Nvidia has launched Cosmos 3 Edge, a new AI model built to help robots and vision-based AI agents interpret and respond to physical environments in real time.
The release marks the latest step in Nvidia’s effort to move artificial intelligence beyond data centers and into factories, hospitals, laboratories, warehouses, and other real-world settings. The company introduced the model on Wednesday as CEO Jensen Huang began a two-day visit to Japan focused on robotics, manufacturing, healthcare, and industrial AI.
Cosmos 3 Edge is a world model, a class of AI systems trained to recognize how objects, people, and environments behave in physical space. Unlike large language models, which focus mainly on text, world models can process visual, spatial, and sensor-based inputs that machines need to operate outside a screen.
The model follows the launch of Cosmos 3 in May and is aimed at edge devices, where AI systems must make decisions close to the robot, camera, vehicle, or machine collecting the data.
That distinction matters for industrial systems. A factory robot cannot always wait for information to travel to a distant data center and back before reacting to a moving object, a worker, or a change on the production line. Edge-based models can reduce that delay and allow machines to respond closer to real time.
Nvidia builds a physical AI coalition in Japan
Nvidia is pairing the Cosmos 3 Edge release with a broader push to build a physical AI network across Japan.
The company said local industrial groups, including Fujitsu, Hitachi, and Kawasaki Heavy Industries, intend to join a new coalition focused on robotics and industrial AI. The group brings together companies with deep experience in manufacturing, engineering, transportation, and automation.
“The next frontier of AI is in the physical world, and this is a once-in-a-generation opportunity for Japan,” Nvidia CEO Jensen Huang said in a Wednesday statement. “Japan invented modern manufacturing. Now, it has the opportunity to reinvent it for the age of intelligent industries.”
Cosmos 3 Edge
Japan offers Nvidia a natural testing ground. The country is home to some of the most established robotics, automotive, electronics, and heavy-industry companies in the world. It is also dealing with an aging population and labor shortages that are pushing businesses to automate more tasks.
Those pressures are creating demand for machines that can perform work in less predictable settings. Traditional industrial robots tend to follow fixed instructions in controlled spaces. Physical AI systems aim to give machines more ability to observe, interpret, and adapt.
The shift could open new markets for Nvidia’s chips, simulation software, robotics tools, and AI models. The company has spent years building a software stack around its graphics processors, giving customers more reasons to stay inside the Nvidia ecosystem.
Japan’s AI market is expected to reach $27.9 billion by 2029, according to the International Trade Administration. Tokyo has encouraged AI adoption across manufacturing, healthcare, government, and other sectors, creating openings for foreign technology companies and local partners.
Ajay Rajadhyaksha, global chairman of research at Barclays, told CNBC last month that Japan holds an advantage in Asia, supported by a mix of AI investment and long-term economic changes.
Nvidia is entering a market that has already attracted large bets from other technology companies. Microsoft recently announced a $10 billion investment in Japan aimed at AI infrastructure and cybersecurity. SoftBank has committed major capital to AI and is seeking partnerships with Microsoft and Sakura Internet to develop domestic computing infrastructure.
From factories to drug discovery
Nvidia’s Japan strategy stretches beyond industrial robots.
The company is extending its reach into biotechnology, drug research, and medical robotics through tools built for agentic AI. These systems can carry out sequences of research tasks, analyze data, generate hypotheses, and assist scientists with parts of the discovery process.
Nvidia pointed to the continued growth of Tokyo-1, an AI drug discovery consortium operated by Xeureka, a Mitsui subsidiary. The project was first announced in 2023 and uses Nvidia’s BioNeMo Agent Toolkit, a software platform built for computational biology and pharmaceutical research.
Japanese drugmakers Astellas Pharma, Daiichi Sankyo, and Ono Pharmaceutical are using Nvidia’s biology tools to support parts of their research workflows, according to the company.
Drug discovery has become one of the more closely watched commercial uses of AI. Pharmaceutical companies spend years and billions of dollars bringing new treatments to market, with many candidates failing before approval. AI tools may help researchers screen compounds, model proteins, analyze biological data, and identify promising targets earlier in the process.
The commercial opportunity remains uncertain. AI can speed up parts of the research process, yet it does not eliminate the need for laboratory testing, clinical trials, regulatory review, and long-term safety data. Nvidia’s role is less about creating drugs itself and more about supplying the computing systems and software that companies use to shorten the research cycle.
Kawasaki partnership brings AI into industrial machines
Nvidia is making a similar push into industrial automation through its partnership with Kawasaki Heavy Industries.
Kawasaki builds robots, aerospace systems, energy equipment, ships, rail vehicles, and heavy machinery. Its involvement gives Nvidia access to industrial environments where physical AI can be tested on machines performing real tasks.
The partnership fits Nvidia’s broader plan to become the computing foundation for robotics. The company already supplies chips and software used in autonomous vehicles, warehouse systems, humanoid robots, and industrial digital twins.
World models such as Cosmos 3 Edge could become a key part of that strategy. Robots need more than object recognition. They must estimate distance, predict movement, understand cause and effect, and respond safely when conditions change.
A robot moving through a factory, for example, may need to recognize a person stepping into its path, detect a misplaced object, and adjust its motion without stopping the entire production line. A medical robot may need to interpret visual information with far less room for error.
These systems remain difficult to build and test. Reliability, safety, training data, and real-world performance are still major barriers. A model that performs well in a simulation may behave differently in a crowded warehouse or hospital.
Nvidia is betting that Japan’s manufacturing base and engineering talent can help close that gap.
The company’s move into Japan reflects a wider shift in the AI industry. The first wave of generative AI centered on chatbots and software tools. The next phase is moving into machines that interact with the physical world.
For Nvidia, that transition could widen its reach far beyond cloud computing. For Japan, it offers a chance to connect its industrial history with a new generation of intelligent machines.
