Xinzhou Wu, Nvidia's Vice President of Automotive, sat down on The Verge's Decoder podcast to discuss the company's autonomous driving strategy and a question that dominates every AV conversation: whether Tesla can achieve true self-driving capability using cameras alone, without lidar sensors The Verge.

Wu joined Nvidia in August 2023 after leaving Xpeng, where he headed autonomous driving. At Nvidia, he oversees strategy, product planning, and engineering for the automotive unit. His three-year tenure arrives as Nvidia's DRIVE AV software now appears in production Mercedes-Benz vehicles, marking the first real-world deployment of Nvidia's full autonomous driving stack The Verge.

Nvidia's approach centers on DRIVE Hyperion, a platform combining onboard computing power with multiple sensor types — cameras, radar, lidar — arranged for redundancy. In 2022, Nvidia opened this sensor architecture to third parties. The underlying code carries ISO 26262 ASIL D certification, the highest functional-safety rating in automotive engineering NVIDIA.

On the podcast, Wu described Nvidia's philosophy as a hybrid: blending traditional perception-and-planning software with AI reasoning models, rather than betting everything on either approach alone The Verge. This positions Nvidia against Tesla's camera-only, end-to-end neural network strategy. When pressed on whether Tesla's Full Self-Driving can reach Elon Musk's stated targets without lidar, Wu's answer reinforced Nvidia's position: the company has spent over a decade and billions of dollars assembling a full-stack solution precisely because sensor diversity matters for safety.

Wu also discussed the evolution from software-defined vehicles — where dozens of control units consolidate onto a few powerful computers — toward what he calls AI-defined vehicles, where machine-learning models increasingly direct vehicle behavior rather than pre-written code. Nvidia has positioned autonomous vehicles as the first mass-produced physical AI agents, a framing that ties the automotive unit to Nvidia's broader AI infrastructure narrative NVIDIA.

One candid moment in the interview involved internal resource pressure. Wu acknowledged competing within Nvidia for chip fabrication and computing power against the company's far larger AI data-center business — the same silicon his automaker partners are also chasing The Verge. Automotive remains a small slice of Nvidia's revenue, though the company has supplied automotive chips for years.

The interview arrives amid a visible gap between technical maturity and market reality. Nvidia's DRIVE AV stack now ships in Mercedes vehicles; DRIVE AGX Thor developer kits are available for edge computing development NVIDIA. Yet U.S. EV sales growth has stalled — the very market autonomous features are designed to catalyze.

The lidar-versus-cameras debate Wu fielded is shorthand for a deeper split in safety philosophy: sensor redundancy, which adds cost and design complexity, against the claim that sufficiently powerful neural networks can extract the same information from cameras alone. Nvidia favors a both-and approach rather than choosing sides. Whether this middle ground holds as vehicles scale to millions of units, or whether real-world safety data eventually vindicate one camp, remains unanswered. Nvidia has tied autonomous driving to its Rubin platform roadmap and an emphasis on open models, signaling it views the automotive business as one expression of its broader AI infrastructure ambitions, not a standalone product line NVIDIA.