NXP CEO Rafael Sotomayor delivered a Keynote address at COMPUTEX 2026, highlighting the potential of physical AI and its real-world applications.
Sotomayor called attention to how much of human motion is not conscious, rather, it’s driven by reflexes. Turning to his own company’s technology, he spoke at length about how NXP applies that to its physical AI products.
Physical AI devices must be low-latency, low-power, and high-security, he stressed. At the edge, in the real world, it is necessary for devices to quickly make their own decisions, like how a human instinctively reacts to situations, he said.
“You don’t scale intelligence by making the brain bigger and bigger, scale intelligence by putting intelligence at the right place,” he said. For physical AI devices, that means some local processing and decision-making capabilities to avoid latency. Furthermore, edge devices need distributed control, to avoid single points of failure, and need to be extremely energy efficient. Leaning on these principles, he highlighted NXP’s work on neural access architecture.
He held up NXP’s innovations in the drone and automotive sectors as examples of this architecture. In a vehicle, the mission critical functions are part of these “reflexive” layers, reacting with microseconds and working alongside other forms of intelligence to ensure it can safely operate, while other intelligent systems do things like navigate and ensure it has stable motion.
“The question is, how do we teach a robot not only how to move, but how to understand?” he asked.
“You need a bridge between perception and understanding. Perception tells you what’s in front of you. The understanding of the world tells you what is going to happen when you interact with that object in front of you,” he said.
Sotomayor pointed out that while humans get years to experience the world and build that understanding, a core challenge for robots is that they do not have the time. NXP’s solution is Vision Language Action Models (VLAs), trained on the cloud, and then deployed at the edge using company’s elQ toolkit. That software tunes the VLAs for use in the real-world and ensures physical AI can be deployed.
“Our goal is to remove objections from a customer,” he said.
A second challenge is trust. Sotomayor emphasized that machines must be trustworthy from the first moment they are deployed. NXP’s solution is a framework that isolates problems, ensures redundancy, and has security injected directly into the hardware. That security includes post-quantum cryptography, to make sure that devices are future-proof and can continue to evolve to meet new threats.
Sotomayor highlighted that NXP is at the forefront of the push to ensure customers can confidently deploy AI in the real world.
“Our job right now is not really to push physical AI before our customers are ready for it. Our job is to actually making sure physical AI has a place to be,” he said.
