'Full-stack autonomous driving' is a phrase that means different things in a keynote and in a 10-K. In its FY2026 annual report — indexed by EdgarBeast and filed at sec.gov — Nvidia writes that, with its ecosystem partners, it is delivering a full-stack end-to-end solution for the AV market under the DRIVE Hyperion platform.
Read the qualifier: 'with our ecosystem partners.' Nvidia is not claiming to be Waymo. It is claiming to supply the layer underneath everyone who wants to be Waymo. Across multiple 10-Ks the description is consistent — DRIVE Hyperion bundles the high-performance, energy-efficient DRIVE AGX compute with a reference sensor-and-software stack so an automaker can run perception, planning, and control inside the vehicle on Nvidia's platform.
That is the boring incumbent's move done at the platform level. Rather than pick a winner in the camera-versus-LiDAR religious war, Nvidia sells the compute and the reference architecture that either camp needs. A vision-only customer and a mapped-LiDAR customer can both buy DRIVE AGX. Sensor neutrality is not a philosophical stance here; it is a market-coverage strategy you can read in the filing.
It also clarifies what 'end-to-end' is doing. It does not mean Nvidia ships a finished self-driving car. It means the platform spans the chain from sensor input to control output as a coherent reference design, so a partner is not stitching together compute, middleware, and a software stack from scratch. The end-to-end claim is about integration completeness, not about owning the deployed service.
Strip the demo lighting and the sec.gov description is precise about Nvidia's actual position in autonomy: it is the supplier of the compute-and-platform layer, deliberately agnostic to whose car and whose sensor philosophy sits on top. That is a more defensible place to stand than any single robotaxi bet. Evidence via EdgarBeast.