The first wave of the AI revolution was a cloud story. Large language models trained on GPU clusters in data centres, inference served over the internet, intelligence delivered as a service from a centralised computation infrastructure. The second wave is a physical story — AI moving out of data centres and into the world, embedded in the machines, vehicles, robots, and devices that inhabit physical space and operate in real time on real-world data.
This physical AI wave depends on a new category of semiconductor: the edge AI chip. Where data centre GPUs prioritise peak performance with little constraint on power consumption and physical size, edge AI chips must deliver useful AI inference within strict power budgets — the battery capacity of an autonomous vehicle, the thermal envelope of a surgical robot arm, the solar panel output of an agricultural drone. Meeting these constraints while maintaining enough computational capacity to run meaningful AI models requires wafer fabrication at the most advanced process nodes available.
The Physical AI Hardware Stack
The physical AI hardware landscape is extraordinarily diverse. At the top end — autonomous vehicles and humanoid robots — NVIDIA's Drive and Jetson platforms, Qualcomm's Snapdragon Ride, and Tesla's custom FSD chips provide the compute for real-time perception, planning, and control. These chips are manufactured on advanced process nodes at TSMC and Samsung, using the same wafer fabrication infrastructure as data centre AI accelerators.
At the intermediate level — smart cameras, industrial inspection systems, warehouse automation — a range of purpose-built AI inference chips from companies including Hailo, Kneron, and Ambarella provide more limited but highly efficient inference at lower power and cost. And at the edge — IoT sensors, embedded controllers, hearing aids, and smart meters — ultra-low-power AI chips from ARM, STMicroelectronics, and Nordic Semiconductor are bringing basic machine learning inference into devices with milliwatt power budgets.
"Physical AI is what happens when machine intelligence leaves the data centre and enters the world. Every autonomous vehicle, every surgical robot, every warehouse AGV is a wafer-derived AI chip operating in real time in physical space."
Agentic AI at the Edge
The most sophisticated physical AI systems are beginning to exhibit agentic behaviour — not just perceiving and responding, but planning, reasoning about goals, and executing multi-step actions toward objectives. Tesla's Full Self-Driving system plans routes, anticipates other drivers' behaviour, and executes complex manoeuvres without human instruction. Boston Dynamics' Atlas robot plans manipulation tasks, adapts to unexpected obstacles, and completes objectives in unstructured environments.
These agentic physical AI capabilities require more computational capacity than simple inference — they need enough silicon to run planning algorithms, world models, and real-time control loops simultaneously. The wafer-derived chips meeting these requirements represent the frontier of physical AI silicon design. AIWafers.com covers this entire spectrum from wafer to agentic physical intelligence.
The Domain for Physical AI Silicon
AIWafers.com — from fab to field, the domain covering the complete physical AI silicon stack.
Acquire This Domain →