Even Nvidia's automotive unit has to fight for GPUs — weekly allocation talks can reach Jensen Huang
TL;DR
Xinzhou Wu said Nvidia's automotive team negotiates GPU allocations almost every week.
An unusual internal shortage: Nvidia automotive chief Xinzhou Wu said even Nvidia has a limited supply of GPU compute, forcing his team to coordinate almost weekly with other groups over capacity for training, testing, and different workstreams. The automotive unit does not have unrestricted access to the GPUs Nvidia produces, and some allocation decisions require CEO Jensen Huang's involvement.
Wu told The Verge's Decoder that internal prioritization considers revenue, strategic priorities, and the size of new markets. Nvidia weighs current income against what Huang calls a "zero trillion dollar business": a market with little current revenue that the company believes could grow into a trillion-dollar opportunity.
Wu said the automotive organization employs thousands of people, mostly in the United States, with teams in China and Europe. It uses GPUs, CPUs, Cosmos, and Nemotron from Nvidia's centralized hardware and software groups, then adapts models, operating systems, and chip platforms to automotive requirements. Each generation of automotive chips also carries a support commitment of 10 to 15 years.
The organization consumes large amounts of compute for autonomous-driving models. Wu said Nvidia runs about five million simulation tests each day and roughly 10 iterations of its end-to-end model daily. A classical stack and an AI stack run in parallel inside the vehicle to compare trajectories frame by frame. Nvidia sells GPUs externally, but the automotive team's allocation still comes from weekly internal resource decisions.
via The Verge
Wu told The Verge's Decoder that internal prioritization considers revenue, strategic priorities, and the size of new markets. Nvidia weighs current income against what Huang calls a "zero trillion dollar business": a market with little current revenue that the company believes could grow into a trillion-dollar opportunity.
Wu said the automotive organization employs thousands of people, mostly in the United States, with teams in China and Europe. It uses GPUs, CPUs, Cosmos, and Nemotron from Nvidia's centralized hardware and software groups, then adapts models, operating systems, and chip platforms to automotive requirements. Each generation of automotive chips also carries a support commitment of 10 to 15 years.
The organization consumes large amounts of compute for autonomous-driving models. Wu said Nvidia runs about five million simulation tests each day and roughly 10 iterations of its end-to-end model daily. A classical stack and an AI stack run in parallel inside the vehicle to compare trajectories frame by frame. Nvidia sells GPUs externally, but the automotive team's allocation still comes from weekly internal resource decisions.
via The Verge
