Nvidia Becomes Uber's Full-Stack Autonomous Driving Partner
Nvidia and Uber have announced an expanded partnership that will see Nvidia's complete autonomous vehicle software platform power Uber robotaxis at commercial scale starting in the first half of 2027. The companies plan to target 28 markets across four continents by the end of 2028, beginning with Los Angeles and San Francisco. The announcement was made by Nvidia CEO Jensen Huang during his keynote at GTC 2026, Nvidia's annual artificial intelligence conference in San Jose.
The scope of the agreement marks a significant evolution from the companies' existing collaboration. Uber had previously agreed at CES this year to feed real-world driving data into Nvidia's AI training pipeline. That arrangement has now expanded into something far more comprehensive: Nvidia will build the full autonomous vehicle stack for Uber at scale — covering perception, prediction, planning, and control — while also handling the training infrastructure and simulation validation that underpin any commercial AV deployment.
What Nvidia's AV Stack Actually Does
Autonomous vehicle development requires solving several interconnected engineering problems simultaneously. The Nvidia platform addresses each layer of the stack. The perception system processes data from cameras, lidar, and radar to construct a real-time 3D model of the vehicle's surroundings. The prediction module anticipates the movements of pedestrians, cyclists, and other vehicles. The planning system charts a path through complex traffic scenarios, and the control layer translates that plan into steering, acceleration, and braking inputs.
Behind the deployed system sits a vast training and simulation infrastructure. Nvidia's DRIVE platform enables AV companies to train neural networks on synthetic data generated in simulation, allowing developers to expose their systems to rare or dangerous scenarios that would be impractical to collect in the real world. Once models perform satisfactorily in simulation, they can be validated against real-world data and progressively deployed.
Ali Kani, Nvidia's vice president of automotive, described the arrangement: Uber is asking Nvidia to be the software partner in 28 cities, starting with Los Angeles and San Francisco. The phrasing reflects Uber's evolving strategy — rather than developing autonomous technology in-house after divesting its AV unit to Aurora in 2020, the company has positioned itself as the distribution network for whoever builds the best self-driving technology.
