NewsNTech
The constraint in physical AI is inference latency: a robot arm or autonomous machine running a real-time control loop cannot wait on a cloud compute round-trip, which has pushed the industry toward edge-deployed models sized to run locally on the hardware doing the physical work.
Nvidia has placed Cosmos 3 Edge inside that layer, announcing the new model alongside an expansion of its physical AI ecosystem in Japan.
What Cosmos 3 Edge represents in the stack Physical AI systems generate sensor data, run perception models, and produce control outputs in a continuous loop. Each step competes for the same millisecond budget.
When that budget is tight, the model must be local. Cloud inference adds variable latency that is acceptable in a chatbot and fatal in a robot.
Keep reading