Key Points:
- A new 4-billion-parameter Cosmos 3 Edge model is now available for real-time, on-device vision reasoning in robotics.
- Japan’s Noetra consortium is constructing the world’s first state-backed 140MW physical AI factory.
- The facility will house 27,500 advanced Rubin GPUs and 13,750 Vera CPUs to power Japan’s national FRONTia project.
- Japan’s top robotics leaders, including Fanuc, Yaskawa, and Kawasaki, plan to join the new Cosmos Coalition to advance open world models.
A massive technological expansion is taking place in East Asia as the global semiconductor leader moves to anchor the future of industrial robotics. A new artificial intelligence model and a series of sweeping partnerships will construct the world’s first national AI infrastructure in Japan. The initiatives, designed to accelerate the deployment of intelligent machines across manufacturing, logistics, and automotive sectors, will provide Japan with an unprecedented computational foundation to lead the physical AI revolution.
The newly released Nvidia Cosmos 3 Edge model brings frontier vision-reasoning capabilities directly to edge computing platforms. This new 4-billion-parameter open-world model, built on the proprietary Nemotron architecture, runs locally on the Nvidia Jetson Thor robotics platform, enabling embodied systems and robots to perceive their surroundings, reason in real time, and generate precise physical actions on-site. By processing data locally without relying on distant cloud servers, the technology drastically reduces latency and ensures operational safety in high-velocity factory environments.
A newly formed Japanese industrial consortium called Noetra is partnering to supply the massive computational power required to train these physical systems. The joint venture will build a state-backed, 140-megawatt (140MW) “Vera Rubin” AI factory at a former electronics manufacturing plant in Sakai City, Osaka Prefecture. The specialized data center represents the world’s first national computing infrastructure dedicated solely to physical AI, serving as the physical engine for Japan’s government-funded FRONTia project.
The planned Osaka facility will feature 13,750 Nvidia Vera central processing units (CPUs) and 27,500 next-generation Nvidia Rubin graphics processing units (GPUs). This massive hardware package divides exactly into 382 custom Vera Rubin NVL72 racks, each utilizing the company’s advanced DSX reference platform. The entire computing cluster will scale up using high-speed Spectrum-X Ethernet networking, providing domestic developers with one of the most powerful computational environments on the planet to train trillion-parameter models.
The FRONTia Project, a state-funded initiative under Japan’s Ministry of Economy, Trade and Industry (METI), lies at the core of this national computing infrastructure. The project, officially titled “Development of Multimodal Foundation Models with a View to AI Robotics and Physical AI,” brings together the country’s manufacturing expertise and real-world data. The initiative is backed by a massive 387.3 billion yen (approximately $2.4 billion) in first-year funding, with total public-private investments projected to reach up to 1 trillion yen (about $6.1 billion) over the next five years.
The Cosmos Coalition in Japan expands the collaborative software ecosystem alongside the hardware buildout. Twenty-two of the country’s leading robotics, manufacturing, and technology firms have declared their intent to join the coalition. This highly prestigious roster includes the world’s two largest industrial robot manufacturers, Fanuc and Yaskawa Electric, alongside other heavyweights like Kawasaki Heavy Industries, Fujitsu, Hitachi, SoftBank, Honda R&D, and Sony Group. The coalition will work to develop open, highly reliable frontier physical AI models, encouraging shared standards across the industry.
This cooperative strategy challenges vertically integrated competitors. Historically, major robotics manufacturers have operated within tightly guarded, highly proprietary software control stacks. By recruiting these industrial giants into an open coalition organized around its own models, the chipmaker is successfully establishing its software platform as the industry standard. This open-stack approach mirrors previous, highly successful software plays in other technology sectors, allowing the hardware provider to supply the underlying technology while letting local manufacturers manage active deployment.
The domestic expansion also focuses heavily on tailoring artificial intelligence to local language and workforce needs. Multiple Japanese enterprises, startups, and academic institutions are actively building specialized regional models using open Nemotron datasets and libraries. For instance, the Institute of Science Tokyo recently developed its “Swallow” family of open foundation models using these resources. This localized training enhances the models’ Japanese-language reasoning and cultural understanding while preserving their underlying capabilities in mathematics and software coding, facilitating domestic enterprise adoption.
Direct partnerships with Japan’s largest industrial groups are expanding beyond the Noetra consortium. Advanced technical support will assist Toyota Motor’s “Woven City” smart-city project in Shizuoka Prefecture, helping to test autonomous vehicles and smart logistics in a real-world sandbox. Simultaneously, collaboration with Hitachi will develop a system that integrates and controls entire factories and power plants, demonstrating that the boundaries of physical AI are expanding far beyond standalone robotic arms.
Ultimately, these product releases and infrastructure partnerships mark a critical turning point for the global industrial economy. By delivering the Cosmos 3 Edge model for localized reasoning and constructing a 140MW national AI factory to train these systems, the project is building the necessary foundation to automate modern manufacturing. As the construction of the Osaka facility begins and the Cosmos Coalition expands, Japan’s rich heritage in precision engineering and robotics will merge with advanced computing, redefining how the physical world operates in the digital era.
