“The Next Round of the AI War Is Sovereign AI” — Japan Commits USD 20 Billion to Build a Homegrown Model

Japan decides on large-scale investment after judging the risks of overseas dependence
Development of a trillion-parameter model to be opened to domestic companies
Global race intensifies to secure sovereign AI as a future core industry

As countries compete fiercely to survive in the era of artificial intelligence (AI), Japan is moving to establish a company dedicated to the development of “sovereign AI,” combining the forces of government and the private sector. Core Japanese companies, including SoftBank, are joining hands with the government to narrow the AI competitiveness gap with the United States and China. This movement is not confined to Japan. Europe, India, Southeast Asia, and the Middle East are also accelerating sovereign AI drives aimed at keeping computing infrastructure and core assets under national control.

Full-Scale Push Into Hyperscale AI, With SoftBank and Others Participating

According to the Yomiuri Shimbun on the 22nd, Japan’s Ministry of Economy, Trade and Industry plans to establish a new joint venture with more than 10 domestic companies, including SoftBank and Preferred Networks, to jointly develop a homegrown AI model. To support the initiative, the ministry will provide approximately USD 6.6 billion over five years starting from fiscal 2026, with more than USD 2.0 billion to be included in next year’s budget. The government will also support the collection of data required for AI training.

Private companies investing in the project will be selected through an open call. Separately, SoftBank is in discussions to invest approximately USD 13.1 billion over six years starting next year in data centers to be used for AI development. In particular, data centers currently under construction by SoftBank in Tomakomai, Hokkaido, and Sakai, Osaka, are expected to be used for the development of domestically produced AI models.

Preferred Networks, a Japanese AI venture, is also expected to play a key role. Founded in 2014 by researchers from the University of Tokyo, the unicorn company is widely regarded as Japan’s leading player in AI deep learning capabilities. Yomiuri reported that the new joint venture is expected to employ around 100 AI experts, mainly from SoftBank and Preferred Networks.

The joint venture aims to develop a model with approximately one trillion parameters. In AI models, a larger number of parameters generally implies higher performance. While the exact parameter counts of major models such as OpenAI’s ChatGPT and Google’s Gemini have not been disclosed, they are widely believed to exceed one trillion. The developed AI model will be opened to Japanese companies for customized use, with plans to further develop AI capable of being embedded in robots.

Japan Increases AI Support Budget by 67% This Year

The establishment of the public-private joint venture is widely seen as part of Japan’s broader sovereign AI policy. To foster the growth of sovereign AI companies, the Japanese government launched the Digital Agency in 2021 as a dedicated AI authority. As of January this year, more than 600 of its 1,180 staff members are from the private sector, outnumbering career bureaucrats in an effort to strengthen the agency’s practical capabilities and avoid bureaucratic inertia. Government spending on AI has also been steadily rising. This year, Japan’s direct AI support budget reached approximately USD 1.4 billion, a 67.4% increase from the previous year.

One of Japan’s most prominent sovereign AI companies is Sakana AI, which achieved unicorn status with a valuation of USD 1.0 billion just one year after its founding last year. Although the company was founded by two foreign nationals who previously worked at Google’s Japan research lab, it is recognized as a Japanese sovereign AI company. During its early growth stage, the Japanese government supported Sakana AI by providing free access to high-cost graphics processing units (GPUs), which typically require hundreds of millions of dollars in upfront investment.

Experts agree that simple subsidy programs alone are insufficient to nurture AI champions. The government must act as a “buyer of first resort” to create a virtuous ecosystem. In Japan, the Digital Agency already injects funding and supplies projects to companies, allowing them to survive, accumulate experience, and strengthen capabilities. Observers note that Japan is effectively replicating the strategy used by South Korea under the Kim Dae-jung administration to foster major IT companies.

Japan’s AI-related support also extends to the semiconductor industry. If AI represents the brain of the future, semiconductors are its limbs. The government is determined to restore competitiveness in semiconductors as well. Infrastructure development is being actively pursued. Although sovereign AI is a highly technology-intensive industry, the ultimate determinants of success are regulatory responsiveness and governance design capabilities. In this context, telecommunications companies are seen as particularly well positioned to benefit from AI infrastructure buildouts, explaining why SoftBank, one of Japan’s three major telecom operators and an investment powerhouse, is aggressively entering sovereign AI development.

Behind Japan’s massive fiscal commitment to sovereign AI lies a deep sense of crisis over the technology gap. Since generative AI and large language models have become core industrial infrastructure, Japan’s economy has consistently lagged in platform and algorithm domains. The government fears that leaving this structure unaddressed could lead to weakened manufacturing competitiveness and the erosion of data sovereignty.

Beyond Technological Dependence, Toward an Era of AI Sovereignty

Japan is not the only country seeking a breakthrough through AI amid structural low growth. In Europe, particularly in countries with weaker manufacturing bases, sovereign AI is increasingly positioned at the core of national strategy. For example, French President Emmanuel Macron described the construction of sovereign AI infrastructure as “our struggle for sovereignty” at the VivaTech technology conference held in Paris in May. At the same event, French AI startup Mistral announced plans to introduce 18,000 GPUs in cooperation with Nvidia. German Chancellor Friedrich Merz also signaled support, calling the planned introduction of 100,000 GPUs by Nvidia and Deutsche Telekom by 2027 “an important step toward digital sovereignty.” While neither country has yet presented a detailed blueprint for sovereign AI, both are prioritizing the acquisition of key assets required for AI development.

The United Kingdom is also moving into concrete action. In January, it unveiled the “AI Opportunities Action Plan,” presenting 50 policy recommendations and committing USD 2.6 billion in investment. Of this, USD 1.3 billion will be allocated to building national computing capacity, with the goal of expanding computing power twentyfold by 2030. This signals a shift from policy declaration to execution. Another USD 650 million will be allocated to a newly established sovereign AI unit, to be used in conjunction with the British Business Bank to create AI growth zones. In parallel, USD 975 million will be invested in installing a supercomputer at the University of Edinburgh for nationwide research use. On the talent front, the UK plans to actively attract overseas experts through the Turing AI Global Fellowship and expand the technical talent base from the education stage through the Tech First program. Additional measures include establishing an AI security institute and launching a sovereign AI industry forum centered on large corporations, completing the policy–industry–regulation linkage.

Emerging economies are also moving swiftly. The Indian government announced “India AI” last year, defining sovereign AI as a national mission. It established an independent India AI division and committed approximately USD 20.0 billion to building a comprehensive AI ecosystem across public and private sectors. Planned initiatives include the construction of public AI computing infrastructure comprising more than 10,000 GPUs, the India AI Innovation Center for developing large language and foundation models, India AI FutureSkills to provide AI education up to the doctoral level, and India AI Application Development initiatives aimed at solving problems within government and public institutions.

In Southeast Asia, Singapore is leading a coalition with Thailand, Vietnam, and Indonesia to develop “Sea-Lion AI.” Although smaller in scale than efforts elsewhere, Sea-Lion AI aims to provide culturally unbiased outputs and serve governments, companies, and academia in Southeast Asia that are often marginalized in the global AI landscape. Elsewhere, the United Arab Emirates has unveiled Arabic-specialized large language models Falcon and Jais, while the Abu Dhabi government established AI71 to lead the development of Falcon 2. Saudi Arabia has committed more than USD 40.0 billion to AI under its Vision 2030 initiative and is currently working with Naver to develop Arabic LLMs and build data centers. Qatar, for its part, is investing USD 2.8 billion in AI and cloud infrastructure under its 2025–2030 national strategy, accelerating the development of its own AI capabilities.