Physical AI.
The mechanical & motion picks-and-shovels behind every humanoid, robotaxi and autonomous platform. NVIDIA's GR00T + Isaac + Cosmos = the compute platform. The 0 names in this basket = the wafers, drives, gears, sensors and magnets it actually runs on.
You cannot 3D-print a strain-wave gear. You cannot software your way past an NdFeB magnet. Every humanoid in the field today carries 24-40 actuators, 60+ sensors, ~3kg of rare-earth magnetics and a sensor-fusion FPGA that nobody else makes. Physical AI is a six-layer mechanical stack — and unlike the GPU race, the bottlenecks here have decade-long lead times.
★ NVDA Q1 FY27 print (May 20): Physical AI revenue >$9B TTM · Edge Computing $6.4B Q1 +29% YoY · Uber robotaxi 2028. Our prior robotics report (38 names) validated by Jensen's framing — and was conservative.
Our robotics thesis survived the earnings test — and was conservative.
We mapped the 38-name Physical AI basket on May 14. Six days later Jensen ran the same playbook on the earnings call. The framework was right — what surprised us was the velocity. We modelled this as a 2027-28 P&L story; the print shows it is already a $9B TTM line two years ahead of the curve.
From motor to manipulation.
Six physical layers behind every humanoid and robotaxi.