on semiconductors targeting China by the Biden administration in October 2022, restricting China's access to advanced U.S. chips. The introduction of U.S. restrictions has led many Chinese companies to expand their sanctions evasion activities, resulting in the smuggling of top-tier semiconductors, including those from NVIDIA, into China through thousands of intermediaries.

Meanwhile, capital from the United States has also fueled the development of Chinese AI military technology companies. This includes funding linked to prominent U.S. venture capital funds, such as Sequoia Capital China, which was affiliated with Silicon Valley venture capital firm Sequoia Capital. Although Sequoia Capital China is in the process of separating its China operations from its U.S. and European businesses and rebranding by March 2024, it continues to attract investments from U.S. university endowments and charitable trusts. Sequoia Capital China was an early investor in EverSec, which provides AI-based open-source data mining and information technology support to the PLA. In November 2021, the PLA Strategic Support Force contracted with EverSec for an AI-based "cyber threat intelligent detection and early warning platform." Additionally, Goldman Sachs invested in 4Paradigm, one of China's largest AI companies, in 2020.

In the domain of AI Large Language Models (LLMs), Chinese companies are aggressively recruiting international AI scientists to enhance China's AI LLM capabilities. China's AI-based defense technology development has been hampered by limited access to training data for specific war scenarios and a shortage of AI engineers. China's computer vision development has been expanded, partly based on China's nationwide surveillance programs, providing AI companies with millions of use cases to develop and test the operational use of AI computer vision technologies.

In summary, China is waging a total national effort to develop military AI through long-term planning. In this process, the technology and semiconductor imports from competitors, particularly the United States, have been crucial, and China is doing its utmost to develop its own AI capabilities while maintaining these imports. It remains to be seen whether China can maintain its position as a high-speed pursuer amidst increasing AI sanctions. China possesses the foundation to develop AI in various fields; particularly, vast data under an authoritarian system will greatly aid AI development. However, limiting the scope to the military sector, the innovative field data for military AI development is indeed limited. The fact that China rarely engages in overseas conflicts to directly test AI means that the data available for developing, training, and improving AI-enabled warfighting capabilities is limited, which is one of the limitations China faces in the U.S.-China military security competition.

The Chinese government has installed extensive AI-based systems to monitor, censor, and restrict the behavior of Xinjiang residents, providing ample funding, data, and operational experience to China's surveillance industry complex, including many companies at the forefront of Chinese AI development. However, the utility of the data is inherently limited depending on the application. Generally, it is far easier to obtain training data on commercial customers or domestic surveillance targets than on adversaries, especially when friendly weapon systems and sensors are not within the adversary's range. The most mature U.S. national security AI applications are in areas such as AI-based analysis of satellite reconnaissance imagery. Even in peacetime, satellites capture numerous photographs of Russian and Chinese military forces, and these photographs can be digitally labeled by human experts to be converted into training data. Training data is the data on which machine learning AI systems learn, and the combination of training algorithms and training data teaches the AI system how to recognize the content of images.

While China may have a data advantage in facial recognition for domestic surveillance or commercial applications like consumer finance, the data relevant to actual military operations is limited in its applicability to military applications. For some military AI applications, such as precision missile targeting or autonomous drone navigation, China has a much lower data advantage compared to the United States.

IV. The Biden Administration's Strategy to Control China's Artificial Intelligence Technology

1. The U.S. Strategy of Technology Control Towards China in Advanced Technology Sectors

The competition between the U.S. and China to lead in military AI innovation and gain an advantage in future warfare is intense, but the competition is already fiercely unfolding across the entire spectrum of economic policy. Shortly after taking office, the Xi Jinping administration announced the "Made in China 2025" plan, seeking to re-establish China as a leading science and technology nation. Although China has not surpassed the United States in technology, science, and advanced industries, it has shown prominence in batteries, solar power, quantum communication, and artificial intelligence. These advancements stem not from China's scientific innovation ecosystem but from innovations based on its manufacturing processes. The Chinese government has fostered intense competition among companies in future key industrial sectors, such as the solar power industry, enabling them to achieve international competitiveness through technological innovation and cost reduction. This innovation is expected to continue, and China has the potential to excel in cutting-edge technologies and space industries, such as lunar unmanned exploration landings (Wang 2023).

The United States has sought to maintain its advantage through technology controls on China. The outcomes of U.S. export controls and investment restrictions on China are critical issues. The U.S. views China as pursuing economic development by illicitly stealing core scientific and manufacturing technologies from U.S. and Western companies or by forcing technology transfers. It believes that restricting these unfair economic practices and controlling China's technological innovation is the path to gaining an advantage in the future U.S.-China geopolitical competition (Horowitz et al. 2022, 101-103).

However, it is difficult to view China's development solely as a result of adopting U.S. and Western technologies. While it is true that the U.S. and the West have led in science, knowledge, and technological innovation, thereby maintaining their economic advantage, it is crucial to focus on the fact that China has developed its own economic development strategies. For example, Apple decided to produce components in China starting in 2007, initially relying on Chinese labor for simple component assembly. Gradually, the role of Chinese workers expanded to the production and assembly of more complex components. Chinese companies, which initially contributed very little to the value chain, now account for nearly a quarter of the value.

It is true that Chinese government subsidies and support policies for businesses have played a significant role in this economic development. However, the knowledge and management systems acquired during mass manufacturing processes, along with the effects of intense competition, are also important factors. Chinese companies have leveraged low-cost labor and maximized management and knowledge systems learned from the West to create a unique manufacturing-based innovation system in China. This innovation is the result of a long process and competition, which is in stark contrast to the advancements made by the United States in its scientific and knowledge innovation ecosystem. This demonstrates the possibility of China's independent AI development under U.S. sanctions in the future (Danzman and Kilcrease 2022).

2. U.S. Strategy of Technology Control Towards China in the Field of Artificial Intelligence

Advancing in the field of AI requires capabilities in scientific research, research personnel, government and military resources and efficient strategies, innovation and corporate performance in the private sector, and semiconductor production capacity, which is the hardware foundation of AI technology. The U.S. and China possess different strengths in their AI competition, with the U.S. increasingly strengthening its strategy of pressuring China, particularly in the semiconductor sector.

The first semiconductors, invented in the 1950s and 1960s for use in missile guidance systems, now power all consumer electronics, but their origins can be traced to defense systems. Semiconductors are critical in warfare, as seen in missiles like the Javelin anti-tank missile, which has been highly effective in Ukraine's war against Russia, and the Starlink low-earth orbit satellite network, which has helped keep Ukraine's communications online. In the future, the military will require far more semiconductors than ever before, and like the civilian economy, AI applications will face the fundamental challenge of requiring immense computing power.

To curb China's military AI revolution, the U.S. must not only block U.S. semiconductor sales but also prevent China from producing its own or acquiring alternatives. To block China's alternative pathways, the Biden administration is leveraging U.S. technological superiority in key nodes of the global semiconductor supply chain, blocking China's access to semiconductor design software, manufacturing equipment, and equipment components—all areas where U.S. technology is critical and nearly irreplaceable.

China relies on imported semiconductors, spending as much annually as it does on oil imports, and must import them from countries that are not only commercial competitors but also geopolitical adversaries. Considering that ASML in the Netherlands plans to release its next-generation EUV lithography equipment within the next one to two years, catching up in the semiconductor industry means not just keeping pace with the status quo but competing at the fastest pace in human history. Currently, only a few companies in Japan are capable of producing the necessary chemicals for advanced materials, and three companies hold monopolistic influence in semiconductor design software production, with similar concentration in manufacturing.

It is well known that China cannot meet its domestic demand for AI semiconductors through domestic production. There is a severe shortage of the semiconductors used to train current AI systems, with NVIDIA reportedly producing almost 90% of the supply. These chips are believed to be training AI systems like OpenAI's, which are absolutely critical to the future of AI and are currently valued at $1 trillion.

In this context, on October 7, 2022, the Biden administration, following the August CHIPS and Science Act, announced its strategy to curb China's semiconductor industry. The U.S. regulations prohibit the export of semiconductor equipment and software capable of producing DRAM below 18 nanometers, NAND flash above 128 layers, and logic chips (non-memory chips) below 14 nanometers to China. The U.S. has prevented the sale of powerful AI training chips needed by companies to build the most powerful computer clusters at the fastest speeds for AI. This regulation means that NVIDIA, the world's largest chip manufacturer, cannot sell its most powerful AI training chips, the H100 and A100, to Chinese companies.

NVIDIA, with over 30% of its total revenue generated from the Chinese market, quickly developed alternative products, the H800 and A800, which can communicate with other chips within a cluster at 400 gigabytes per second instead of the 600 gigabytes per second threshold set by previous regulations, thereby circumventing U.S. controls. While slower than the cutting-edge chips, the H800 and A800 are still useful for building powerful AI applications. Chinese tech companies, including ByteDance (owner of TikTok), Baidu, Alibaba, and Tencent, have reportedly ordered $5 billion worth of H800 chips following the announcement of the regulations.

A year later, on August 9, 2023, the Biden administration issued an executive order notifying, and in some cases prohibiting, U.S. persons from making certain investments in China related to semiconductors and microelectronics, quantum information technology, and AI systems. This allows the U.S. to pursue export controls on the A800, discussed earlier. The U.S. government has also restricted NVIDIA's exports to certain Middle Eastern countries, citing close collaboration with China in key technology development. Another target is the cloud services market, where the U.S. government is considering measures to prevent Chinese companies from accessing Amazon Web Services (AWS) and Microsoft Azure, which offer powerful computing capabilities anywhere. This is due to concerns that these services could be used to circumvent export control measures, and furthermore, existing export regulations on semiconductor equipment are also reportedly being updated.

V. The Future of the U.S. Strategy to Counter China's Artificial Intelligence

Semiconductors represent one of the most complex supply chains in the world, requiring a multilateral response. Several important features must be considered for the success of U.S. quantum policy toward China, necessitating the establishment of a multilateral framework. First, Europe's pivotal role in semiconductor trade. One of China's main obstacles in advancing semiconductors is access to advanced photolithography technology and testing equipment, such as ASML's in the Netherlands, and U.S. negotiations regarding trade regulations with the Netherlands required significant pressure. However, such pressure may not be feasible in the future. As economic ties between China and Europe, particularly Germany, deepen, large-scale trade relationships may lead Europe to re-evaluate its alignment with U.S. priorities. Changes in U.S. security commitments in Europe, such as troop reductions in Ukraine or a reduced role in NATO, hinted at by candidates in the lead-up to the 2024 U.S. presidential election, could further accelerate these shifts.

Second, the technological superiority of U.S. and Taiwanese semiconductor production, especially cutting-edge semiconductor manufacturing, is a critical factor in the global balance of power. As the ability of advanced manufacturers to increase transistor density, known as Moore's Law, slows down, it presents an opportunity for China to narrow the gap. Although the West is known to be more than a decade ahead, future breakthroughs could alter the dynamics. China's companies, as innovators, continue to advance, as evidenced by the 7-nanometer chip featured in Huawei's new phone.

Third, regarding the feasibility of next-generation technologies, while Samsung, Intel, and TSMC are leading in next-generation 3D semiconductor technology, Chinese competitors like SMIC are indeed skilled pursuers. 3D semiconductors are the most complex devices in history, and while there is no guarantee that companies outside of China can successfully scale production, if technological advancement slows due to complexity and increased costs, it could become a more achievable goal for government-backed Chinese firms.

Fourth, concerning clean energy production elements, given the immense energy demand for charging capabilities and cooling required for artificial intelligence, the United States and its allies might find it difficult to power these data centers without China's support. China produces three-quarters of the world's solar panels and was the first to bring fourth-generation nuclear technology to market.

Finally, regarding the resource equation, the United States is heavily reliant on Chinese companies for essential materials in semiconductor production, such as gallium and germanium. Major U.S. companies estimate that decoupling from these resources will take at least another decade. This poses a potential bottleneck for U.S. semiconductor production, as China has already begun implementing export controls on key materials (Brill 2023).

In this context, as the Biden administration expands its national security trade and investment tools, it is crucial whether allies will embrace this agenda. To date, U.S. allies and strategic partner countries have largely aligned with the U.S. economic strategy toward China. Following its unilateral export control announcements, the U.S. has forged alliances with countries included in the semiconductor task force 'FAB4,' such as Japan, the Netherlands, and South Korea. European Commission President Ursula von der Leyen has also urged the European Union to introduce corresponding measures to support the U.S. Japan, a powerhouse in semiconductor equipment and materials, announced export control measures on 23 items, including advanced semiconductor lithography and cleaning equipment, effective July 2023. The Netherlands also previously announced strengthened export controls. The U.S.-China semiconductor dispute intensified particularly in May 2023 when Chinese authorities banned purchases of U.S. company Micron's semiconductor products, citing security risks. As the U.S. restricted exports of advanced semiconductor equipment and added China's state-owned semiconductor firm YMTC to its export control list last year, China retaliated by initiating export controls on 30 items, including gallium and germanium, leading to tensions between the two countries over semiconductors.

The United States attempted to prioritize overseas investment screening at the G7 Summit in Hiroshima in May 2023. According to the leaders' statement, "We recognize that appropriate measures designed to address risks from outbound investment may be important to complement existing targeted controls on exports and outbound investment, and that we can work together to protect our sensitive technologies from being used in ways that threaten international peace and security." The European Union's economic security strategy, released in June 2023, included measures such as preventing excessive investment in EU companies possessing sensitive technologies, controlling exports of goods that could be used militarily, and preventing the acquisition of key EU infrastructure or companies by third countries.

Multilateral frameworks to curb China's AI development will continue to strengthen in the future. Simultaneously, China will accelerate its domestic production while seeking to circumvent sanctions for its own AI development. In this scenario, questions will arise regarding which technologies should be controlled, at what level of maturity they should be controlled, and whether consensus can be continuously reached with allies and partners. Coordination of control measures with allies and partners is not always straightforward, and for many mature technologies and their supply chains, policy coordination among a broader range of countries may be necessary to effectively slow China's acquisition. For emerging technologies, coordination issues become more complex as it is unclear how related industries will evolve, which countries possess the greatest capabilities or obstacles in that technology, and at what stage technology maturity controls should be implemented, as commercial and potential defense applications develop.

For example, key East Asian countries such as South Korea, Japan, and Taiwan maintain their own screening systems, and each jurisdiction preserves a robust and highly globalized technological ecosystem. Nevertheless, the system promoted by the United States is quite unique, and engaging allies in this type of regime will require concerted efforts, politically intensive processes, and significant time investment from leaders.

A larger issue is whether current U.S. export controls and investment restrictions, even when combined with multilateral export controls and the investment screening systems of U.S. allies and partners, are sufficient to prevent the inflow of foreign technology, expertise, and capital into China's defense sector. China's military-civil fusion strategy poses many challenges to maintaining export controls, and control and restriction policies are becoming more complex due to the rapid pace of technological advancement and the increasing globalization of research and development for new technologies and their supply chains (O’Dea 2023).

Chinese companies have been able to develop workarounds due to the slow development and implementation of export controls. For instance, in March 2023, the Australian Financial Review reported that iFlytek, a Chinese speech recognition company added to the entity list in 2019, was circumventing restrictions on purchasing advanced U.S. chips by renting time on cloud computing servers equipped with advanced Nvidia chips to train its AI models. In September 2023, Huawei, a primary target of U.S. export regulations, released the Mate 60 smartphone, featuring a 7-nanometer chip manufactured by China's leading semiconductor manufacturer, SMIC. The 7-nanometer manufacturing process is a relatively advanced one, suggesting that SMIC has achieved technological progress faster than anticipated or has been able to evade export controls. Furthermore, at an event in Beijing in October 2023, China's giant search engine company Baidu announced a new version of its language model, ERNIE 4.0, which it claims rivals the performance of ChatGPT's AI models. Baidu stated that tens of thousands of chips were used to train ERNIE 4.0, and it was confirmed that Nvidia chips were ultimately used. Under regulations in 2023, the U.S. government did not explicitly name the H800 chip, but it is widely understood to be subject to new regulations (Wired October/17/2023).

Ultimately, U.S. export controls will face a series of challenges in deterring the transfer of technology to China. First, export controls are a temporary solution that can delay, but not completely prevent, China's acquisition and development of key technologies. Second, the end-user-focused approach requires extensive resources to track the proliferation of new companies acting on behalf of the Chinese military. Additionally, ownership and transaction data for identifying these companies may be inaccurate or unobtainable. Third, for many technologies, the U.S. does not possess sufficient control over the supply chain to unilaterally implement effective controls. Fourth, the existing regime focuses on non-proliferation rather than restricting the transfer of dual-use technologies, and multilateral coordination is difficult because it requires the consent of all member states.

VI. Policy Challenges for South Korea

The future of humanity becomes more uncertain when the strategic competition between AI and great powers, particularly the U.S. and China, which are leading in AI technology, is combined. If advanced countries leading in AI technology, other than the U.S. and China, fail to establish common norms and become entangled in the logic of U.S.-China strategic competition, the global security environment is likely to worsen. Currently, the U.S. and China are fiercely competing in all areas necessary for AI technological development; this competition must evolve towards creating common international norms and sanctions regimes before it escalates into military confrontation within the broader logic of U.S.-China strategic competition. It is important to strive for common regulatory regimes for AI within the window of opportunity presented by technological gaps, rather than relying solely on military sanctions, so that the current advantage held by the U.S. and its allies over China does not become solely based on military sanctions. The discussion on the military use of AI at the U.S.-China summit during the Asia-Pacific Economic Cooperation (APEC) in San Francisco in November 2023 can be seen as a step forward in the international community's concerns and efforts surrounding AI.

South Korea has pursued comprehensive AI policies, including establishing its own development plans for AI technology, advancing the ROK-U.S. technology alliance, enhancing cooperation with Western countries, strengthening the relationship between government and corporations, and further developing internationally indispensable technologies following its existing memory semiconductor technology (Bae 2023; Yoon 2023; Kim 2022). These efforts must continue. In particular, it will be important to make efforts to ensure that South Korea's economic interests are not harmed as U.S. economic containment policies against China regarding AI technology accelerate. In terms of AI technology, the South Korean semiconductor industry is not expected to be significantly affected by the intensified sanctions against China by the U.S. and the European Union for the time being. This is because Samsung Electronics operates NAND flash memory plants in China, and SK Hynix operates DRAM and NAND flash memory plants, but neither company produces AI semiconductors (Choi 2023). However, it will be necessary to consider changing geopolitical factors as South Korea expands its AI technology innovation in the future.

More importantly, it is necessary to closely monitor the multifaceted impacts of the expansion of military AI technology and the changing political and economic environment surrounding it. South Korea's future challenges are threefold. First, it is crucial to understand how the strategic competition and AI competition between the U.S. and China will become intertwined and how the military security environment surrounding the Korean Peninsula will change. The U.S. is currently pursuing containment of China in the semiconductor sector for AI competition, deepening bilateral containment while evolving multilateral frameworks. However, China is simultaneously accelerating its independent technological development through various efforts to circumvent sanctions. This competition will ultimately accelerate the overall development of military AI, and the nature of U.S.-China military security competition will also change. It is necessary to conduct specific research on how the acceleration of tactical and strategic AI use will affect the security environment of the Korean Peninsula. In particular, U.S.-China military competition is expanding into a full-fledged nuclear weapons competition, and nuclear weapons competition utilizing space technology and AI technology is becoming more active. It is a critical question whether nuclear weapons, previously deterred by the logic of mutually assured destruction in the 20th century, could be operated under different strategies in the new environment. Given the potential for escalation of U.S.-China military competition across Asia, including Taiwan, the possibility of AI-based nuclear war must always be considered.

Second, securing the military supply chain necessary for South Korea's military innovation is important. In the process of pursuing independent AI development, it is crucial for South Korea to maintain an innovation ecosystem that includes semiconductors, research personnel, mineral resources, and electricity, thereby laying the foundation for developing its military capabilities. South Korea has presented a basic plan for Military Innovation 4.0, which aims to prepare for intelligent warfare based on cross-domain integration, unmanned-unmanned complex combat, and hyper-connectivity and hyper-convergence. To this end, it aims to foster an AI science and technology powerhouse by securing advanced military capabilities based on fourth industrial revolution science and technology in a timely manner. This requires securing an economic supply chain for AI development. As global supply chains are being reorganized amidst U.S.-China technological competition, South Korea needs to strive for the redesign of a sustainable military supply chain. It will be important to simultaneously emphasize efforts to develop the ROK-U.S. alliance into an industrial and technological alliance and maintain broad strategic cooperation with strategic partner countries, while also managing these efforts to avoid deteriorating relations with major trading partners such as China.

Third, normative efforts will be necessary to prevent the strategic competition between the U.S. and China from militarizing, particularly the militarization of AI, which has catastrophic effects. As discussed earlier, the G7 countries have issued a non-binding code of conduct for AI development. In early November, the UK hosted an AI Safety Summit where representatives from 28 countries pledged to cooperate in managing the risks of AI. In early December, members of the European Union reached a political agreement on the AI Act, a landmark piece of legislation designed to mitigate the risks of the technology and set global regulatory standards. While these efforts discuss the broad development of AI, discussions on military AI, particularly autonomous weapons systems, are urgently needed. South Korea must redouble its efforts to establish international standards for the military use of AI, which could have catastrophic consequences, not only for our security but also globally. ■

References

Kim, Sang-bae. 2022. “Competition Among Digital Platforms in the Non-Face-to-Face Era: Complex Geopolitics of U.S.-China Technological Hegemony.” East Asia Institute. *[EAI Working Paper] Series on Post-COVID World Political and Economic Order ②*. February 8.

Bae, Young-ja. 2023. "Prospects for U.S.-China Technological Competition in 2023 and South Korea's Technological Diplomacy Strategy." East Asia Institute. *Korean Diplomacy 2023: Prospects and Strategies* 8.

Yoon, Jeong-hyun. 2023. “Innovation in Military AI Technology and International Normative Constraints: Implications for Balancing Security and Responsibility.” *Jeju Peace Institute Policy Forum* No. 2023-04.

Bremmer, Ian and Mustafa Suleyman. 2023. “The AI Power Paradox: Can States Learn to Govern Artificial Intelligence—Before It’s Too Late?” *Foreign Affairs*. September/October. 102-5.

Brill, Jonathan. 2023. “America’s Shaky Semiconductor Supremacy Over China.” Forbes. https://www.forbes.com/sites/jonathanbrill/2023/12/20/americas-shaky-semiconductor-supremacy-over-china/?sh=12b773071aca

Choi, Jasmine. 2023. “US-China Semiconductor War Enters ‘Round 2.’” Business Korea. https://www.businesskorea.co.kr/news/articleView.html?idxno=202813

Danzman, Sarah Bauerle and Emily Kilcrease. 2022. “The Illusion of Controls: Unilateral Attempts to Contain China’s Technology Ambitions Will Fail.” Foreign Affairs. December 30.

Fedasiuk, Ryan, Karson elmgren, and Ellen Lu. 2022. “Silicon Twist: Managing the Chinese Military's Access to AI Chips.” Center for Security and Emerging Technology. June.

Flournoy, Michèle A. 2023. “AI Is Already at War: How Artificial Intelligence Will Transform the Military.” Foreign Affairs. November/December. 102-6.

Horowitz, Michael C., Lauren Kahn, and Laura Resnick Samotin. 2022. “A Force for the Future: A High-Reward, Low-Risk Approach to AI Military Innovation.” Foreign Affairs. May/June.101-3.

Huang, Sihao and Bill Drexel. 2023. “China Goes on the Offensive: in the Chip War What the United States Should Do to Keep Its Lead.” Foreign Affairs. October 11.

Kissinger, Henry A. and Graham Allison. 2023. “The Path to AI Arms Control: America and China Must Work Together to Avert Catastrophe.” Foreign Affairs. October 13.

Krepinevich, Andrew F. Jr. 2024. “The Big One: Preparing for a Long War With China.” Foreign Affairs. January/February. 103-1.

Luo, Shuxian. 2022. “Addressing Military AI Risks in U.S.–China Crisis Management Mechanisms.” China International Strategy Review 4.2: 233-47.

Lushenko, Paul. 2023. “AI & the future of WARFARE.” Bulletin of the Atomic Scientists. https://thebulletin.org/2023/11/ai-and-the-future-of-warfare-the-troubling-evidence-from-the-us-military/

Muhammed, Can and Alena Vieira. "The Chinese Military-Civil Fusion Strategy: A State Action Theory Perspective." *The International Spectator*. 57.3 (2022): 85-102.

O’Dea, Ann. 2023. “How the chip war and the future of AI are inextricably linked.” Silicon Republic. https://www.siliconrepublic.com/future-human/chip-war-semiconductors-supply-tech-geopolitics-chris-miller.

Scharre, Paul. 2023. “America Can Win the AI Race: It Has the Resources—Now It Needs a Plan.” Foreign Affairs. April 4.

Wang, Dan. “China’s Hidden Tech Revolution: How Beijing Threatens U.S. Dominance.” Foreign Affairs. March/April 2023

Wired. 2023. “The US Just Escalated Its AI Chip War With China.” https://www.wired.com/story/the-us-just-escalated-its-ai-chip-war-with-china/

■ Jeon Jae-seong_Director of the National Security Research Center at the East Asia Institute, Professor at Seoul National University.

■ In charge and edited by: Lee Ju-yeon_Research Fellow at EAI

    Inquiries: 02 2277 1683 (ext. 205) | jylee@eai.or.kr