AI Factory and Grid

Jensen Huang's announced architecture for national scale AI infrastructure, first deployed in Japan through the Nvidia and SoftBank partnership:

"The AI Grid will have AI factories for developing the AI models. It'll have AI RAN to distribute the AI models all over Japan. The architecture of the AI Factory and the AI RAN that we worked on together is revolutionary. The world has nothing like this. Japan will be the world's first."¹

Architecture

AI Factory, the production layer. Where models are trained, developed, and refined. Analogous to a manufacturing facility: raw inputs (data, compute, algorithms) go in, trained AI models come out. The AI Factory is the capital intensive node, requiring massive GPU clusters (Nvidia's H100, B200, and Blackwell generation), power infrastructure, and cooling.¹ It is where the intelligence is made.

AI RAN, the distribution layer. RAN stands for Radio Access Network, traditionally the cellular infrastructure that carries wireless communications. Nvidia's AI RAN is a next generation version that carries not just bits but intelligence: model inference, agentic responses, and AI services distributed across the network. Where the AI Factory makes the intelligence, AI RAN puts it everywhere.

AI Grid, the combined system. Factory plus RAN equals a national infrastructure for intelligence production and delivery. The Grid is to the AI era what the electrical grid is to the industrial era: without it, no AI economy is possible.

The highway analogy

Jensen invokes the highway framing explicitly at the Japan summit:

"You cannot have an automotive industry if there are no roads. Now you have built the AI roads for Japan, on top of it, all kinds of new services, new companies can flourish."¹

Masayoshi Son extends this: SoftBank's cell towers "used to just carry bits for telecommunication and internet surfing. However, now with this intelligence network, densely connected, it becomes one big neural brain, the infrastructure of intelligence for Japan."¹

The Japan AI Grid turns SoftBank's existing carrier infrastructure into a national intelligence delivery system. Son is not building the AI applications, though his subsidiaries will use them; he is building the road on which all Japanese AI applications will run.

Why national AI infrastructure matters

The AI Grid is the physical prerequisite for what Huang calls sovereign AI. His argument:

"Every country will produce its own intelligence. How is it possible that a company does not create its own AI? A company is intelligence, giving your brain away to somebody else."¹

The same logic applies at national scale. A country that outsources its AI infrastructure outsources its intelligence production. The AI Grid is the answer: national scale compute owned and operated domestically, producing AI tuned to the country's data, culture, and security requirements.

SoftBank is treating the Japan AI Grid as the world's first sovereign scale AI infrastructure. Every other carrier and country will need to follow. Japan's early move is both strategic, capturing the pattern before competitors, and symbolic, since Japan is positioning to lead the AI era after missing the software era.

Access model

Critically, the Japan AI Grid is not a closed enterprise system. Son's stated intent is to subsidize access for researchers, students, and startup companies at near free rates: "almost free to try out their new models, try out their applications of AI."¹ The grid is public utility infrastructure, not a private moat. Roads generate value not by restricting access but by attracting the maximum amount of traffic; the more startups build on the Japan AI Grid, the more value accrues to the infrastructure owner.

Why it matters

Huang puts a figure on the stakes at the same summit: "The industry of work is not one trillion, it is 100 trillion dollars."¹ Infrastructure sized to a market that large cannot be built as ordinary corporate capital expenditure, which is the reframe the AI Factory and Grid model performs: it treats AI infrastructure spending as nation building. It gives sovereign AI a concrete physical form: a country cannot claim AI sovereignty without an AI Grid of its own, and the architecture is the implementation of the sovereignty principle. It also predicts where regulation is headed, since a country that has already built its own AI Grid is positioned as the incumbent provider if and when domestic AI infrastructure requirements become mandatory.