The arms-dealer thesis is a compounding thesis, and 2025 shows it widening. NVIDIA Corporation's May 20, 2025 grant US12307788B2 claims sensor fusion for autonomous machine applications using machine learning — note "machines," not just "vehicles."

Read the generalization in the claim. The CPC tags — G06V 20/588 and G06V 20/58 scene perception, B60W 60/00272 autonomous maneuvering, G06T 7/292 tracking, B60W 2556/35 sensor context — describe perception software framed for autonomous machines broadly. NVIDIA is extending the car perception stack into the wider robotics field its Isaac platform targets.

“In various examples, a multi-sensor fusion machine learning model—such as a deep neural network (DNN)—may be deployed to fuse data from a plurality of individual machine learning models.”— U.S. Patent No. 12,307,788 source

The capex read is that NVIDIA captures perception-software margin across an expanding surface. Where earlier grants fenced off automotive perception, this one fences off autonomous-machine perception generally — robots, drones, industrial systems. Every robotics builder running on NVIDIA compute is a candidate customer for NVIDIA perception software.

For a public-equities reader, this is the arms-dealer position compounding. NVIDIA sells the compute to nearly everyone in robotics; owning the perception software on top means capturing margin at two layers across the entire field, not just cars. The switching costs accumulate.

The honest limit: a fusion patent is a method, not a market share, and it does not disclose how much robotics-perception revenue NVIDIA books or at what price. It establishes deliberate IP accumulation at the perception layer across all autonomous machines.

The takeaway for the money desk: NVIDIA's autonomy moat is widening from cars to all robotics. Read its machine-perception patents as the arms dealer extending the same two-layer margin capture across the whole field.