Camera-only 3D perception pressures the lidar mid-market

Winners: camera-first robotics companies and consumer-robot makers — perception capability at $30 of hardware compounds directly into gross margin. GPU vendors win either way; camera-only autonomous-driving stacks get fresh validation.

Pressured: lidar manufacturers in mid-tier applications — warehouse robotics, delivery, drones — where near-lidar accuracy at 1% of the cost is good enough. Premium safety-critical lidar (highway autonomy) keeps its moat for now.

Signals: watch for design wins where lidar gets dropped from a shipping product's BOM; replication of the 4% gap by independent teams; follow-up results in rain, fog, and night conditions, the documented weak spots.

Difficulty to commercialize: 6/10. Weights are public but non-commercial, so productization means retraining; integrating perception into certified robotic systems carries a long validation tail, and the adverse-weather gap caps near-term addressable applications.