Nvidia auto chief says mainstream Level 4 arrives in under 5 years

The most interesting thing Nvidia's automotive chief said this week was not the autonomy timeline, it was the admission that even inside Nvidia, GPUs are rationed. Xinzhou Wu, VP of automotive, told Nilay Patel on The Verge's Decoder podcast that he meets other division leaders "almost on a weekly basis" to divvy up training, testing and development compute, and that CEO Jensen Huang sometimes has to referee. "Even Nvidia has a limited supply of GPU for compute," Wu said, which is a line worth sitting with if you are an OEM banking on Nvidia's Drive stack landing on time.

On the roadmap itself, Wu was more bullish than the industry consensus. He predicted mainstream Level 4 autonomy, where the car drives itself without a human, will arrive in consumer vehicles in "less than five years." More concretely, he said Nvidia's ADAS technology will ship "on the ADAS side in all Mercedes vehicles and some other partners as well, all over the United States" by the end of 2026, per coverage of the interview summarized by TechBuzz. The scale claim behind that push is Nvidia running roughly 5 million validation tests a day on synthetic data, an argument that simulation, not fleet miles, is now the bottleneck-breaker.

Why this matters if you are not building an AV stack yourself: the mental model where Tesla's fleet data is an unbeatable moat gets softer if a lidar-inclusive, simulation-heavy competitor can credibly claim to be closing the gap. Wu conceded that "for the basic L2++ technology, Elon is probably ahead of everybody," but framed Level 4 as "more open," which is Nvidia's polite way of saying the race is not decided. The strategic wrinkle is China. Wu, who spent five years at XPeng before joining Nvidia, said the company still supplies in-car inference chips to Chinese OEMs and collaborates via open-source models like Cosmos and Alpamayo, which threads a needle around US export controls that could tighten at any point.

The honest caveat is that almost all of this is a single-sourced pitch from Nvidia's own VP on a podcast. The reporting does not give you a per-quarter GPU allocation, does not name the non-Mercedes US partners, and does not spell out how the promised small-scale 2026 Level 4 trial handles liability or state approvals. Take the specifics as reported, not settled.

The forward-looking read is simple. If Wu is even directionally right, the OEMs that lock in Nvidia Drive partner status now get access to shared validation data and simulation scale they cannot build alone, and the ones still deciding will be queued behind Nvidia's own training runs at a moment when Nvidia is openly saying its own chips are the bottleneck.