The world's largest industrial robot maker just bet its factory floor on AI that sees, thinks, and moves—and open-sourced the brain.
The Summary
- ABB Robotics announced a partnership with Nvidia at Computex, focused on integrating physical AI into manufacturing systems
- Nvidia released Cosmos 3, the first open omni-model for physical AI reasoning and action, enabling robots to understand and respond to real-world environments
- ABB, which ships more industrial robots than anyone, is now building on foundation models that let machines reason about the physical world—not just follow programmed paths
- The open model approach means every warehouse, factory, and logistics operation can now access the same AI capabilities powering ABB's industrial systems
The Signal
ABB Robotics revealed its Nvidia partnership from the Computex floor, with Business Line Managing Director Craig McDonnell explaining how physical AI will reshape manufacturing. This is not a research collaboration. ABB is the world's dominant industrial robotics supplier, and they're integrating AI that can perceive, reason about, and act in physical space.
The timing syncs with Nvidia's launch of Cosmos 3, which they're calling the first open omni-model for physical AI. Omni-model means it handles multiple modalities: vision, spatial reasoning, manipulation planning. Physical AI means it reasons about cause and effect in the real world, the kind of common-sense physics humans take for granted but robots have historically been terrible at.
"The first open omni-model for physical AI reasoning and action gives every robotics builder access to foundation models trained on real-world physics."
Here's why this matters beyond one partnership:
- Industrial robots today follow precise, pre-programmed paths. They break when anything unexpected happens.
- Physical AI models learn from sensor data to adapt in real time—routing around obstacles, adjusting grip pressure, recovering from errors.
- Open-sourcing this model means startups, research labs, and regional manufacturers can build on the same foundation ABB is using.
The convergence is deliberate. Nvidia needed a massive real-world deployment partner to prove physical AI works at scale. ABB needed AI capabilities to compete as manufacturing gets more dynamic and less predictable. Traditional robotic programming takes weeks for each new task. Foundation models can generalize across tasks with fine-tuning measured in hours.
The Computex announcement setting is strategic. Taiwan is the semiconductor capital and an advanced manufacturing hub. This is where you debut technology that will scale globally.
The Implication
If ABB is betting its installed base on physical AI, expect every industrial automation company to follow. The compute advantage shifts from whoever has the most engineers programming robots to whoever has the best training data and model access. That's a different game with different winners.
Watch for the speed of deployment. If ABB can roll out adaptive robots faster than competitors can hire robotics engineers, the labor economics of manufacturing shift again. And if the model is open, the real competition moves to implementation, integration, and the quality of sensor data feeding these systems. The robots are about to get much harder to outthink.