Physical AI is moving from an emerging label into a defined engineering discipline. The Physical AI Expo North America, scheduled for May 18–19, 2026 in San Jose, is built around the systems that allow intelligence to sense, reason and act in the real world.
The focus marks a shift from chatbots and digital workflows toward robots, autonomous machines and industrial automation. These systems must operate amid physical constraints, changing environments and people, making reliability and infrastructure as important as model capability.
The event brings together robotics specialists, enterprise technologists, developers and infrastructure providers to examine how prototypes become dependable production systems.
Physical AI connects models with machines
Software-based AI produces information. Physical AI converts perception and reasoning into movement, inspection or control. A warehouse robot may identify an object, plan a route and handle it; an industrial system may interpret sensor readings and adjust a process.
That chain requires more than a capable foundation model. Systems need sensors, spatial reasoning, low-latency compute, dependable connectivity and mechanical control. They also need feedback that confirms whether the intended action actually succeeded.
The conference agenda reflects this full stack. Topics include enterprise robotics, autonomous systems, industrial automation, data platforms, compute, developer tools and human-AI collaboration. Reliability, transparency and safety are treated as production requirements rather than separate ethical discussions.
That breadth is necessary because physical deployment exposes dependencies that a software demonstration can hide. A model may reason correctly while a sensor supplies stale data, a gripper slips or network latency makes the instruction unsafe.
Production scale is the central challenge
Day one of the event focuses on enterprise strategy, autonomous intelligence and the infrastructure required to support deployment. Day two moves deeper into robotics, real-world systems and the engineering path from prototype to production.
The distinction captures an important market tension. Many organisations can demonstrate a robot completing a selected task in controlled conditions. Far fewer can maintain performance across different sites, equipment, workers and unexpected events.
Production readiness requires repeatable evaluation, service procedures and clear ownership across hardware and software teams. Organisations must know how updates are tested, how failures are logged and how a machine reaches a safe state when conditions exceed its operating boundary.
Economic evidence matters too. Physical systems require capital, integration and ongoing maintenance. A compelling deployment needs measurable throughput, quality or safety improvements rather than a novelty premium.
Industry convergence reveals where value may emerge
Speakers include leaders from Nvidia, Airbus Acubed, Qualcomm, Hitachi, Hyundai’s software organisation, JPMorgan and General Robotics. Their backgrounds span chip design, aviation, manufacturing, finance and robotics research.
That mix demonstrates that Physical AI is not one product category. It is a convergence of models, compute, machinery and domain operations. Infrastructure vendors may provide the processing layer, robotics companies the embodiment and industrial organisations the environments and process expertise.
Manufacturing and logistics are likely early markets because they offer structured sites, repetitive tasks and measurable outputs. Automotive and defence bring greater autonomy and safety demands, while service industries introduce more unpredictable human interaction.
The strongest partnerships will combine domain knowledge with technical capability. A model developer cannot define every factory constraint, and an industrial operator may lack the expertise to evaluate embodied reasoning. Joint testing can expose risks before systems are connected to critical equipment.
San Jose is an appropriate setting because the next phase of AI will depend on more than improvements inside data centres. It will require a surrounding ecosystem of sensors, edge compute, actuators, simulation, safety controls and skilled integration.
The Physical AI Expo is itself an industry event, so its programme should be viewed as a map of commercial priorities rather than proof that every promised deployment is mature. Even so, the agenda shows where investment and engineering attention are converging.
The defining question for Physical AI is no longer whether machines can demonstrate intelligent action. It is whether organisations can make that action reliable, governable and economically useful across the untidy conditions of the real world.
