Nvidia's auto chief: cars are becoming AI-defined
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Xinzhou Wu, Nvidia's head of automotive and former XPeng autonomous-driving lead, frames the industry as evolving through three eras — mechanical, 'software-defined vehicle,' and now 'AI-defined vehicle,' where generative AI rewrites most of the car's software instead of merely running on it.
- China's auto sector completed a wholesale pivot to central-compute architectures between 2018 and 2023, with both new and legacy Chinese OEMs consolidating dozens of ECUs into one or two computers because, per Wu, 'that's how you compete.'
- Legacy US automakers like GM and Ford have largely failed to consolidate their ECUs into central compute; GM told Decoder it doesn't need to, while Ford's skunkworks-built truck remains untested, and only EV-native startups like Tesla and Rivian have succeeded.
- Mercedes-Benz has adopted Nvidia's central-compute platform across its current vehicle generation, and Wu said Nvidia is working with 'all' basic OEMs globally to push them toward one-or-two-computer architectures.
- Nvidia's automotive unit competes internally against the company's booming AI business for compute, engineering talent, and capacity; Wu described his three years navigating that internal fight as 'a rapid learning experience.'
- Chinese OEMs had a structural advantage over US incumbents because they designed cars as EVs from a clean sheet rather than transitioning from gas platforms weighed down by legacy ECUs, Wu confirmed, with the Chinese government subsidizing the shift.
Why it matters: Nvidia's automotive team — shipping central-compute systems to Mercedes and courting every major OEM — faces a structural bottleneck: its engineers must outmaneuver Nvidia's own AI cloud customers for limited compute capacity. Carmakers still running dozens of independent ECUs, most legacy US players, risk falling further behind Chinese OEMs and EV-native startups that have already consolidated their architectures.
Ask SkimNews
