Nvidia's vice president of automotive, Xinzhou Wu, sat for an interview on The Verge's Decoder podcast, addressing the company's autonomous driving strategy, its relationship with automakers, and the question every AV conversation eventually reaches: whether Tesla can deliver true self-driving without lidar The Verge.

Wu has spent three years at Nvidia as of the interview, having resigned as Xpeng's head of autonomous driving and joined the company in August 2023 — an exit he announced himself on Weibo with a photo alongside Jensen Huang Reuters. His current title is Vice President of Automotive, and he leads strategy, product planning, and engineering execution for that business unit NVIDIA.

The interview arrives with Nvidia's automotive stack now shipping in production vehicles. DRIVE AV software made its debut in the Mercedes-Benz CLA, and Nvidia's broader system is deployed across newer Mercedes EVs The Verge. That platform, DRIVE Hyperion, combines a high-performance onboard computer with a sensor architecture Nvidia opened up to third parties in 2022, built around redundancy and diversity of sensing modalities for safety-critical operation NVIDIA. The core development processes behind it carry ISO 26262 ASIL D certification, the top automotive functional-safety rating NVIDIA.

On the podcast, Wu described Nvidia's approach as blending a "classical" perception-and-planning stack with AI reasoning models rather than betting entirely on either paradigm The Verge. That framing puts Nvidia at odds with Tesla's camera-only, end-to-end neural network approach to Full Self-Driving. Asked directly whether Tesla FSD can hit Elon Musk's stated capability targets without lidar, Wu's response sits at the center of an argument that has run since Tesla abandoned radar and ultrasonic sensors in favor of vision alone. Wu has previously said Nvidia has spent billions of dollars over more than a decade building a full-stack autonomous driving solution, sensors included The Verge.

Wu also discussed the software-defined vehicle concept — consolidating dozens or hundreds of electronic control units onto a handful of centralized, powerful computers — a shift Nvidia has separately framed as evolving further into what Wu calls "AI-defined vehicles," where onboard models increasingly drive vehicle behavior rather than static code NVIDIA. Wu has described autonomous vehicles as the first mass-deployed physical AI agents NVIDIA, a framing consistent with Nvidia's positioning at recent GTC events, where Wu has presented sessions such as "Achieving Scalable Autonomous Vehicle Driving With NVIDIA" NVIDIA.

One of the more candid moments in the interview involved internal resource competition. Wu acknowledged he has to fight for compute capacity within Nvidia against the company's booming AI data-center business — the same silicon and fabrication capacity that Nvidia's largest and most lucrative customers are also drawing on The Verge. Automotive remains a comparatively small fraction of Nvidia's revenue against its data-center business, even though Nvidia has supplied chips to the auto industry for years and positions itself as a key supplier across the sector.

The Decoder interviewer also characterized the current US EV adoption cycle as "fully off track," a framing raised in the context of the broader conversation rather than a claim Wu endorsed outright. It's a notable gap between the state of AV software maturity — Nvidia's stack now in shipping Mercedes vehicles, DRIVE AGX Thor developer kits available to accelerate on-vehicle edge development NVIDIA — and the commercial reality of slowing EV sales growth in the US market that autonomous features are meant to help sell.

The interview follows a March 2026 Verge piece in which a reporter road-tested Nvidia's system against Tesla FSD directly The Verge, and a companion video published by Nvidia showing Huang and Wu driving a DRIVE AV-equipped vehicle through San Francisco NVIDIA. Nvidia used its CES 2026 special presentation to tie autonomous driving into its broader Rubin platform roadmap and a push toward open models NVIDIA, suggesting the automotive unit is being positioned less as a standalone product line and more as one expression of Nvidia's wider AI infrastructure ambitions.

The lidar-versus-vision debate Wu was pressed on is not new, but it has hardened into something closer to a proxy war between two philosophies of safety validation: sensor redundancy that adds cost and complexity, against a bet that sufficiently capable neural networks can extract equivalent information from cameras alone. Nvidia's public position, reinforced through Wu's remarks, favors hedging with both approaches rather than resolving the argument outright. Whether that hybrid stance proves durable as vehicles ship at scale, or whether one camp's approach clearly wins out in real-world safety data, is the open question the industry has not yet answered.