James Peng, cofounder and CEO of Chinese robotaxi startup Pony AI, was reviewing customer data with his team, and he was facing a puzzle. Every day, one of his users would book a robotaxi at the same time: right after lunch.
“After a while, we called him and asked why he always took a ride at that time,” Peng recalled to Fortune.
The user’s answer? “The environment is great. It’s clean. I use it as my napping place!”
For Peng, the regular napper is a sign of how quickly riders are adapting their behavior as self-driving cars start to take over China’s—and the world’s—streets.
“Without a driver in the loop, we have to find creative ways to do a lot of things,” Peng says. If a passenger leaves a door open—a common problem for robotaxis—the car might chirp at a passerby in a “cute voice,” in Peng’s words, asking them to close it. If that doesn’t work, Pony AI will reach out to China’s army of delivery gig workers, asking them to close the door and maybe “clean up [the car’s interior] a little bit,” says Peng.
And Chinese robotaxis are now in service well beyond China. Pony AI says it’s now delivering 26 rides per car per day, or somewhere north of 25,000 daily in total, with operations in the United Arab Emirates, Qatar, and Singapore, and is planning to expand into Europe; its counterparts are also expanding around the world. (By comparison, Waymo is present in only two non-U.S. cities: London and Tokyo.)
A number of structural advantages have made China a fertile test bed for self-driving cars.
Peng sees a world with fewer human-driven cars as “inevitable,” citing safety and convenience. “People love to drive; they don’t love driving all the time,” he says. But he doesn’t see that trend as a danger sign for the labor force, noting, “AI will change what we consider ‘work.’ ”
Chinese consumers are also more willing to consider self-driving cars. Around 85% of Chinese drivers reported being comfortable with robotaxis without human supervision, compared with 39% of U.S. drivers, according to a 2023 survey from PwC. Far fewer Chinese—just 35%—drive a car, meaning they may be more willing to hire a robotaxi.
Finally, there’s government support for automated driving, which Beijing sees as a strategic industry. Local governments offer pilot zones, subsidies, and speedy permits for high-level autonomous driving, while national regulators have issued guidelines to move robotaxis from testing to commercial services in dozens of cities.
HSBC predicted last year that robotaxis could soon capture around 6% of China’s total taxi and ride-hailing market, generating $40 billion a year in fare revenue; meanwhile, UBS estimated that the size of the robotaxi market in China could reach $183 billion a year—if self-driving cars completely replace human-driven taxis.
Peng calls Baidu a “magnet” that attracted talented engineers and researchers from across the industry, several of whom went on to found their own ventures.
Baidu was an early supporter of AI among China’s tech sector, including making a concerted pitch to hire Geoffrey Hinton, an early AI pioneer.
Pony AI’s cofounder and chief technology officer, Tiancheng Lou, is also a Baidu alumnus, as is Tony Han, CEO of competing robotaxi firm WeRide, who was chief scientist of its autonomous driving unit. Even Dario Amodei, Anthropic’s cofounder, spent a year at Baidu.
In late 2016, Peng made the jump to startup founder, establishing Pony AI in Silicon Valley. It started testing self-driving cars in California in 2017, then in China in 2018.
By late 2025, Pony AI claimed its robotaxis were operating at breakeven in Guangzhou; by March, it said robotaxis in nearby Shenzhen were breakeven as well. Pony AI reported record usage of its robotaxis over the Chinese New Year holiday, hitting an average of 26 orders a day per vehicle. Each robotaxi generated an average of 338 Chinese yuan ($48.91) a day.
Pony AI generated $60.8 million in revenue in the first nine months of 2025, a 54% year-on-year jump. But investing in a new technology is expensive. The company spent $156.9 million on research and development between January and September 2025, contributing to a $152.2 million net loss over the same period.
Pony AI’s shares have performed poorly since the Hong Kong IPO, down by around 30% from the original offer price. WeRide, which had its own secondary Hong Kong listing at the same time as Pony AI, has seen its shares drop by about as much.
Pony AI is now venturing into international markets: specifically, Dubai in the United Arab Emirates; Doha in Qatar; Seoul, Hong Kong, Singapore, and Luxembourg. (Peng also mentioned the startup will soon expand to another European market, though didn’t specify which one.)
85%
35%
Pony AI’s long-term plan isn’t to own the self-driving cars, but rather provide the technology that keeps them running. The startup positions itself as the “virtual driver”—providing the AI, software, and platform—while partners fund and operate the physical fleet.
“That’s where we create the greatest economic and societal value,” Peng explains. “Our most important motivation is to scale as fast as possible, and put as many ‘drivers’ into the market as possible. Everything else could be done by somebody else.”
Partners, meanwhile, can earn a return from vehicle ownership. “If you can earn, say, 5% return by owning vehicles instead of 3% from keeping that capital in the bank, it’s a good business to be in.”
One market that’s not in the cards for Pony AI? The U.S., which takes a dim view of Chinese cars owing to concerns over data security. The outgoing Biden administration barred the sale of Chinese “connected vehicles” from 2027 onward.
“We’re unlikely to run large-scale commercial operations in the U.S. anytime soon,” Peng says. “But I think R&D activities and exchanges of ideas are still permitted.” The company retains a research team in Silicon Valley.
Pony AI’s robotaxis—including models like the GAC Aion V, a compact crossover SUV—are spacious and comfortable, similar to other luxury EVs getting churned out by China’s many factories.
A voice welcomes passengers as they enter the car; a screen mounted in the rear cabin allows them to start the journey and monitor what’s happening around the vehicle on a real-time map. A cute robot-like avatar—at the time wearing a horse costume to celebrate the incoming Year of the Horse—shared updates on the car’s journey.
The trip feels oddly smooth, as the robotaxi cleanly shifts lanes and slows down to avoid hitting other vehicles and bicycles on the road. It’s a quieter journey than what a human-driven taxi normally feels like.
Peng understands that difference. “The human driver does more than just driving: They clean, they charge the car, they have conversations with the passenger, or even, in some cases, comfort passengers,” he says.
Investors are jumpy, too. In early 2026, U.S. trucking shares collapsed after a little-known karaoke-turned-AI firm announced its product could help increase freight volumes by 300% without adding staff.
Peng takes a measured view. “AI will not destroy the workforce; it will change what we consider ‘work,’” he says. “A lot of the fear is overblown.”
Yet he’s certain that a world with fewer human-driven cars is “inevitable,” citing efficiency, safety, and convenience. “People love to drive; they don’t love driving all the time,” he says.
What about the roads themselves? Urban infrastructure, after all, is still pretty dumb—forcing robotaxis to be designed around transit systems that are decades old. Will that limit how far robotaxis can go?
Peng is realistic on that front. Perhaps, he acknowledges, the roads will get smarter in 20 or 30 years, and make autonomous driving safer and more efficient. But he’s not going to wait around for the roads to be rebuilt.
“If we want autonomous vehicles to really be part of everyday life, they have to cope with the roads we have now. That’s the beauty of AI: We can train our AI ‘drivers’ to be smart enough to be on the existing infrastructure.”
This article appears in the April/May 2026: Asia issue of Fortune with the headline “The world’s consumers are ready for robotaxis. James Peng of Pony AI wants to make sure they’re riding in his”
