What is Japan’s national AI infrastructure for physical AI?

It is an NVIDIA Vera Rubin AI factory being established by Noetra with METI support. It will provide large-scale computing for multimodal AI models used in robotics, digital twins, AI agents and other real-world applications.

How large will the Noetra and NVIDIA AI factory be?

The planned facility includes 13,750 NVIDIA Vera CPUs, 27,500 NVIDIA Rubin GPUs and 140 megawatts of data center capacity.

What is the FRONTia Project in Japan?

FRONTia is METI’s project titled “Development of Multimodal Foundation Models with a View to AI Robotics and Physical AI.” It is intended to develop highly reliable multimodal foundation models using Japan’s industrial expertise and real-world data.

Physical AI refers to AI systems that can operate in or understand real-world environments. In this initiative, it includes applications such as robotics, digital twins and industrial AI agents.

Japan Launches World’s First National AI Infrastructure for Physical AI

Q: Will Japanese companies be able to use the models developed through FRONTia?

Noetra plans to make pretrained weights from its multimodal foundation models broadly available to domestic model developers and enterprises, alongside NVIDIA software and open-model resources.

Japan is launching a national AI infrastructure for physical AI with support from METI, Noetra and NVIDIA. The Vera Rubin AI factory will underpin the FRONTia Project, helping develop multimodal models for robotics, digital twins, AI agents and industrial applications.

Japan is pairing its industrial expertise with NVIDIA’s newest AI infrastructure to build what is described as the world’s first national AI infrastructure for physical AI. The project is designed to move AI beyond screens and chatbots, giving Japanese companies a shared foundation for smarter factories, robotics, logistics, healthcare and other real-world systems.

Japan’s National AI Infrastructure Takes Shape

NVIDIA is working with Noetra Corp. to establish an AI factory that will serve as the computing backbone for Japan’s physical AI ambitions. The initiative has support from Japan’s Ministry of Economy, Trade and Industry (METI) and brings together public policy, industrial data and advanced AI systems in a single national-scale effort.

The project is especially notable because its purpose is not simply to train general-purpose language models. It is intended to develop AI that can understand and operate in the physical world, where machines must interpret images, language, sensor data and changing environments together.

Project operator: Noetra Corp.
Technology partner: NVIDIA
Government support: Japan’s METI
Primary focus: national infrastructure for physical AI
Target sectors: manufacturing, logistics, healthcare, telecommunications and more

The NVIDIA Vera Rubin AI Factory Behind the Plan

At the center of the deployment is an NVIDIA Vera Rubin AI factory built with 13,750 NVIDIA Vera CPUs and 27,500 NVIDIA Rubin GPUs. It is planned to deliver 140 megawatts of data center capacity, a scale aimed at supporting frontier AI training rather than isolated enterprise experiments.

The facility will use NVIDIA Vera Rubin NVL72 racks and the NVIDIA DSX™ platform, with NVIDIA Spectrum-X™ Ethernet networking and NVIDIA BlueField® DPUs. Together, these components are designed to help infrastructure operators bring large AI systems into production more quickly while improving reliability, efficiency and token throughput per megawatt.

CPU count: 13,750 NVIDIA Vera CPUs
GPU count: 27,500 NVIDIA Rubin GPUs
Planned capacity: 140 megawatts
Architecture: NVIDIA Vera Rubin DSX AI factory architecture
Networking: NVIDIA Spectrum-X Ethernet

How the Japan National AI Infrastructure Supports FRONTia

The AI factory will provide the computing foundation for Japan’s FRONTia Project, formally titled “Development of Multimodal Foundation Models with a View to AI Robotics and Physical AI.” METI launched the program to combine Japan’s strengths in manufacturing technology, onsite operational know-how and real-world industrial data with global AI capabilities.

Multimodal models are important here because physical AI needs more than text. A robot, digital twin or industrial agent may need to reason over visual information, instructions, machine signals and context from a live environment. FRONTia’s goal is to create highly reliable models suited to those kinds of practical tasks.

Opening the building blocks to more Japanese organizations

Noetra plans to make the pretrained weights of its multimodal foundation models broadly available to domestic model developers and enterprises. That approach could give more organizations a starting point for building specialized systems, rather than requiring each one to develop foundational capabilities from scratch.

Foundation model access: pretrained weights from Noetra’s multimodal models
Development tools: NVIDIA Nemotron™, NVIDIA Cosmos™, NVIDIA Isaac™ GR00T open models and NVIDIA NeMo™ libraries
Potential applications: AI agents, digital twins, robotics and physical AI
Core objective: accelerate Japanese development and deployment of reliable industrial AI

Why Physical AI Matters to Japan’s Industrial Strategy

Japan has deep roots in modern manufacturing, and the new infrastructure is intended to translate that operational expertise into an advantage in the next generation of intelligent machines. Rather than treating AI as a standalone software layer, the initiative focuses on systems that can support work in factories, supply chains, care settings and communications networks.

That direction aligns with Japan’s AI Robotics Strategy, released in March. The strategy sets a goal for Japan to capture more than 30% of the global AI robotics market by 2040, representing an estimated $133 billion opportunity.

Strategic market goal: more than 30% of the global AI robotics market
Target year: 2040
Estimated opportunity: $133 billion
Industrial advantage: manufacturing knowledge and real-world operational data

Japanese industry and government leaders planning national physical AI infrastructure

What Makes This Different From a Typical AI Data Center

A conventional AI deployment may serve one company, one model or a limited set of digital use cases. This effort is positioned as national infrastructure: a shared computing base intended to support organizations across Japan and eventually train trillion-parameter-scale AI models.

Its emphasis on openness also matters. By pairing large-scale compute with broadly available model weights and software, the project aims to widen participation in physical AI development. The practical test will be whether this shared foundation helps Japanese researchers and businesses turn advanced models into dependable tools for complex, safety-conscious environments.

Scale of ambition: support for trillion-parameter-scale model training
Deployment model: a computing foundation available across Japan’s ecosystem
Technical priority: multimodal AI for real-world environments
Business priority: faster deployment of agentic and physical AI applications

Key Takeaways

Japan’s new project combines METI, Noetra and NVIDIA around a national physical AI initiative.
The planned AI factory includes 13,750 Vera CPUs, 27,500 Rubin GPUs and 140 megawatts of data center capacity.
It will underpin the FRONTia Project, which is focused on reliable multimodal foundation models for robotics and physical AI.
Domestic developers and enterprises are expected to gain access to pretrained model weights and NVIDIA software resources.
The infrastructure supports Japan’s longer-term aim of becoming a major force in the global AI robotics market by 2040.

What Comes Next for Japan’s Physical AI Push

The most meaningful outcome will not be the size of the AI factory alone, but what Japan’s industrial ecosystem builds with it. If shared models, computing capacity and factory-floor expertise come together effectively, FRONTia could become a practical bridge between frontier AI research and the machines, workflows and services people rely on every day.