Foxconn will use Nvidia chips to build platforms for self-driving vehicles, Reuters reported Friday, deepening a partnership that already spans robotaxi development with Stellantis and Uber.
The report gives few specifics on timelines or customers, but it lands on fertile ground. Foxconn and Nvidia have been building toward automotive compute for three years, and Foxconn already assembles Nvidia’s AI servers across Texas, Wisconsin and California.
The automotive collaboration went public in October 2025, when Foxconn announced it would work with Nvidia, Stellantis and Uber to scale Level 4 robotaxis, vehicles that drive themselves without human help inside defined zones. The systems are built on Nvidia’s DRIVE AGX Hyperion 10, a reference platform that combines the in-vehicle computer with a safety-certified operating system and a sensor suite of cameras, radar and LiDAR. Foxconn’s job in that arrangement is high-performance computing and sensor integration for a global rollout.
The stack is demanding. Level 4 operation needs powerful computing, smart power management, reliable connectivity and thermal control, plus a sensor array working in concert to read the road. That is why Nvidia positions Hyperion as a reference design rather than a chip: it sets the baseline a vehicle maker has to meet.
“Autonomous mobility is a strategic priority within Foxconn’s EV program,” Foxconn chairman Young Liu said at the time. Nvidia CEO Jensen Huang framed it the other way: “Level 4 autonomy isn’t just a milestone for the auto industry. It’s a leap in AI capability.”
From AI factories to Kaohsiung streets
The relationship started with data centers. In 2023 the companies announced plans for “AI factories,” GPU infrastructure for training models, built on Nvidia’s Grace Hopper superchips. Foxconn, the world’s largest contract electronics manufacturer, has since become one of Nvidia’s key server builders while pushing its own electric vehicle program through the MIH open platform consortium.
“Foxconn and Nvidia share an excellent partnership. Our team is bringing the most advanced AI data center solutions to the United States,” Liu said at a GTC event in Washington, where the company also displayed Nvidia’s GB300 NVL72 system.
The car work moved into the real world this year. In June, Nvidia expanded its DRIVE Hyperion robotaxi program with more manufacturers and ride-hailing partners. In Taiwan, Foxconn is designing and building fully automated EVs starting in Kaohsiung, where the city government is upgrading infrastructure for large fleets. Foxconn has said it plans an official robotaxi service there in 2028.
| Date | Milestone |
|---|---|
| Oct 2023 | AI factories partnership announced, built on Grace Hopper superchips |
| Oct 2025 | Robotaxi collaboration with Nvidia, Stellantis and Uber on DRIVE AGX Hyperion 10 |
| June 2026 | DRIVE Hyperion expansion; Foxconn automated EV work begins in Kaohsiung |
| Sept 2026 | Reuters reports Foxconn to use Nvidia chips for self-driving platforms |
Why the division of labor works
Nvidia sells the same playbook it used in data centers: the chip, the software stack and the reference design, while partners handle manufacturing and deployment. Halos, its safety-certified operating system, sits alongside DRIVE AV software. Foxconn brings manufacturing scale and, through MIH, a base of carmakers that might otherwise build nothing themselves.
The competitive set is real. Qualcomm sells Snapdragon Ride, Mobileye supplies complete driver-assist stacks to most of the industry, and Tesla builds its own silicon and fleet data loop. Nvidia’s bet is that Level 4 autonomy needs data-center-class compute in the car, which favors its hardware.
Foxconn’s own position has been shifting for years. The company built its name assembling iPhones, and its electric vehicle push has been a strategic bet on finding a second act as the smartphone market matures. Contract manufacturing of automotive compute fits that logic: the company does not need to sell its own brand of car if it can build the nervous system for everyone else’s. Its EV output so far has been modest, with prototypes and low-volume programs, which is why the robotaxi work matters as a reference customer.
The economics remain unproven. Sensor suites are expensive, Level 4 operation is confined to geofenced areas and favorable weather, and the robotaxi services running today serve a handful of cities. Whether platform suppliers or fleet operators capture the value is an open argument in the industry.
Regulation shapes the timeline as much as engineering. Level 4 deployment requires approvals jurisdiction by jurisdiction, and the operating zones are drawn partly by regulators, not just by mapmakers. Taiwan’s willingness to prepare Kaohsiung’s infrastructure is a signal that at least one government wants the industry to move faster.
There is a supply chain angle too. Foxconn’s server business has grown fast enough that AI infrastructure now anchors its earnings calls, and automotive compute would diversify a revenue base that currently leans heavily on a handful of data center customers. Nvidia, for its part, has pushed hard into physical AI, robotics and autonomous systems as the next demand wave beyond cloud training, and Foxconn is one of the few partners with the manufacturing footprint to build that hardware at volume.
The labor question follows the factories. Robotaxi fleets threaten driving jobs the way server automation reshaped warehouse work, and the politics of that displacement are already visible in city council fights over pilot programs in the United States. Taiwan’s plan is framed as an industrial strategy first, which has so far kept that debate quieter there than in American cities where Waymo and its rivals operate.
What Friday’s report adds is supply-side depth. If Foxconn builds self-driving platforms as a contract manufacturer, automakers without in-house software teams get a shortcut to Level 4 hardware, and Nvidia’s automotive revenue gets a second engine alongside AI data centers. The next signal to watch is which carmakers commit.