RoboBusiness talk: Physical AI's bottleneck is deployment, says Inbolt CEO

The robotics industry often treats data as the main hurdle for physical AI, focusing on better simulation and curation. Cohen counters that this applies in the lab, not the factory, where the integration tax around perception and motion is the actual constraint. His company, Inbolt, applies vision-guided control across its factory footprint, suggesting a practical alternative to the data-centric view.

Cohen's background includes applied mathematics at École Normale Supérieure Paris-Saclay and entrepreneurship studies at École Polytechnique and HEC Paris. Inbolt has raised €20 million ($23.2 million U.S.) since 2019 and operates in factories across Europe, the U.S., and Japan. The RoboBusiness talk will use production data from Stellantis, Toyota, and Ford to argue physical AI is already commercially proven, though perhaps not in the areas receiving the most attention.

The stakes for this argument are significant for manufacturing decision-makers. If Cohen's deployment-focused perspective gains traction, it could shift investment toward real-time control integration rather than model scale. The outcome likely hinges on whether the evidence from major automakers convinces attendees that closing the loop at servo frequency delivers more value than pursuing larger models, potentially reshaping priorities for robotics deployments.