Much of the conversation around AI centres on increasingly capable AI models.

In Physical AI, however, intelligence is a stack of sensors, compute and software, all working as a single unit to deliver localised intelligence to the machine in the real world in near real time.

Yet intelligence alone does not enable a machine to operate in the physical world. Whether an autonomous vehicle navigates an unfamiliar road, a robotic system adjusts to changes on a production line, or a surveillance platform monitors activity in challenging environments, every decision depends first on the machine's ability to accurately perceive its surroundings and respond appropriately. That is why Physical AI represents more than an evolution in artificial intelligence; it marks a broader shift in how intelligent systems are conceived, designed and engineered.

Unlike digital AI, which works with information that already exists, Physical AI must continuously interpret a world that is dynamic, uncertain and constantly changing. Every action begins with data captured by sensors, processed by semiconductors and translated into decisions by intelligent software. These technologies have traditionally evolved as separate disciplines, each optimised for its own purpose. Physical AI is dissolving those boundaries. System performance is no longer determined by the capabilities of individual components, but by how effectively they function together.

This convergence is changing long-established engineering priorities. The reliability of an AI model depends on the quality of the sensor data it receives. Compute and memory determine how efficiently that information can be processed within latency, power and thermal constraints. Software, in turn, must interpret continuously changing inputs while maintaining accuracy, consistency and responsiveness. Improvements in one layer cannot fully compensate for limitations in another because every stage influences the next. This requires extreme hardware-software co-design, an area where new entrants are often better placed to build than incumbents.

The implications extend well beyond robotics. Industrial automation, autonomous mobility, and defence depend on machines that can understand and respond to the physical world with a high degree of reliability. As these systems become more capable and more widely deployed, engineering success will increasingly depend on designing semiconductors, including sensors, compute and memory, as well as software as an integrated foundation rather than as independent siloed technologies.

Physical AI is therefore changing more than the capabilities of intelligent machines. It is changing how the physical and digital worlds interact. The next phase of artificial intelligence will be defined not only by advances in algorithms, but by the convergence of sensing and semiconductor technologies that together form the foundation for machines operating in the built world.