Nvidia Japan AI Factory Plan Lists 140MW Capacity

A 140-megawatt AI factory will use 13,750 Vera CPUs and 27,500 Rubin GPUs for Noetra Corp.’s physical AI plan, according to Tech Wire Asia.

The planned system will support FRONTia, a Ministry of Economy, Trade and Industry project for multimodal foundation models in robotics and physical AI.

The new platform forms the hardware base for the national programme.

NVL72 Hardware Defines The AI Factory

The AI factory is based on NVL72 racks and the DSX platform.

The configuration also includes Spectrum-X Ethernet and BlueField DPUs across the compute, networking and data-processing stack.

FRONTia is designed around industrial data, domestic manufacturing expertise and open multimodal foundation models.

Ryosei Akazawa, Minister of Economy, Trade and Industry, described the programme as the core of the country’s physical AI ecosystem.

The company plans to release pretrained model weights to domestic developers and enterprises.

The software components named for the programme include Nemotron, Isaac GR00T and NeMo.

FRONTia Targets Robotics And Factory Workloads

Manufacturing, logistics, healthcare and telecommunications were named as target sectors for the initiative.

Noetra chief executive Hironobu Tamba described physical AI as dependent on computing, data and foundation technologies that exceed the resources of one company.

The AI Robotics Strategy was released in March.

The report listed a target of more than 30% of the global AI robotics market by 2040 and put the government’s estimate for that share at about $133 billion.

The wider national package also creates a Cosmos Coalition for robotics and industrial technology groups.

Prospective members include FANUC, Fujitsu, Hitachi, Honda R&D, Kawasaki Heavy Industries and Kubota; a separate part of the list named NEC, Preferred Networks, SoftBank, Sony, TIER IV and Yaskawa Electric.

Edge Model Extends The Physical AI Stack

Cosmos 3 Edge arrived with the coalition expansion.

The report described it as a four-billion-parameter edge model for robots and vision systems that need local environment interpretation and action generation.

The model runs on Nemotron and is intended for RTX GPUs, DGX systems and Jetson modules, including T2000 and T3000.

The company described the model as adaptable for machines, vehicles, sensors and operating environments.

Fujitsu is exploring a shared control platform with FANUC, Yaskawa Electric and Kawasaki Heavy Industries.

The planned system uses Nvidia robotics and simulation tools across robot learning, simulated production environments, model work, digital twins and checks before deployment.

Industrial AI research work was listed for NEC, Hitachi, OMRON and Preferred Networks.

SoftBank’s work covers a physical AI platform built with simulation and robotics software, alongside AI-enabled radio access networks through AI Aerial.

Blackwell Systems Sit Beside The AI Factory Plan

The same announcements also include Blackwell-based supercomputers for scientific and quantum workloads.

Those systems are separate from the AI factory but sit within the same national infrastructure push.

Jensen Huang, Nvidia founder and chief executive, called physical AI the next frontier and described the country’s opportunity as once-in-a-generation.

The broader package also covers robotics models, edge systems and scientific computing rather than only the AI factory buildout.

Mujin is assessing Cosmos for autonomous industrial robots running on MujinOS.

TRON K.K. is developing factory data workflows for assembly, picking, inspection, material handling and digitisation.

The report did not identify the site or construction timetable for the 140-megawatt AI factory.