scripod.com

Yes, even Nvidia's head of automotive is fighting for compute

Decoder with Nilay Patel
In this episode, the head of Nvidia's automotive division, Xinzhou Wu, provides an insider's perspective on the electric vehicle and autonomous driving landscape. He discusses the technological shifts, strategic challenges, and competitive dynamics shaping the industry, particularly in the US and China.
Xinzhou Wu explains the transition from software-defined to AI-defined vehicles, where AI rewrites car software, and notes that Chinese automakers have a head start due to building on EV platforms. He outlines Nvidia's strategy of providing a full platform—including chips, operating systems, and open-source models—to help OEMs mass-produce autonomous technology. Wu contrasts Nvidia's open platform approach with Tesla's vertical integration, arguing that LiDAR is necessary for Level 4 safety and redundancy. He addresses safety through ISO 26262 standards and a redundant stack combining an end-to-end neural model with a classical safety-based system. Wu predicts Level 4 autonomous driving will become mainstream in less than five years, citing Waymo as the only current safe deployer, and discusses how data-sharing challenges between OEMs lead to regional model variations.
06:18
06:18
AI rewrites car software
20:42
20:42
Capturing a share of the 13 trillion miles driven annually.
36:38
36:38
Synthetic data closes the data gap.
46:55
46:55
The model can reason and explain its driving decisions in language.
1:01:54
1:01:54
LiDAR is necessary for Level 4 safety and redundancy.
1:12:26
1:12:26
Level 4 will become mainstream in less than five years