On an ordinary evening in Wuhan, a central Chinese city of 13 million people, you can open an app, tap a button, and watch a white car pull up with an empty driver's seat. You get in, tap "start trip" on a screen, and the car merges into traffic, changes lanes, turns, and stops — all on its own. A few years ago this was a carefully supervised experiment. Today it is a paid commercial service that millions of people have used.
China has pushed into driverless taxis faster and more openly than almost anywhere outside the United States. The scale is real, the money is starting to follow — and so, increasingly visibly, are the problems.
The Leader of the Pack: Baidu's Apollo Go
The best-known Chinese robotaxi service is Apollo Go ("萝卜快跑"), run by internet and Artificial Intelligence (AI) company Baidu. Launched in Beijing in 2020, it expanded to Wuhan in 2022 and began fully driverless commercial operations there the same year — meaning no human safety driver sits inside the car at all.
The growth in rides has been steep. Baidu reported about 1.4 million orders in the first quarter of 2025, roughly 3.1 million in the third quarter, and 3.4 million in a single quarter by the end of the year — a year-on-year jump of more than 200 percent at peak. By February 2026, Apollo Go said it had crossed 20 million cumulative fully driverless orders across 26 cities worldwide, operating a fleet of more than 1,000 vehicles. In every mainland China city it serves — including Beijing, Shanghai, Shenzhen, Wuhan, Chengdu, and Chongqing — it now runs without an in-car safety driver.
Perhaps the most-watched claim came in August 2025, when Baidu co-founder and chief executive Robin Li told an earnings call that Apollo Go had reached per-vehicle break-even in Wuhan. If true and sustained, that is a threshold the industry has chased for years: proof that removing the driver can make a single car profitable, not just technically capable.
It Is Not Just Baidu
Apollo Go sits at the front of a competitive field. Pony.ai and WeRide form, with Baidu, the first tier of Chinese autonomous-driving companies, each with a different emphasis — Baidu on scale, Pony.ai on replication, WeRide on global expansion.
Pony.ai offered a revealing glimpse of the economics in mid-2026. Its robotaxi business revenue surged nearly 700 percent year on year in the second quarter, and in the cities of Guangzhou and Shenzhen its per-vehicle economics turned positive; one seventh-generation car in Shenzhen posted a peak daily net income of 394 yuan across an average of 25 orders, close to the productivity of a conventional ride-hailing car. But its Chief Financial Officer (CFO) also stated bluntly that the company would need to deploy 40,000 to 50,000 robotaxis before free cash flow turns positive — against a global fleet of under 2,000 vehicles at the time. The gap between "a few profitable cars" and "a profitable company" is roughly twenty-fold.
That gap explains the dominant business model emerging in China: rather than one company buying and operating every car itself, operators are partnering with carmakers, local governments, and ride-hailing platforms. It spreads the capital cost, routes real passenger demand into the system, and lets the autonomous-driving firm focus on the software.
Why Wuhan, and Why So Fast?
Wuhan is the clearest example of how China moves quickly. Local authorities embraced robotaxis as a symbol of technological progress and gave operators unusually large, continuous service zones rather than tiny isolated test blocks. The result is what many analysts call the largest autonomous-ride service area in the world — a place where driverless cars operate not in a fenced demo district but across real, mixed, high-speed urban traffic.
Three conditions make this possible across China. First, dense cities generate enormous ride demand, which is exactly what an early robotaxi needs to gather data and fill seats. Second, a mature electric-vehicle and sensor supply chain keeps hardware costs falling; the sixth-generation Apollo RT6 uses eight Light Detection and Ranging (LiDAR) sensors, a dozen cameras, and other sensors at a vehicle cost far below earlier generations. Third, local governments compete to host the technology, offering permits, infrastructure, and political support.
The Night 100 Cars Froze
The same scale that makes Wuhan impressive also magnifies failure — a lesson delivered on the evening of March 31, 2026.
Starting around 9 p.m., more than a hundred Apollo Go vehicles stalled almost simultaneously on roads across the city, including elevated ring roads with fast-moving traffic and little room to pull over. Videos circulated of rows of white driverless cars stopped in live lanes, hazard lights flashing, while ordinary drivers swerved around them. Some passengers were stuck for an hour or more; several reported emergency buttons and customer-service lines that were slow or unresponsive, and a few stepped out into traffic on their own. There were reports of minor rear-end collisions, but no serious injuries. Wuhan police attributed the disruption to a "system malfunction."
The incident was significant less for what happened than for what it revealed. These vehicles combine onboard sensors with constant cloud communication and high-definition maps, which means a central software or connectivity problem can affect many cars at once — a kind of systemic risk that a city full of independent human drivers does not have. It also exposed a tension at the core of the business model: robotaxi economics depend on cutting staff costs, yet a fleet-wide emergency demands a large, fast-responding human backup operation. In Wuhan that night, it was traffic police who walked onto the elevated roads to help passengers, car by car.
How It Compares With the United States
The global robotaxi race is largely a two-country contest. In the U.S., Alphabet's Waymo is widely regarded as the operational leader, with a fleet of several thousand vehicles running broad, all-hours commercial service in Phoenix, San Francisco, and Los Angeles, and more than 500,000 paid trips per week. Tesla promises a very different, camera-only approach at potentially far lower vehicle cost, while Amazon-backed Zoox has begun limited service.
China's first-tier operators are broadly comparable on core technology and, in Wuhan, have been willing to deploy over larger continuous zones. The two markets have also learned the same hard lesson: Waymo clusters have frozen during a San Francisco power outage, and driverless cars have repeatedly blocked traffic and emergency vehicles in U.S. cities. Independent analysts describe the autonomous-driving leaders in both countries as close, while noting the gap between performing well on average and handling rare, high-impact "edge cases."
Chinese companies are now pushing abroad. Apollo Go launched fully driverless commercial service in Dubai in March 2026, has pilots or plans in Abu Dhabi, London, Seoul, and Swiss cities, and has partnered with Uber and Lyft on overseas deployments. For these firms, foreign markets are not just extra cars — they may offer better unit economics and a longer profit runway.
The Open Questions
Three questions will decide whether China's driverless bet pays off. Can per-vehicle break-even in one city be reproduced across many, and can operators survive the years of losses needed to reach tens of thousands of cars? Can they build emergency and remote-support systems robust enough that a repeat of the Wuhan outage does not endanger passengers or choke a city? And will regulators — now scrutinizing safety more closely — keep allowing expansion, or slow it down?
What is no longer in doubt is that China has moved robotaxis out of the laboratory and into ordinary life. A visitor to Wuhan can already ride in a car with no driver, paying ordinary fares. Whether that becomes the future of urban transport, or a scaled-up lesson in the limits of automation, depends on whether the industry can make the cars as reliable in a crisis as they are on an ordinary evening.