Nvidia launches Cosmos 3 Edge model and expands its physical AI push in Japan

Nvidia Corp. has unveiled Cosmos 3 Edge, a compact world model built to run vision reasoning and robot control directly on edge devices, alongside a wave of partnerships that pushes its physical artificial intelligence platform deeper into Japan’s robotics and manufacturing base.

The model can handle 4 billion parameters and is built on Nvidia’s Nemotron family. It offers on-device vision reasoning and generates robot policies, and developers can adapt it to specific robots, vehicles and sensors in about a day. Cosmos 3 Edge runs on edge graphics chips and Nvidia’s Jetson platform, including the newly announced T2000 and T3000 modules, as well as RTX GPUs and DGX systems.

The release extends Cosmos, Nvidia’s world foundation model platform for physical AI, which the company uses to generate and score training data for robots and autonomous machines. Nvidia introduced the broader Cosmos 3 generation of omnimodal world models in June.

Nvidia also rolled out new Metropolis libraries that it says let developers build and operate Cosmos-based video intelligence systems at least six times faster, using coding agents for training and deployment.

Most of the news was aimed at Japan, where more than 20 companies said they intend to join the Nvidia Cosmos Coalition. The list runs from industrial robotics names such as FANUC Corp., Yaskawa Electric Corp. and Kawasaki Heavy Industries Ltd. to Fujitsu Ltd., Hitachi Ltd., NEC Corp., SoftBank Corp., Sony Group Corp., Honda R&D Co. Ltd. and Toyota-backed Preferred Networks Inc.

Fujitsu is already building a collaborative control platform with FANUC, Yaskawa Electric and Kawasaki Heavy Industries. SoftBank is building a physical AI development platform on Cosmos, Omniverse and Isaac Sim. Groove X Inc. is using Jetson to power its LOVOT companion robots, while Enactic Inc. is fine-tuning Nvidia’s Isaac GR00T model for elder-care robots and Telexistence Inc. is applying the tools to retail automation.

“The next frontier of AI is in the physical world and this is a once-in-a-generation opportunity for Japan,” said Nvidia founder and Chief Executive Jensen Huang, who is in Japan this week for a run of announcements.

Nvidia and the Japanese government separately unveiled what they billed as the world’s first national AI infrastructure. Noetra Corp. will build an AI factory powered by 27,500 Nvidia Rubin GPUs and 13,750 Vera CPUs, drawing 140 megawatts, with backing from Japan’s Ministry of Economy, Trade and Industry. The project, called FRONTia, will develop open multimodal foundation models for robotics, digital twins and intelligent manufacturing.

The government is putting about $2.4 billion toward the effort, according to CNBC. Japan is aiming to capture more than 30% of the global AI robotics market by 2040, an opportunity Nvidia and its partners value at $133 billion.

Huang framed the buildout as a chance for Japan to reclaim its manufacturing edge, saying the country that invented modern manufacturing is now building the AI factories for the next industrial revolution.

The push builds on Nvidia’s growing sovereign AI business, in which governments fund domestic computing capacity rather than rely on foreign cloud providers. It also underscores how central physical AI, the use of models that let machines perceive and act in the real world, has become to the company’s growth beyond the data center.

Image: Nvidia