The way Pony.ai announced its new domain controller—rather than with the typical fanfare of a Silicon Valley keynote, but rather as part of a larger Nvidia partnership story that already has Stellantis, Uber, and Lucid vying for headlines—is subtly telling.
It’s the kind of move that causes you to stop. After nearly ten years of trying to persuade investors, regulators, and doubtful passengers that it can be trusted with a steering wheel, the company is now wagering that the next chapter won’t be about cars at all. It has to do with the silicon within them.
| Pony.ai — Company Snapshot | Details |
|---|---|
| Founded | 2016 |
| Headquarters | Fremont, California & Guangzhou, China |
| CEO | James Peng |
| Core Business | Level 4 autonomous driving, robotaxi services |
| Key Partner | NVIDIA (collaboration since 2017) |
| New Hardware | Domain controller built on NVIDIA Drive Hyperion, powered by Drive AGX Thor |
| Peak Performance | 4,000 FP4 TFLOPS (dual-chip configuration) |
| 2026 Fleet Target | 3,000+ vehicles across 20+ cities |
| Autonomous Miles Logged | Over 34.1 million globally |
| Stock Listing | Nasdaq: PONY |
| Notable Markets | China (all four Tier-1 cities), Germany, UK, South Korea, Japan, Switzerland |
The Drive AGX Thor chip powers the new controller, which is based on Nvidia’s Drive Hyperion platform and can be configured with a single SoC or doubled up via NVLink for a total of 4,000 FP4 TFLOPS of compute. Most people will ignore that figure, but it’s important. In addition to raw horsepower, level 4 autonomy—the kind in which the vehicle actually drives itself within a predetermined area—requires redundancy, or the capacity to fail gracefully when something goes wrong. In 2025, shipments from Pony’s former controller, known as Fangzai, increased by over 500% year over year and found their way into cars in Germany, the UK, South Korea, Japan, and Switzerland. That is a quiet kind of achievement. It also suggests that the brains that drive everything from mining trucks to delivery vans may be the true business here, not robotaxis at all.
However, Pony’s robotaxi aspirations are anything but subtle. The company claims to have achieved unit-economics breakeven in two Chinese metropolitan markets, a milestone that Waymo and Cruise have found difficult to publicly attain. By the end of 2026, it hopes to have a fleet of more than 3,000 vehicles spread across more than 20 cities. The number is aggressive. Perhaps too combative. There have been many missed deadlines in the robotaxi sector, and Pony has experienced its fair share of difficulties. You get the impression that the company has realized that owning everything is costly and time-consuming when you hear James Peng explain the new “asset-light, AI-empowered” model, in which traditional taxi operators manage the fleet while Pony supplies the software.
Riders in Shenzhen got a taste of what autonomous driving actually feels like earlier this year. Depending on the vehicle’s generation and the passenger’s patience, it can be surprisingly smooth at times or a little aggressive at others. Along with Beijing, Shanghai, and Guangzhou, Shenzhen recently became the fourth Tier-1 Chinese city to issue Pony a citywide commercial driverless permit. That combination is unique to the company. Even though the general public still finds it hard to believe that a steering wheel could turn on its own, it’s a regulatory moat that’s difficult to ignore.
However, the story is reframed by the Nvidia angle. Nvidia is wiring up the entire industry, from Stellantis and Foxconn to Lucid and now a deeper push with Pony, without favoritism. Observing this, it seems as though Jensen Huang, who has never built a car, has quietly emerged as the most significant figure in autonomous mobility. Pony appears to be aware of this. The 3,000-vehicle target is audacious. It remains to be seen if it will arrive on schedule. However, the wager on Nvidia? It appears to be settled already.
