Japan’s Role in the Era of AGI (Theory of AGI, AGI and Society, and Japan’s Choices: Part 3)
Introduction
In Part 1 of this series, we introduced theoretical research suggesting that AGI lies on a continuous extension of the scaling race. Next, in Part 2, we argued that the monopolization of advanced, post-AGI artificial intelligence by a handful of nations or entities poses a major “localization risk,” and that our future direction should aim toward building a multipolar, interdependent ecosystem network of AI.
In this final installment, Part 3, we focus on the realities of present-day Japan to discuss what concrete role the country can play within the current technological and international landscape. In particular, we lay out Japan’s potential as a candidate for a third or fourth pole, along with the compute infrastructure indispensable for supporting the scaling that leads continuously to AGI—focusing on the core challenges of gigawatt-scale data centers and power supply.
Across three installments, this series discusses the possibility that today’s AI development race is continuous with AGI, as well as its implications, from the following three perspectives:
- Part 1: Introducing recent learning theory research suggesting that Transformer-based models can asymptotically approach Solomonoff induction, demonstrating the possibility that AI and AGI are more closely continuous than previously thought.
- Part 2: Based on the potential continuity between AI and AGI, examining the significance of the scaling race led by US and Chinese tech giants, as well as strategic considerations for a desirable future.
- Part 3 (This Article): Considering the research and policy paths Japan should take, and presenting several concrete proposals.
Can Japan Become a Candidate for a Third or Fourth Pole?
In the previous Part 2, we showed that the localization of advanced AI development capabilities in a limited number of countries, such as the United States and China, could pose a major risk. If one assumes that AGI itself might be inherently dangerous to begin with, halting its development could be an option; however, in a reality where nations and corporations are pursuing development races driven by scaling laws backed by national security policies and economic rationality, such a scenario is unrealistic.
Therefore, one rational option that remains is to envision a future where as many third, fourth, and subsequent poles—nations and coalitions capable of developing AGI—are fostered as possible, with the resulting AIs forming the kind of interdependent, ecological AI network described in Part 2.
In this article, “third and fourth poles” refers to nations or economic spheres that can autonomously possess AGI development capabilities alongside the US and China. When it comes to undertaking this role as a third pole, Japan possesses highly promising conditions.
1 Geopolitical Conditions
It is obvious that the United States and China are the two giants of AI development. However, when considering alternatives outside the US and China in terms of national security, technical standards, and economic partnerships, Japan occupies an ideal geopolitical position to play a role in turning an AI alignment network (a multipolar interdependent structure) into reality.
2 Technological Foundations and Human Talent
The scale and accumulated expertise of Japan’s research community meet a certain standard, meaning there is ample room to project a presence in global competition if appropriate investments are made. The current decline in the prominence of Japanese AI research can largely be explained by factors other than human talent (without going into excessive detail here: an insufficient scale of R&D investment, especially the lack of the massive compute infrastructure required for cutting-edge AI R&D, as well as a lack of organizational agility and flexibility in research institutions, among other issues). Furthermore, Japan holds strengths in peripheral areas such as semiconductor manufacturing equipment, materials, energy technologies, robotics, and high-performance computing (HPC) including supercomputers; under the right policies, it is entirely possible to build an advantageous technological foundation by integrating these domains.
3 Long-Term Motivations
Possessing AGI technology holds the key to the future development of science, technology, and the economy. To break into the current US–China duopoly in the AGI race as a latecomer, massive investments in compute and energy infrastructure are required, as discussed later. For this, actions that may not make sense from a short-term economic rationality perspective will likely be necessary. However, from a long-term perspective, Japan, as a resource-poor nation, has a powerful motivation to maintain the status of a major developed nation that it secured through tremendous effort and sacrifice since the 19th century, and to continue advancing its international competitiveness. If this motivation is also consistent with contributing to the stability of human civilization as a whole, the justification becomes all the more compelling.
Among the three items mentioned above, the final one—long-term motivation—carries profound significance. In past industrial revolutions, a massive gap of more than tenfold in per capita GDP emerged between countries that successfully industrialized and those that did not. In economics, this phenomenon is known as the “Great Divergence,” marking the first clear distinction in human history between developed and developing nations. Japan’s current status as an advanced nation can be seen as the result of having barely caught the last bus of the Second Industrial Revolution through significant stretching of its limits and societal restructuring during the Meiji Restoration.
The automation of intellectual labor anticipated from AGI is projected from numerous angles to have an impact far exceeding that of the Industrial Revolution, which automated physical labor. There is little debate that this outcome will lead to a reshuffling among developed nations. Looking at history, a country’s industrial structure and economic standing are directly linked to the standard of living of its citizens. If we learn from history, should Japan fail to maintain AGI technology domestically, “digital colonization”—relying on external sources for intellectual production methods—will advance, leaving structural vulnerabilities across national security, industrial competitiveness, and the educational system. If Japan’s retention of AGI development capabilities as a third or fourth pole could also significantly contribute to the stability of the world order following the advent of AGI, then there is all the more reason not to forgo capitalizing on Japan’s current promising position.
Even if being the first to achieve AGI is not guaranteed, that is not a major issue. What truly matters is whether the domestic capability to develop and operate AGI is retained. This gap between the “haves” and “have-nots” could bring about the next Great Divergence, rivaling the economic disparities created by the Industrial Revolution. Maintaining AGI technology within one’s own country forms the foundation of national strength, directly tied to intellectual production, security, and industrial competitiveness; it is no longer merely a research topic, but national strategy itself. In 20th-century industrialization, steel manufacturing, electric power, and shipbuilding became symbols of national strength. In the same way, investing in AGI is nothing less than building strategic infrastructure to secure autonomous intellectual production capabilities as a nation.
Japan’s Challenge: Balancing Scaling and Agility
So, what should be done right now to achieve this? To consider this question, we must first reflect on why the United States and China are currently leading the pack as the two frontrunners in frontier AI development.
Just a few years ago, scaling laws were either not yet widely known, or even if known, there was no way to tell how long they would continue to hold. Making large investment decisions in compute back then, driven purely by the “conviction” that scaling laws would persist, involved extreme risk-taking. A shared structural dynamic seen in both the US and China is that leading AI companies were nurtured by capital from deep-pocketed tech giants being channeled into startups where top talent could make large investment decisions swiftly and freely. The defining feature of this domain is the necessity to balance two things: continuously ramping up the scale of technical investment and facilities in a doubling game (scaling), while steering massive investments through real-time decision-making that rapidly incorporates breakthrough insights on a week-by-week basis (agility). From a business management standpoint, balancing scaling and agility at such a magnitude is historically unprecedented. A handful of companies in the US and China succeeded in achieving this balance, whereas other countries failed to do so. This can be considered the fundamental reason behind the current US–China duopoly.
With this framing in mind, let us reassess the current situation in Japan. First, regarding scaling, frontier AI development today exhibits a prominent self-sustaining nature, where tech giants and mega-startups attract massive capital on the premise of future commercial value. The only way to back the enormous funds required to sustain the scaling race over the long run is for the developed AI to generate real-world economic value. However, raising the ecosystem from its current state—where it starts at a significant deficit—to a stage where a virtuous cycle of AI development and investment can spin on its own will likely require some unconventional measures. Next, regarding agility, international organizations have repeatedly pointed out that large Japanese organizations struggle with flexibility and speed in decision-making. Generally, smaller organizations hold an advantage over large ones when it comes to speed and flexibility. The fact that an ecosystem capable of concentrating sufficient capital into startups tackling promising domains has not yet matured can be seen as an ongoing challenge for Japan. Consequently, for Japan to become a third pole, an institutional design that supports scaling at the national strategic level without stifling agility in the private sector is essential.
Developing Gigawatt-Scale Data Centers for AGI Development
The purpose of the AI Alignment Network in publishing this article is to present the realistic outlook of AGI development being uncovered by academic research, and by urging Japan to establish an autonomous stance as an international third pole, ultimately mitigate the “localization risk” of future AGI technology being overly concentrated in specific regions or corporations.
While detailed proposals on industrial policy or corporate strategy fall beyond the scope of this article, we propose the national development of gigawatt-scale data centers as an exploratory concept to aid future discussions.
The Bottleneck in the Era of Scaling: Compute
As stated repeatedly, modern AI development is characterized by a virtuous cycle of upfront investments and outcomes driven by scaling laws. To approach AGI, securing the compute necessary to keep this scaling sustainable is indispensable. Training the latest large language models (LLMs) alone requires compute on the order of tens of thousands of GPU-years, and they consume immense amounts of power during inference as well. In the United States, plans are underway across the country to construct multiple data centers approaching gigawatt (GW) capacity to support this demand.
In Japan, by contrast, companies capable of undertaking frontier AI development have not sufficiently emerged, and a mechanism to autonomously drive the scaling cycle of “investment → compute → AI development → service operations → growth → reinvestment” has yet to be established. Within this loop, the most effective element to prime first is the foundational provision of compute.
Gigawatt-Scale Data Centers as a Bootstrap
While Japan has previously built compute infrastructure for AI research, its scale has remained limited to several megawatts to tens of megawatts. However, considering the high likelihood that performance improvements via scaling will continue—as discussed in Part 1—investing in gigawatt-scale compute infrastructure can be justified far more rationally than in the past. Building gigawatt-scale data centers with the explicit objective of AGI development can serve as an effective move to bootstrap this scaling. For example, during the initial phase of data center development, it would be practical to combine government-backed long-term funding facilities, electricity tax incentives, and priority access allocations for researchers and startups.
Ensuring Agility
Even if scaling can be initiated, agility remains indispensable for fostering innovation. In this field, small, experimental startups tend to hold an edge over large organizations; an environment that allows diverse hypotheses to be tested on a small scale with rapid iterations of trial and error is far more likely to yield results than direct initiatives run by governments or large enterprises.
Therefore, establishing a mechanism that opens up parts of these gigawatt-scale data centers to startups and researchers at costs competitive with international standards would be highly effective. This would stimulate emergence and rejuvenation across Japan’s entire AI research and development ecosystem.
Toward Becoming “The Easiest Country in the World to Develop AGI”
By combining such infrastructure development with an open support system, Japan could position itself as “the easiest country in the world to develop AGI.”
Rather than a mere technological investment, this represents a national readiness initiative across scientific, industrial, security, and ethical fronts, while also contributing to global risk diversification in the era of AGI.
The Bottleneck in Gigawatt Data Center Construction: Electric Power
The Limitations of Conventional Infrastructure
While Japan’s power infrastructure boasts high reliability, its flexibility to accommodate massive new demand has its limits. Power grid constraints, siting restrictions, and complex regulatory procedures make it difficult to secure the power required for gigawatt data centers in the short term. Renewable energy is also important, but given its output volatility and grid constraints, it is ill-suited for baseload supply. Furthermore, the practical deployment of fusion energy—considered the next-generation breakthrough power source—is likely to take longer than AGI development, which could unfold within the next 3 to 10 years. What is more, even assuming conventional power generation development, building new power plants and transmission lines generally takes more than a decade. This temporal gap is a major issue and a challenge that must be resolved.
Integrating Gigawatt Data Centers with Power Generation Facilities
As one solution to this challenge, we propose the integrated development of gigawatt data centers alongside dedicated power generation facilities. Because gigawatt-scale data centers have steady and predictable electricity demand, incorporating power generation facilities as part of their site makes practical sense. Additionally, by keeping the generation facilities off the public grid and strictly dedicated as part of the data center, it could become possible—assuming certain regulatory relaxations and legal provisions—to launch operations far more rapidly than constructing standard new power plants.
Small Modular Reactors (SMRs) represent one candidate power source; however, since the commercialization and construction of new technologies may require more time than the data center itself, phased development might be appropriate—such as co-locating near existing power generation facilities during early construction or meeting power demands through interim solutions like gas turbines. Over the longer term, transitions toward safe, highly efficient fusion energy or space-based solar power could also be brought into view as they become viable.
It should be noted that integrating data centers with power generation facilities must be made compatible with inherently safe designs that do not compromise emergency shutdown capabilities in the event of an AGI malfunction. This is contingent upon incorporating multi-layered safety designs to prevent AI loss of control or military misuse, as well as connecting to frameworks for international auditing and cooperation.
Conclusion
Throughout this series, we have introduced theoretical perspectives suggesting that artificial general intelligence (AGI) is not a sudden “leap,” but potentially lies along the continuous extension of current scaling laws, and we have examined the societal implications of this prospect.
In Part 1, through theoretical analysis of Transformers, we outlined emerging findings that models may asymptotically approach Solomonoff induction in the limit of infinite compute and data. In Part 2, we pointed out the risks of AGI and subsequent advanced AI concentrating within a handful of nations or corporations, noting that establishing a multipolar, interdependent network structure could contribute to securing international stability.
Finally, here in Part 3, we considered the role Japan could play as a third or fourth pole. Japan possesses clear strengths in its geopolitical conditions, technological foundations, and accumulated human talent, coupled with long-term motivations to preserve its status as an advanced nation and secure its international competitiveness. Building on this, we proposed that securing gigawatt-scale data centers and their power supplies represents a promising area of investment to serve as the foundation supporting the scaling that leads continuously to AGI.
In closing, we conclude this article with the following proposal:
We, the AI Alignment Network (ALIGN), are an independent non-profit organization aiming for a hopeful future for AI and humanity. With the rapid advances in AI technology in recent years, the “international localization risk”—in which future artificial general intelligence (AGI) technology becomes overly concentrated in specific nations or corporations—is becoming tangible. We believe that this risk to human civilization can be mitigated if Japan, a peace-oriented nation sharing universal values such as freedom, human rights, and democracy, retains AGI development capabilities as an international “third pole” following frontrunners the United States and China; opens up its achievements to help increase the number of subsequent “fourth” and “fifth” poles among nations and enterprises; and acts to form an “AI ecosystem” where AI and humans are interdependently and symbiotically linked. Grounded in this philosophy, we propose that Japan secure and strengthen its AGI development capabilities and share those achievements with the world, thereby contributing to global stability and harmony following the advent of AGI.
Read from the beginning (Go to Part 1)
This article is a reprint from the ALIGN Blog (October 14, 2025). Original: https://www.aialign.net/blog/transformer-agi-part3
Koichi Takahashi