Nations fueled by artificial intelligence (AI) nationalism – the sentiment that a nation must develop its own technology to serve its interests – are racing to achieve AI supremacy. Undoubtedly, the United States and China are at the forefront, adding yet another battleground to their increasingly tense relationship.
While there are various components necessary for having advanced AI capabilities, hardware to support AI and energy to run the models are two critical factors. Examining AI hardware and energy needs, the U.S. foreign policy approach to gaining AI hardware supremacy, and pre-existing U.S. capabilities reveal an opportunity for South Korea-U.S. cooperation.
AI Hardware and Energy Needs
A computer requires three fundamental hardware components to operate: logic chips or processors, memory chips, and storage chips. AI requires much of the same hardware but more advanced and complex, especially with logic chips, which are often also called semiconductors, microchips, computer chips, and integrated circuits.
While computers use general-purpose central processing units (CPUs) to execute computations and graphic processing units (GPUs) to render three-dimensional graphics, process videos, and execute parallel processes, AI uses specialized CPUs with AI accelerators to run the virtual machine and advanced GPUs for machine learning, deep learning training, and inference. AI-specialized processors include application-specific integrated circuits (ASICs) like Google’s Tensor Processing Units (TPUs), neural processing units (NPUs), field-programmable gate arrays (FPGAs), and Cerebras’ Wafer-scale Engines. Sometimes, such advanced chips are simply called “AI chips,” and while definitions of AI chips vary, Georgetown’s Center for Security and Emerging Technology defines it as a chip with GPUs, FPGAs, and certain types of ASICs for specialized AI calculations implemented as a core on system-on-a-chip.
Because AI chips process complex computations, they need a lot of power and energy. The intense energy usage can lead to AI equipment overheating, which then necessitates cooling solutions requiring even more energy. In July 2024, Johnny Liu, the president of Chief Telecom, emphasized that an AI data center will need at least 10 kilowatts of power grid capacity compared to the four to six kilowatts required for a traditional data center. Energy used by AI can exacerbate global warming and contribute to increased carbon dioxide emissions: AI already uses 33 times more energy than typical software, and a ChatGPT request uses 10 times more electricity than a Google search.
U.S. Foreign Policy Approach to AI Hardware
The U.S. foreign policy effort to sustain its technological edge on AI hardware revolves mostly around China and is an extension of the China-U.S. “chip war” that began in the Trump administration.
Prior to the 2000s, the United States was the global leader in computer chip manufacturing, producing 37 percent of the world’s chips, but by 2022, the U.S. was only producing 12 percent of chips. Asian companies like the Taiwan Semiconductor Manufacturing Company (TSMC) and South Korea’s Samsung or SK Hynix had gained a lead within three decades. Currently, U.S.-based Nvidia is the global leader in AI chips, controlling between 70 percent and 95 percent of the market. However, the majority of Nvidia’s AI chips are manufactured in Asia, particularly Taiwan, through partnerships with TSMC and SK Hynix.
Simultaneously, China continued to signal its objective of replacing the United States as the global technology leader. In 2017, China released its “New Generation Artificial Intelligence Development Plan,” which included the objective of becoming the world’s primary AI leader by 2030. While China lags in quality high-end chips, it dominates the market for legacy chips – mature-node computer chips of 28 nanometers or above – projecting a production of 33 percent of global legacy chips by 2027.
The continued reliance on East Asia for chips and the goal of containing China’s AI lead led to the 2022 U.S. CHIPS and Science Act, which earmarked $52 billion in funding for semiconductor chip companies in the form of tax incentives for manufacturers with locations in the United States. Following the CHIPS Act, the Biden administration instituted an export control policy on AI technology, banning China from accessing high-end chips, U.S. chip design software, U.S.-built semiconductor manufacturing equipment, and other U.S.-built components. Yet, this foreign policy approach to AI advancement and supply chain resilience is flawed for a few reasons.
First, barring China from accessing advanced chips and hardware does not stop China from advancing its AI capabilities. While behind, Chinese technology companies like Huawei and SMIC are progressing in AI chip development, and China could eventually catch up to the United States. In the meantime, China can use smuggled or previously imported chips. In August 2024, Chinese AI engineers were also found accessing Nvidia’s AI chips through cryptocurrency methods that access servers abroad.
Additionally, while legacy chips typically do not have the capacity to run AI processes, China can develop AI models that can be trained on fewer advanced chips, use multiple less powerful chips to form one higher-performing package for AI models, or develop smaller AI models that require less computational power.
Second, despite the projected chip production increase in the U.S. – estimated to reach nearly 30 percent of leading-edge chips by 2032 – East Asia continues to hold dominance in the industry. There are three major semiconductor manufacturers in the world: Taiwan’s TSMC, South Korea’s Samsung, and the United States’ Intel. While TSMC and Samsung provide foundry services – the ability to accept architectural and system designs from clients, which allows for the production of semiconductors that fit the specific client’s needs and functionality – Intel only announced its foundry services plan in June 2024.
Furthermore, the process of bringing chip fabrication plants to the United States has not been easy: construction delays, language barriers, cultural differences, complex regulatory processes, and more have delayed the operations of TSMC’s facility in Arizona as well as Samsung’s chip fabrication plant in Texas.
Not only is the CHIPS Act not really leading the U.S. to AI supremacy, it risks straining relations with key Asian allies by draining their chip manufacturers and threatening their supply chains. South Korea – commanding 16 percent of the global market share – depends largely on Chinese production facilities: Samsung’s Xi’an plant produces 40 percent of its NAND chips while SK Hynix manufactures half of its DRAM chips in Wuxi and Dalian. In March, SK Hynix’s CEO emphasized that China is a critical production base and a key market. With semiconductors comprising 19.1 percent of South Korea’s exports as of July 2024, maintaining strong ties with China is essential to sustain its leadership in chip development.
Due to the CHIPS Act’s inherent economic risk, South Korea introduced its own K-Chips Act, offering tax breaks and credits for companies investing in “national strategic goods” like semiconductors. Although the United States extended a waiver in late 2023 allowing Korean semiconductor firms to bring U.S. chip equipment into China, Washington continues to ask South Korea to impose restrictions on semiconductor technology exports to China.
Centering foreign policy on containing China and limiting other nations’ supply chain ties with China risks threatening allies’ supply chain strategies while not achieving AI hardware leadership for the United States. Instead the U.S. should work with allies to enhance AI infrastructure and strengthen collective resilience.
The Opportunity for South Korea-U.S. Cooperation
A key opportunity lies in clean energy. While the United States faces risk in AI hardware, it leads in renewable energy – a crucial component for powering AI technology. With data centers projected to consume up to 9 percent of U.S. electricity generation by 2030, the U.S. has begun actively addressing this challenge as highlighted in the August 2024 Department of Energy’s Recommendations on Powering AI and Data Center Infrastructure. These recommendations include assessments of technology timelines to limit carbon emissions, aimed at adhering to a least-cost path to the Biden administration’s goal of net-zero emissions by 2050. The United States is well-positioned for this approach, having attracted over $405 billion to clean technology and energy development since 2021.
In 2023, renewable energy made up less than 10 percent of South Korea’s power generation, far below the global average of over 30 percent. To support its $19 billion chip industry investment by 2027 and meet AI energy demands, South Korea must triple its clean energy deployment by 2030. As the leading clean tech nation, the United States can help meet this demand by deepening collaboration between U.S. and Korean labs to develop clean energy technologies as outlined in the first-ever U.S.-South Korea Next Generation Critical and Emerging Technologies (CET) Dialogue from December 2023. Additionally, efforts such as the bipartisan Partner with Korea Act, creating 15,000 annual visas for skilled professionals, could allow Korean nationals to gain experience in U.S. cleantech and later bring back their expertise to advance South Korea’s green tech sector.
By supporting South Korea’s renewable energy development, the United States would be strengthening a critical ally’s AI and chip industry, which is a step toward building collective resilience – how nations can collaborate to both build supply chain resiliency from China. Other examples of collective resilience include the U.S. National Science Foundation and South Korean Ministry of Science and ICT’s promotion of joint research and development in semiconductor development as well as pledged cooperation between the new U.S. National Semiconductor Technology Center and Korean Advanced Semiconductor Technology Center for public and private research efforts.
By shifting the strategy to developing AI hardware and required clean energy together as well as building collective resilience with South Korea, Washington and Seoul can both maintain their AI leadership.