AI Chips AI 芯片
NPUs, GPUs and robot-specific SoCs — the compute substrate for perception, planning and on-board AI.
Modern robots increasingly think on board. Running perception, mapping and learned control policies in real time — within tight power and thermal limits on a mobile platform — takes purpose-built compute: GPUs and robotics SoCs that pack neural accelerators alongside conventional CPUs.
As embodied-AI models grow, on-board compute is becoming a primary design constraint and a strategic supply point. The platform that owns robot compute also tends to own the software stack and developer ecosystem around it, which is why chip vendors are investing heavily in robotics-specific modules and reference designs.
What to watch
- On-board inference for large embodied-AI policies
- Power and thermal budgets as the real ceiling
- Compute platform lock-in shaping the software stack
Key Players
Representative companies operating in this part of the value chain.
GPUs and Jetson/Thor robotics compute platforms powering on-board AI.
Visit website →Edge AI system-on-chips for autonomous machines and vehicles.