On the Release of the NAGI Initiative
Introduction
Last October, in 2025, we published a three-part article series on this blog ([Part 1], [Part 2], and [Part 3]). We discussed how AGI (artificial general intelligence) could emerge as an extension of current AI technology, the “localization risk” that arises when development capabilities are concentrated within a handful of countries and corporations, and the multipolar strategy needed to mitigate that risk.
Back in October when we published those articles, we had already synthesized this discussion into a policy proposal: the “Nippon AGI Infrastructure (NAGI) Initiative,” a proposal advocating that Japan develop large-scale compute and power sources necessary for frontier AI training as national infrastructure. While the first edition was shared only with relevant stakeholders for feedback and revision, we are now releasing this second edition to the public.
[Nippon AGI Infrastructure (NAGI) Initiative, Second Edition (PDF)]
Why ALIGN Is Making a Policy Proposal for Japan
Some might wonder why an organization championing AI alignment is putting forward a policy proposal concerning Japan’s national infrastructure. ALIGN (AI Alignment Network) is a network of members with diverse backgrounds, and we are not an organization with a single, unified view on concrete technical strategies or policies. However, one conclusion we have reached in pursuing our mission is that the multipolarization of AGI development capabilities is essential, and the single most effective starting move right now is building out Japan’s compute infrastructure.
Our mission at ALIGN is to study the conditions under which AI can be safe and beneficial for all of humanity, and to help bring them about.
What we aim for is an “AI ecosystem” where diverse AIs detect each other’s deviations and act as mutual checks, with the very structure of the network ensuring safety. To foster such diversity, there must be multiple development hubs across the world possessing distinct technical backgrounds, design philosophies, and cultural contexts.
If AGI development capabilities remain concentrated in a handful of entities, errors in judgment or runaway behavior by any one of them could ripple across the entire system. As ecology teaches us, systems with low diversity are fragile in the face of environmental fluctuations. Moreover, because AGI technology has the power to fundamentally alter the productive structure of the economy, any country that abandons its development effectively surrenders the very foundation of its economic future; thus, the development race is virtually unstoppable. Under this constraint, increasing the number of poles capable of development and building a multipolar, interdependent network is the best strategy for reducing overall risk for all of humanity.
In efforts surrounding AGI safety, independent non-profit organizations like FLI, CAIS, and GovAI conduct research and offer policy recommendations alongside industry and government. While it is purely by chance that ALIGN is based in Japan, being based here means the Japanese government and society are the parties we can engage most directly and effectively. And as discussed in the next section, if Japan possesses the conditions to serve as the launchpad for multipolarization, it is only natural for ALIGN to direct this proposal toward the Japanese government.
Why Japan Can Be the Starting Point for a Third Pole
Currently, the countries capable of making massive compute investments toward AGI are primarily the United States and China. This is a de facto bipolar structure, and this structure inherently carries risks. In a bilateral relationship, only the dynamics of confrontation or subordination can operate, leaving no third party to check the failure or runaway behavior of either side. Furthermore, competition between two parties creates pressure to prioritize development speed, creating the danger that ensuring safety will be sidelined. From an alignment perspective, this is the most alarming structure possible.
Breaking this bipolar structure requires, first and foremost, a third pole. Only with three parties do the possibilities for mediation and collaboration arise; by sharing its technology and expertise, a third pole can pave the way toward fourth and fifth poles.
A third pole must meet at least three conditions: possessing the technical foundation required for AGI development, occupying a geopolitical position not wholly subordinate to either the US or China, and being capable of serving as a conduit for technology diffusion based on trusted relationships with other countries. Very few nations meet all these criteria simultaneously. India lacks the foundation for advanced semiconductor manufacturing, the UK is closely integrated with the US in national security (Five Eyes) and faces constraints as an independent pole, and the EU struggles with securing compute infrastructure at scale as well as decision-making speed.
Japan combines competitiveness in advanced semiconductor manufacturing equipment, a legacy of basic science demonstrated by its Nobel laureates, a geopolitical position not fully absorbed by either the US or Chinese camps, and a cultural position capable of connecting both Asia and the West to mediate between differing value systems. What is missing is the large-scale compute capable of running frontier AI training domestically, along with the stable power supply to support it.
Yet nations do not act on favorable conditions alone. Spurring action requires rational incentives. Two incentives exist here. The first is economic security. There is a qualitative divide between being able to use AGI technology and being able to build it, and access through APIs can be restricted or cut off at the provider’s discretion. National investment in compute infrastructure is a rational choice for escaping this structure of dependency. The second is reducing risk for all of humanity through multipolarization. By establishing a technical foundation and expertise as a third pole and sharing them with other nations, Japan can help expand multipolarization even further.
Realizing that these two incentives point in the same direction was the direct motivation behind our proposal to the Japanese government.
Overview of the NAGI Initiative
The NAGI Initiative is a policy proposal designed to address these compute and power bottlenecks. Written for Japanese policymakers, the proposal frames its argument around economic security rather than front-loading the AI ecosystem perspective that ALIGN emphasizes. The crucial point, as discussed in this post, is that both goals converge on the very same means.
Rather than a project to develop a specific AI model, we propose building a shared foundation as national infrastructure, enabling all researchers, companies, and startups within the country to conduct frontier training.
Specifically, we propose developing a cluster of ultra-large-scale compute facilities dubbed “Giga Data Centers (GigaDCs)” integrated with low-environmental-impact power generation facilities. The plan calls for launching operations rapidly in the short term using available power, followed by a phased expansion toward dedicated power sources over the medium to long term. Please refer to the proposal text for details regarding design requirements, service design, technical architecture, regulatory frameworks, risk analyses, KPIs, and the roadmap.
We welcome feedback and critique regarding this initiative.
This article is reprinted from the ALIGN Blog (April 16, 2026). Original text: https://www.aialign.net/blog/nagi-v2
Koichi Takahashi