Pony.ai

One driving system, three businesses

Pony.ai is an autonomous driving company founded in Fremont, California, in late 2016^[2]. Its structure rests on a core system it calls the virtual driver, and on top of that system it runs three business units: Robotaxi, Robotruck and personally owned vehicles^[1]. The idea behind that arrangement is that the same underlying driving software should be able to serve a ride-hailing fleet, a long-haul truck and a car sold to an individual owner, rather than each use case being built from scratch.

Founders

The company was founded by James Peng, who is chief executive, and Tiancheng Lou, who is chief technology officer^[2]. Peng spent 11 years at Baidu and Google before founding Pony.ai, and holds a PhD from Stanford and a bachelor's degree from Tsinghua^[2]. Lou was a two-time Google Code Jam champion and was the youngest T10 engineer in Baidu's autonomous driving division^[2]. Between them, the founding pair combines academic machine-learning training, large-company engineering experience and a public competitive-programming record.

Robotaxi as the public face

Pony.ai says it was the first to launch and offer a public-facing robotaxi service in both California and China^[2]. A public-facing service means rides are open to ordinary passengers rather than only to employees or invited testers, which is a different kind of undertaking from closed-road testing: the vehicle has to handle routes, pickups and passengers without a safety driver making the decisions.

Commercialisation moves in 2026

The company has described a series of steps toward commercial robotaxi operation. In February 2026 it advanced robotaxi commercialisation with Toyota on mass-produced bZ4X vehicles^[2]. In March 2026 it launched a joint robotaxi fleet with Chenqi Mobility, with an initial delivery of more than 100 Gen-7 vehicles^[2]. In the same month it integrated robotaxi access into Tencent's mobility service, placing its rides inside a consumer app that people already use to book transport^[2].

On the technical side, in April 2026 it launched PonyWorld 2.0, described as a self-improving physical AI engine for autonomous driving^[2], and announced a new autonomous driving compute platform built on NVIDIA DRIVE Hyperion^[2]. The compute platform is the hardware layer that runs the driving software; a "self-improving" engine refers to training systems that generate and reuse driving scenarios so the software keeps getting better with less manual labelling.

Outside China, in June 2026 the company expanded its Singapore service with ComfortDelGro through access in a consumer app, and announced a testing programme in Luxembourg with Bolt and Stellantis^[2].

Where it sits in the field

Pony.ai belongs to the group of companies that began as robotaxi developers and then widened out. Two things follow from its design choices. First, because the virtual driver is shared, engineering work done for one business unit can in principle carry over to the others, which matters in a field where development is expensive and slow^[1]. Second, the sequence of partnerships — with an automaker on mass-produced vehicles, with a mobility app, with a fleet partner, and with ride-hailing and manufacturing partners abroad — reflects the fact that running a driverless service requires vehicles, a booking channel and local operating permission, none of which a software company holds on its own. The company's own account of its history places it among the earliest to put a public robotaxi service on the road in both California and China^[2], and its later work has been about turning that position into vehicles, cities and partners.