Nvidia revealed this week it will provide computer chips and self-driving software to companies running robotic taxi fleets, with services set to start operating in 2027.
The move grows Nvidia’s car business, though it still makes up a tiny slice of what the company earns. In the quarter that ended in October, chips for cars and robots brought in $592 million, about 1% of everything Nvidia sold during that time.
The chip company said last October it was teaming up with Uber to build technology for taxis that drive themselves.
Test drive shows promise with some challenges
Then in December, Nvidia rolled out software that lets cars handle city driving on their own. Mercedes-Benz will put this system in vehicles coming out in late 2026, giving them the ability to manage tricky urban roads like those in San Francisco.
Nvidia also gives carmakers its computer processors and testing software. The company sells its Drive AGX Thor car computer for around $3,500 each. Nvidia says this helps carmakers spend less money on creating self-driving features and get them to buyers faster.
The company works with each manufacturer to adjust the technology for different car models, including things like how fast the vehicle speeds up. “Some say, ‘Hey, I need your help on training and optimizing my software on your chip, but I’ll take care of simulation myself,'” said Ali Kani, who runs Nvidia’s car platform division.
The business of taxis without drivers has picked up speed in the last year. Waymo, which Alphabet owns, runs the biggest operation right now.
Nvidia set up test drives in December to show off what its technology can do. Reporters and business experts spent an hour riding around San Francisco in a 2026 Mercedes-Benz CLA with the automated driving system running.
A Mercedes-Benz safety worker sat in the driver’s seat during the test but let the car drive itself for 90% of the trip. San Francisco has tough driving conditions, with sharp hills, lots of traffic lights, and delivery trucks that block traffic lanes. The car handled most of it without problems.
The driver did need to step in once. The car ran into a messy situation with two buses and a Waymo robotic car all trying to get through a four-lane street that had parked cars and trucks unloading on both sides.
Nvidia calls the system “Level 2 Plus Plus” technology, putting it in the same category as Tesla’s Full Self-Driving feature. The company said the system will one day offer “park-to-park” driving, which means handling everything from leaving one parking space to pulling into another. That ability won’t come with the first Mercedes-Benz CLA models though.
“Any parking situation that you feel is intimidating, that car will solve for you,” said Ola Källenius, who leads Mercedes-Benz Group, speaking at an Nvidia event on Monday.
The Mercedes-Benz car started selling in Europe last year and arrives in America this year, Kani said. Early models came with lane keeping and driver help features. The cars got lane-changing abilities added through software downloads and will receive hands-free highway driving, city navigation, and park-to-park capabilities throughout this year.
Dual systems aim to keep passengers safe
Nvidia runs two separate computer systems in Drive-equipped cars to keep things safe. The main system uses what Nvidia calls a vision-language model, which relies on computers interpreting what cameras see to figure out where to drive.
A backup safety system works with set rules, like always stopping at stop signs, and takes over when the main computer isn’t sure what to do. The company thinks new advances in computer technology that creates content, run by its graphics chips, will make self-driving better.
Nvidia set 2028 as when it wants point-to-point self-driving ready for regular car buyers. Down the road, the company wants cars that respond to voice commands from riders. “With transformers and generative AI, we can do much more,” Wu said.
Get $50 free to trade crypto when you sign up to Bybit now
