The compute layer for physical AI: companies, chips and open-source projects that run AI models on and for robots. Covers on-robot SoCs, robot compute boards, edge-AI accelerator chips, runtimes, compilers and quantization, VLA and robot-model serving stacks, edge-cloud split inference and teleop latency, and fast-inference robot foundation models. 41 entries, a selection by Abdul Al Ali (not a census), facts as of 8 Nov 2026.