Uber will use Nvidia’s new generative world model simulation tool, Cosmos, and cloud-based AI supercomputing platform, DGX Cloud, to support the development of autonomous vehicle technology, the companies announced at CES 2025.
Cosmos is being marketed to robotics and autonomy companies as a tool that generates physics-based videos from a variety of inputs, including 9,000 trillion tokens generated from 20 million hours of (likely copyrighted) video, which can be used to simulate realistic industrial and driving environments. Nvidia DGX Cloud would provide Uber and other companies with access to high-performance AI infrastructure to train, fine-tune, and deploy AI self-driving models.
Uber shared few details about how it plans to use these Nvidia tools. The ride-hail and delivery giant has spent the last year racking up 14 partnerships with AV companies across verticals, from Waymo in robotaxis to Aurora Innovation in trucking to Serve Robotics in sidewalk robotic delivery.
Perhaps Uber will hand over its vast data on how and where people use ride-hail and delivery, working in tandem with such partners. The company doesn’t develop its own AV technology, preferring instead the asset-light approach of partnering with others. That’s not for lack of trying; Uber’s history with AV development is troubled.
Uber launched its self-driving unit, Uber ATG, in 2015 through a partnership with Carnegie Mellon University’s National Robotics Center. A year later, the company acquired self-driving truck startup Otto, which was founded by one of Google’s star engineers, Anthony Levandowski. Uber later shuttered its trucking tech unit to focus on self-driving cars. But some damage had been done by bringing Levandowski on board. Waymo accused him of stealing trade secrets, which they alleged were then used by Uber. The case ended in a settlement in 2018, and Levandowski was later sentenced to 18 months in prison before he was pardoned by then-President Donald Trump.
Also in 2018, an Uber self-driving vehicle testing in autonomous mode in Arizona struck a pedestrian, who later died as a result of her injuries. The next year, Uber spun out Uber ATG after closing $1 billion in funding from Toyota, but then in 2020, sold it off to autonomous vehicle startup Aurora Innovation.
Today, Uber seems committed to being the bridge between rider and driver, whether that’s a human or robot driver. But that doesn’t mean Uber is comfortable moving slowly.
During the company’s third-quarter earnings report, Uber CEO Dara Khosrowshahi explained that the company is launching with Waymo in only two cities this year — Austin and Atlanta — due to the investment that goes into launch.
“What you want to get is the proper liquidity in a city anytime that you launch,” Khosrowshahi said in response to an analyst question. “You want to go into a city with the proper investment in your depots, in your infrastructure, in your mapping, etc., so that you can start getting a return on capital.”
It takes time to lay the groundwork for AV deployment in a new city, but it looks like Uber thinks that with Nvidia’s world models and cloud platform, it can get to scale quicker.
“By working with Nvidia, we are confident that we can help supercharge the timeline for safe and scalable autonomous driving solutions for the industry,” Khosrowshahi said in a statement.
Uber would not share any additional details with TechCrunch.
Rebecca Bellan covers transportation for TechCrunch. She’s interested in all things micromobility, EVs, AVs, smart cities, AI, sustainability and more. Previously, she covered social media for Forbes.com, and her work has appeared in Bloomberg CityLab, The Atlantic, The Daily Beast, Mother Jones, i-D (Vice) and more.
Rebecca studied journalism and history at Boston University. She has invested in Ethereum.
Read More
Rebecca Bellan