Xinzhou Wu, Nvidia’s head of automotive, says even his own division has to argue for scarce GPU capacity inside a company riding the AI boom. In an interview with The Verge’s Decoder, Wu said Nvidia’s automotive group meets with colleagues “almost on a weekly basis” to allocate compute for training, testing, and other internal work, and that CEO Jensen Huang sometimes gets pulled in to settle priorities.
That is the awkward position for Nvidia’s car business. The company is one of the most valuable firms in the world because AI developers cannot buy enough of its GPUs. Wu’s team is trying to use the same corporate machinery to sell automakers chips, software, models, and safety systems for autonomous driving, a market that moves more slowly and demands support for far longer than the data-center crowd usually tolerates.
Wu said Nvidia’s automotive organization has “thousands” of employees. Most are in the United States, with teams also in China and Europe. He described the group as a dedicated automotive unit that draws on Nvidia’s centralized hardware and software teams, plus cross-company work on foundation models such as Nemotron and Cosmos.
Cars are being rebuilt around fewer computers
Wu argued that the auto industry is moving toward what carmakers call the software-defined vehicle: a car controlled by a small number of powerful computers rather than a mess of separate electronic control units, or ECUs, scattered across each subsystem. That shift allows over-the-air updates and makes it easier to add driver-assistance and autonomous features after sale.
Wu said the next step is what Nvidia calls an “AI-defined vehicle,” where generative AI helps write or replace much of the software inside the car. That is Nvidia’s preferred framing, and it conveniently points back to the company’s strongest business. Still, the mechanism is plain enough: consolidate the car’s electronics, run more workloads on shared compute, and treat driving software as something that can be trained, updated, and improved rather than frozen at the factory.
According to Wu, Chinese automakers moved faster because they had less legacy architecture to drag along, and because both newer brands and established global players in China were forced to match the local industry’s pace. Wu previously worked at Chinese automaker XPeng and said that experience shaped his view of how quickly the shift can happen.
He said Nvidia is working with major automakers on the same transition elsewhere. He pointed to Mercedes as a partner using a central-computer-based architecture, and The Verge said Nvidia’s autonomous driving system is already used in newer Mercedes electric vehicles.
Nvidia is selling more than chips
Wu said Nvidia’s automotive pitch spans hardware, operating systems, open-source models, data infrastructure, high-definition maps, autonomous-vehicle software, and Halos, its safety operating system for autonomous vehicles. He said maps remain important for higher levels of autonomy, including Level 3 and Level 4 systems.
The catch is the auto industry’s product cycle. Wu said suppliers must be prepared to support a car platform for 10 to 15 years, including current-generation chips. From a Silicon Valley point of view, he called that commitment almost absurd. From an automaker’s point of view, it is table stakes: cars stay on roads for years, and recalls are more expensive than app updates.
Inside Nvidia, Wu said compute allocation is judged by revenue, market size, and longer-term strategic bets. He said Huang talks about searching for “zero trillion dollar” businesses, meaning markets that do not yet exist at scale but could become major businesses. Autonomous vehicles are one of those bets, according to Wu.
Wu’s broad claim is that autonomy will become necessary for automakers and that “everything that moves” will eventually be autonomous. That remains a supplier’s forecast, not a delivered fact. For now, even Nvidia’s auto boss has to wait in line for the GPUs needed to prove it.
This story draws on original reporting from The Verge.
