Table of Contents

As the applications of artificial intelligence (AI) in critical sectors like healthcare, smart cars, and military affairs continue to multiply, countries are seeking to realize technology’s economic and national security benefits. To capitalize on these opportunities and spur technological innovation, countries are scrambling to build healthy AI ecosystems.

As South Korea tries to secure its economic competitiveness, AI will be a crucial testing ground, and a good point of comparison for its efforts is Japan—another high-tech regional power in Asia that is seeking to capitalize on AI. Although most conversations on AI revolve around the capabilities of leading powers like the United States and China, South Korea and Japan also offer unique strengths that should not be overlooked.

Hana Anderson
Hana Anderson was a James C. Gaither Junior Fellow in the Carnegie Asia Program.
More >

With similar export-led growth models, Seoul and Tokyo have cemented their status as high-tech powers. The two countries are already both key players in the global supply chain for cutting-edge semiconductors, which are needed to build AI systems. But both countries are also aiming to become hubs for AI innovation so that they can reap AI’s national security and economic benefits. To do so, South Korea and Japan have laid out plans to catch up to the United States, China, and major EU economies.

To understand the global contours of AI ecosystems, one should consider not only the two leading players, the United States and China, but also mid-tier players such as Japan and South Korea. By focusing on three core building blocks of a thriving AI economy, it is possible to build a general understanding of the strengths and weaknesses of South Korea’s and Japan’s respective ecosystems.

The Building Blocks of AI

AI broadly refers to machines that are capable of simulating human-like behavior to accomplish a variety of tasks, ranging from language processing to complex decisionmaking. Over the past thirty years, the field of AI has become increasingly diverse and sophisticated, leading to the rise of subfields such as machine learning, which is currently at the center of AI-driven research and development. AI has the potential to spur rapid productivity gains across the private and public sectors with applications ranging from online customer service platforms to semiautonomous weapons. Governments around the world understand that AI-driven systems will have dramatic implications for national security, economic prosperity, and military power.

Jacob Feldgoise
Jacob Feldgoise was a James C. Gaither Junior Fellow in the Carnegie Asia Program.

Modern AI systems require three core components: data, computing power, and algorithms. (Although additional inputs such as talent and government policy are also needed to develop a robust AI ecosystem, they are beyond the scope of this piece.) However, while national comparisons on computing power can be made by looking at the semiconductor supply chain and cloud computing, it is difficult to make comparisons about algorithms and data.

Accurate AI systems also rely on large quantities of data to learn how to identify patterns and make predictions. In addition to quantity, these data sets must also be high quality and diverse because AI systems inevitably reflect the limitations of their underlying data. It is challenging to size up the quantity, quality, and diversity of a country’s data resources because that is proprietary information—primarily held by individual private companies who have little incentive to share it. To date, there is no clear way to rank countries based on their access to good data. This remains a fruitful area for future work.

Algorithms provide the instructions for how to process data inputs; machine-learning algorithms draw insights from data on their own, rather than being explicitly programmed. Algorithmic breakthroughs are quickly published in open-source academic publications, allowing that information to easily diffuse across national borders. This also makes it difficult to compare countries’ prowess in algorithm development—of which algorithm research is one component. Instead of looking at algorithms, this analysis compares South Korea’s and Japan’s contributions to cutting-edge AI research, as both countries are global leaders in this area.

Lastly, firms need computing power to train and deploy AI systems. Firms can increase their computing power through two main mechanisms: purchasing cloud computing services or securing semiconductors to build an in-house data center. Cloud computing or an in-house data center is also needed to store and manipulate large data sets. Small- and medium-sized enterprises usually rely on cloud computing services to build their AI systems. Large firms, meanwhile, often have sufficient resources and internal demand to build their own data centers. Both types of computing power rely on the fundamental building block of the digital age: the semiconductor. South Korea and Japan each play a crucial role in the semiconductor supply chain. Looking at two of these critical factors—computing power and cutting-edge AI research—helps shed light on the strengths and weaknesses of South Korea’s and Japan’s AI ecosystems.

Computing Power: Semiconductors

To maintain a resilient AI ecosystem, countries need to secure continued access to cutting-edge semiconductors. A steady supply of semiconductors is especially necessary for building local computing capacity with advanced AI capabilities—supporting in-house data centers and public cloud computing alike. Once a niche topic, semiconductors are now firmly embedded in public discourse, primarily due to the global shortages triggered by the coronavirus pandemic, which delayed sales of countless products from ovens to smartphones. While the United States, China, and Taiwan are typically identified as the predominant players in the global semiconductor supply chain, South Korea and Japan also play pivotal roles.

South Korea’s strength is fabricating leading-edge microchips. The country has steadily grown its share of global semiconductor manufacturing capacity from virtually zero in 1990 to 21 percent in 2020. Over the same period, Japan’s share declined from 19 percent to 15 percent.

Japanese officials have acknowledged that the country is behind on semiconductor manufacturing and is seeking to improve. However, in the short term, Japan will continue to depend on semiconductor imports. The share of domestic Japanese demand for semiconductors satisfied by imports (import penetration) has steadily risen since 2000. Meanwhile, over the same period, South Korea’s import penetration has steadily declined. This indicates that, while Japan’s reliance on semiconductor imports has increased, South Korea’s has dropped (see figure 1). Furthermore, much of Japan’s capacity is geared toward manufacturing less-advanced chips, while South Korea and Taiwan lead at the cutting edge.

Box 1: Figure 1 Methodology Notes

Import penetration measures the share of domestic demand that is satisfied by imports. It is calculated by the following formula:  Imports / (Production - Exports + Imports) . A 0 percent score would indicate low dependence on imports, while 100 percent would indicate high dependence on imports. Values can exceed 100 percent due to re-exports as well as inconsistencies between production and trade data. For example, a semiconductor product that is re-exported (imported and then exported) would increase the numerator of the import penetration formula without increasing the denominator—allowing it to exceed 100 percent. That South Korea and Japan’s import penetration statistics have both exceeded 100 percent at various points in time suggests that each country likely re-exports a significant share of semiconductor imports. For more information, see explanation of “Import Penetration” and “Export Share of Production” in the OECD’s STAN Indicators reference document. Also, South Korea’s production data was calculated using the country’s Index of Industrial Production (2000-2021), with a 2015 base year. Index data was multiplied by South Korea’s 2015 semiconductor production value to estimate production over the entire time period.

For now, only South Korea and Taiwan are in the enviable position of being able to fabricate the most advanced semiconductors, which are needed to train the most competitive AI systems. Innovation in semiconductor fabrication is measured in process nodes, whereby more advanced technology can etch wafers with greater precision. Today, South Korea’s Samsung and Taiwan’s TSMC are the only companies in the world that can produce commercially viable chips at the most-advanced 5 nanometer node, which they do by using next-generation extreme ultraviolet (EUV) lithography machines. When it comes to fabricating the most advanced chips (4–5 nanometers), as of 2021, Samsung is responsible for an estimated 31.3 percent of global capacity while TSMC is responsible for 66.8 percent. The remaining 1.9 percent is noncommercial research capacity maintained by the Interuniversity Microelectronics Centre (IMEC) in Belgium.

Both TSMC and Samsung are working on commercializing 3 nanometer fabrication, an advance that would secure their lead in the industry and keep their services in high demand for years to come. A misstep by either company in this sector-leading oligopoly would likely create problems for everyone; if Samsung and TSMC are unable to meet global demand for AI chips, prices will rise and wait times will lengthen.

Still, the semiconductor supply chain is highly complex, and fabrication is only one piece of the puzzle. Japan also plays an essential role in the semiconductor supply chain, particularly as a supplier of semiconductor manufacturing equipment (SME) and of fabrication materials, such as wafers and photoresist. Japan is a major manufacturer of SME, controlling about 31 percent of global market share. For AI chips, which are produced at the 90 nanometer node or below, Japan’s Nikon is the only company besides the Netherlands’ ASML that can produce photolithography equipment (a critical type of SME) at scale.

In 2019, Japanese companies controlled 56 percent of global market share for wafers and produced about 90 percent of the global supply of photoresist. Although only around 7 percent of a chip’s added value comes from these input materials (as opposed to 34 percent for the fabrication services that a company like Samsung supplies), a shortage of these materials––whether engineered or unintentional––would still create a debilitating bottleneck in the semiconductor supply chain (see figure 2).

Box 2: Figure 2 Methodology Notes

The corresponding product code for photoresist in the Harmonized Tariff Schedule of Korea is 3707901010. South Korea’s photoresist import values from Japan in each year was directly available in KITA’s statistics tool. To calculate South Korea’s imports from other countries in each year, South Korea’s photoresist imports from Japan were subtracted from South Korea’s total photoresist imports from all countries. To calculate the percentage labels for each year, South Korea’s photoresist imports from Japan were divided by South Korea’s total photoresist imports from all countries.

South Korea has been the target of an engineered bottleneck in EUV photoresist—an essential material for manufacturing cutting-edge chips using EUV lithography. In 2018, South Korea’s Supreme Court ordered Japanese companies to compensate South Korean victims of forced labor from World War II, sparking a torrent of hostility between the two countries. In July 2019, Japan imposed new restrictions on exports of two critical semiconductor materials (photoresist and hydrogen fluoride) to South Korea, delaying exports to South Korean chipmakers for several months. In the first half of 2019, around 90 percent of South Korea’s photoresist imports had come from Japan.

To minimize the economic damage, the South Korean government and domestic chipmakers sought to diversify the country’s supply of photoresist. However, South Korea is still heavily dependent on Japan for chip-making materials. In 2021, 79.5 percent of South Korea’s photoresist was still imported from Japan, and that share declined to 77.4 percent in the first seven months of 2022 (see figure 2). In the short term, South Korea’s semiconductor ambitions will likely continue to depend on a positive relationship with Japan. Still, for the most advanced semiconductors, both countries’ AI ecosystems rely in part on a stable supply from Samsung (as one of the world’s few suppliers), and Samsung in turn relies on a stable supply of photoresist (and other materials) from Japan. More broadly, global market demand for AI chips depends on South Korea’s proficiency in fabricating advanced semiconductors and Japan’s dominant role in the supplying of input materials.

Computing Power: The Cloud

While large firms can access computing power by building their own data centers, smaller firms are increasingly accessing computing power through the cloud because they lack the resources and demand to build in-house data centers. Many smaller firms build and deploy their AI systems on public cloud platforms—such as Amazon Web Services, Microsoft Azure, and Google Cloud—which allow multiple customers (companies or government entities) to share computing resources, reducing the costs for each client. As AI becomes ubiquitous and the importance of cloud computing grows, countries that supply cloud computing systems will profit from selling such services to others, and countries that adopt cloud computing systems will have access to more computational power.

All else being equal, countries that adopt cloud computing technologies more widely will likely experience disproportionately large benefits as AI diffuses throughout the global economy. The countries that are adopting cloud computing most heavily can be identified by measuring expenditures on cloud computing services. Currently, based on their similar levels of spending on cloud computing (as a share of GDP), Japan and South Korea appear to be on roughly equal footing with respect to cloud adoption (see figure 3). This suggests that Japanese and South Korean companies are likely investing substantial resources in AI readiness. To give a global perspective to these numbers, both countries outspend China as a share of GDP, but neither comes close to the United States, which is expected to spend more than 0.7 percent of GDP on cloud computing in 2021.

On the supply side, the countries that build the most globally competitive cloud computing platforms will capture a disproportionate share of cloud services sales. Global market share for public cloud infrastructure and platform services is dominated by U.S. and Chinese companies. As of 2020, the top five providers by revenue were Amazon Web Services, Microsoft, Google, Alibaba, and IBM. Notably, neither Japanese nor South Korean companies have produced a globally competitive cloud services platform. In fact, about 70 percent of South Korea’s market is controlled by foreign companies, as is around 80 percent of Japan’s market.

Still, South Korean and Japanese companies are investing similar amounts of resources in cloud computing, and both countries are outinvesting China as a share of GDP. In addition, other factors, such as government regulations, will impact companies’ ability to collect, store, and use data to develop AI systems; this topic would be a valuable subject for further study.

Cutting-Edge AI Research

South Korean and Japanese universities are also global leaders in cutting-edge AI research. Robust AI research environments signify that both countries may have a high concentration of top-notch AI talent, which local companies can draw upon. The most direct way to observe the impact of each country’s contributions to such AI research is to look at its share of citations of papers presented at academic conferences, where groundbreaking advances are often published.

Historically, South Korea’s share of citations was smaller than Japan’s, but the two countries’ performances have slowly converged since the early 2000s, with South Korea’s share surpassing Japan’s for the first time in 2019 (see figure 4). By contrast, both countries fall below China’s share of citations, and China, in turn, falls well below the United States.

The countries’ respective shares of AI conference citations per capita gives a sense of which countries are most effectively using their limited human capital to produce cutting-edge AI research. Given that Japan’s population is more than twice the size of South Korea’s, this metric shows that South Korea is punching above its weight (see figure 5). Since 2012, South Korea’s per capita share of AI conference citations has been higher than Japan’s, which suggests that South Korea’s companies and academic institutions may be gaining a leg up on their Japanese peers.

Conclusion

South Korea and Japan have developed competitive advantages in the global AI ecosystem, most notably in semiconductors; South Korea and Japan control critical nodes in the semiconductor supply chain, and South Korea is a leading fabricator of cutting-edge chips, while Japan is a leading supplier of chip-making materials. But the two countries still trail the United States and China in other AI-related metrics. With respect to cloud computing, South Korea and Japan are both outinvesting China as a share of GDP but lagging the United States. Finally, regarding cutting-edge AI research, South Korea has caught up to Japan, but both countries are far behind the United States and China.

Maintaining competitiveness in computing power and cutting-edge research is a vital endeavor for Japan and South Korea as they seek to gain an advantage in AI innovation. However, other factors such as spiraling U.S.-China economic tensions and historical animosity between Seoul and Tokyo will affect the prospects of both countries’ AI ecosystems. One looming challenge to sustaining robust AI ecosystems will be semiconductor supply chains. As technological and economic tensions escalate between Washington and Beijing, both Japan and South Korea must contend with U.S. efforts and Chinese attempts to reshore critical segments of the semiconductor supply chain. Japan and South Korea, which have historically benefitted from a network of globalized supply chains, must navigate these tensions without alienating either great power.