Nvidia auto chief says his team competes for scarce GPU compute
Xinzhou Wu told The Verge that Nvidia’s automotive group must vie internally for GPU capacity as the company’s AI business strains supply.
By Maya Lindqvist · Senior Technology Correspondent
3 min read
Nvidia’s automotive business is competing inside the company for the same scarce computing resources that power the broader AI boom. Xinzhou Wu, Nvidia’s head of automotive, told The Verge’s Decoder podcast that his team works with other groups nearly every week to decide how GPU capacity is assigned.
Wu said Nvidia has an internal priority system for compute used in training, testing and other work. He said Nvidia CEO Jensen Huang sometimes gets involved when teams need help settling those decisions.
The comments show how the demand for Nvidia’s chips affects even units inside the company. Nvidia has become one of the world’s most valuable companies as AI developers race to buy its GPUs, while its automotive group is trying to sell carmakers on more powerful in-car computers and autonomous-driving systems.
Wu said the debate over resources includes current revenue, future market potential and strategic value. He described Nvidia as a company that looks for new businesses that could become very large, while also weighing what generates money now.
Automotive group builds on Nvidia’s AI work
Wu said Nvidia’s automotive team numbers in the thousands and works across hardware, software, models and infrastructure. The group is based mostly in the United States, with staff also in China and Europe, he said.
According to Wu, Nvidia’s automotive unit is a separate organization that adapts work from the company’s centralized hardware and software teams for carmakers. He said the group also draws on Nvidia foundation-model projects including Nemotron and Cosmos.
Wu said Nvidia uses virtual teams that span research, software and hardware groups to work on open-source foundation models. The automotive organization can then build on that work for autonomous-vehicle development, he said.
He also described Nvidia as organized around centralized chip work, including GPU, CPU and other chipset planning. The auto team’s role, he said, is to turn that companywide technology into a platform for vehicles.
Wu says car architecture is changing
Wu told The Verge that the auto industry is moving from “software-defined” vehicles toward what he called “AI-defined” vehicles. He said cars have shifted from mainly mechanical and electrical systems toward machines whose abilities can be upgraded through over-the-air software.
He said automakers are moving toward architectures built around one or two central computers instead of many separate electronic control units. Wu said he saw that shift accelerate in China between 2018 and 2023, where both newer automakers and established companies had to adapt to compete.
Wu said Nvidia is working with global automakers on that transition. He cited Mercedes as a partner using a central-computer-based architecture in its current generation of vehicles.
The shift is slower for some carmakers because the auto business carries long obligations, Wu said. He said Nvidia, as a supplier, commits to supporting automotive chipsets, platforms and autonomous-vehicle technology for 10 to 15 years, which can slow change compared with Silicon Valley software cycles.
Nvidia is offering automakers more than chips, Wu said. He said the company provides autonomous-driving technology, operating systems, open-source models and Halos, which Nvidia describes as a safety operating system for autonomous vehicles.
This story draws on original reporting from The Verge.