The dominance of the United States and China in artificial intelligence (AI) and large language model (LLMs) technologies has raised sovereign concerns about other nations’ domestic AI capabilities.
In response, “Sovereign AI”—a concept referring to a country’s capacity for developing and operating AI systems—has emerged as a global policy priority.
Indigenous AI technologies are seen as critical as they increasingly shape modern lives. Domestic foundation AI models can perform with a deeper understanding of a country’s linguistics and cultural nuances, while also reducing reliance on foreign technologies.
Despite its robust economy and a strong R&D foundation, Japan is no exception in the global race for sovereign AI. Since the dawn of the AI age in the early 2020s, the country has lagged behind the US and China in AI innovation.
In response, through a combination of targeted innovation policies, Japan is seeking to reverse this trend and foster an indigenous AI ecosystem. As more countries pursue sovereign AI capabilities, Japan’s initiatives and outcomes could hold important implications for similar efforts around the globe.
What is ‘Sovereign AI’?
The recent explosion of AI and LLM technologies promises substantial gains in efficiency and productivity across societies. From personal assistance and corporate activities to academic research and public infrastructure, AI is enhancing performance and boosting output across diverse domains.
However, the global benefits of AI innovation are expected to be unevenly distributed. In nearly all performance benchmarks, the top-performing AI models are dominated by US and Chinese tech firms.
UNCTAD’s global development report predicts the AI market will reach US$4.8 trillion in 2033, but those gains will likely be concentrated among a small number of companies and regions.
The concept of “sovereign AI” has gained currency in response to these concerns. Sovereign AI refers to a nation’s ability to develop and sustain indigenous AI capabilities—including foundational model technologies, data systems and supporting infrastructure—that ensure autonomous control over AI development and deployment while minimizing reliance on foreign providers.
The primary advantage of AI sovereignty lies in its ability to generate outputs that are tailored to the specific needs of its user base. Since the performance of foundation models relies heavily on training data, domestic models trained in local languages and informed by historical and cultural context are better positioned to serve national needs than foreign models designed for universal use.
Developing domestic AI capacity also helps reduce foreign influence over critical national functions. As AI technologies become more embedded in critical infrastructure and national defense systems, reliance on foreign providers increases the risks of interference or vulnerability during crises.
Moreover, developing domestic models and data facilities can help reduce the risk of foreign access to sensitive information. Since the training of AI models requires vast volumes of data that often includes sensitive data, governments can better safeguard and govern access to that data by promoting AI technologies and infrastructure domestically.
Finally, sovereign AI carries major economic implications. As UNCTAD and other reports estimate, the global AI market is expected to surge in the coming decades. By cultivating domestic AI enterprises, countries can capture a greater share of the resulting economic value – rather than allowing it to flow to foreign technology providers.
In recent years, a growing number of countries have joined the sovereign AI race. These include not only established AI investors such as the UAE, Taiwan and Singapore, but also emerging players like Vietnam, Thailand and Indonesia.
AI policy pillars
Japan’s AI Strategy 2022 outlines the government’s broad ambition to harness AI to boost industrial competitiveness and tackle national and global challenges. While the government strategy does not explicitly use the term “sovereign AI”, it advances three core initiatives aimed at fostering an indigenous AI ecosystem.
The first is the Ministry of Economy, Trade and Industry’s (METI’s) Cloud Program, launched under the Economic Security Promotion Act. Backed by more than 100 million yen ($6.8 million), the program designates cloud services as “Specified Critical Products” and subsidizes domestic providers to reduce foreign reliance.
The Cloud Program’s implementation emphasizes the risks of market concentration among dominant foreign providers and aims to raise Japan’s domestic GPU cloud service sufficiency from the current level of 30%.
Funding has already been allocated to leading firms and institutions—including GMO, Sakura Internet, Softbank, KDDI, and the University of Tokyo—to expand GPU-equipped data centers across the country.
The second initiative is the development of ABCI 3.0, a public supercomputer built by the National Institute of Advanced Industrial Science and Technology (AIST).
Funded with 36 billion yen ($232 million) through METI’s Economic Security Fund, ABCI 3.0 is the third generation of Japan’s open computing infrastructure designed for AI development and real-world application bridging.
The mainframe, constructed by Hewlett-Packard Enterprise (HPE) and powered by Nvidia’s GPUs and network system, delivers a theoretical peak performance of approximately 6.2 exaflops (half-precision, 16-bit). This makes it the most powerful public supercomputer in Japan to date.
Since January 2025, AIST has offered ABCI 3.0 access to companies and universities for foundational model development. AIST also plans to develop its own multimodal AI models using the supercomputer for societal and industrial use.
The third pillar is the Generative AI Accelerator Challenge (GENIAC), a METI-sponsored accelerator program aimed at supporting domestic foundational model projects.
GENIAC aims to strengthen Japan’s competitiveness in foundational AI by providing computing resource assistance, supporting demonstration projects for data utilization, organizing matching events and facilitating collaboration with global tech companies for selected start-up and research projects.
Since its launch in February 2024, two rounds of GENIAC’s six-month cycle have assisted 30 projects. The selected projects include the development of LLMs specifically designed for processing the Japanese language, autonomous driving systems optimized for domestic traffic conditions, and foundational model technologies in research and industry activities.
Shortages and shortfalls
Despite these investments, Japan’s path to sovereign AI is far from straightforward. The country still faces significant challenges in reaching its AI goals.
One major obstacle is the shortage of skilled IT labor. Like many other countries, Japan struggles to meet the rising demand for IT workers. METI projects that by 2040, Japan will face a shortfall of 3.2 million workers in AI and robotics technologies.
A key reason behind Japan’s IT labor shortage is the lack of advanced educational programs in computer and data science, along with low enrollment in doctoral programs in these fields. The country also suffers from brain drain, driven by comparatively low salary levels.
Japan also faces a shortage of training data for foundational AI models. While the scale and quantity of training data are essential in developing AI models, the quantity and diversity of Japanese language text data lags far behind that of English or Chinese.
This shortfall is exacerbated by the absence of major domestic internet platforms—such as Facebook, X, or WeChat—that generate vast amounts of user data for model training.
Some policymakers question the concept of AI sovereignty per se. While countries like Japan strive to build sovereign AI systems, they still depend on foreign providers like Nvidia for critical infrastructure like data centers and computing components.
This reliance can paradoxically increase foreign dependence, potentially giving external actors access to valuable data or control over AI infrastructure.
In the end, Japan must strike a balance between pragmatic compromises and long-term strategic goals in its pursuit of AI autonomy. Success may lie in how “AI sovereignty” is framed—by prioritizing areas for domestic investment, identifying acceptable areas of foreign dependence and carefully choosing trusted international partners.
Given the high stakes involved in AI governance and development, Japanese policymakers face difficult choices about strategy and resource allocation. The extent to which Japan can achieve a meaningful level of AI sovereignty will have far-reaching implications–not only for its own future, but for the global community as well.
Atsushi Sumikawa is a recent graduate of Georgetown University’s master’s program in national security and emerging technology policy.
