What It Takes to Make Robotaxis an Everyday Service
In September, invited passengers in Zagreb began taking fully driverless Robotaxi test rides. For Pony.ai and our partners, the rides were a step toward offering a fully driverless service people could use for everyday trips. We welcomed guests, including members of the media, and received encouraging and constructive feedback to help us improve our technology and service.
At Web Summit 2026 in Lisbon, our co-founder and CFO Leo Wang used Zagreb to explain how a Robotaxi service develops once the first cars are on the road. An operator needs to build a safety record in local traffic, put enough vehicles in service to keep waits short and make the economics work for each vehicle. On a panel moderated by FOX Business correspondent Susan Li, Leo measured progress by how many vehicles Pony.ai could put on the road “collecting fares every day.”
Pony.ai’s third-quarter results, released on November 18, offer a closer look at what goes into that measure. Broader service coverage gave passengers more trips they could book, and lower insurance premiums reduced a recurring cost. In Europe, the joint deployment model is helping Pony.ai expand with local operators and mobility platforms.
Adapting the AI Driver to Zagreb
Narrow streets, frequent roundabouts and roads shared with trams were among the conditions Pony.ai had to account for in Zagreb. In a fireside chat with Bloomberg technology reporter Gian Volpicelli, Leo said adapting the system to the city took about five to six months. The first few dozen vehicles helped the team build a safety record on local roads. Commercial ride-hailing began in April, followed by fully driverless test rides in September. By the end of October, the Zagreb operation had covered more than 200,000 kilometers and completed several thousand passenger trips.
On the earnings call, our CTO Tiancheng Lou explained how PonyWorld, the proprietary world model we use to train our Virtual Driver, supported that work. The virtual training environment lets the driving system practice interactions with simulated road users before it encounters those situations on public roads.
In a new market, that means accounting for local traffic rules and driving habits. As the fleet encounters more situations on the road, those observations help improve the fidelity of the simulation and inform the next round of training. Road testing then shows how the system performs under the local conditions it has trained for.
Putting Enough Vehicles on the Road
A service area tells riders where they can go. The number of cars available nearby helps determine how long they wait for pickup. At Web Summit, Leo estimated that a typical European city might need at least 100 Robotaxis to offer pickups in five to eight minutes.
A shorter wait helps make a ride worth booking again. The experience inside the car matters too. In a CGTN interview, Leo said privacy stood out when Pony.ai looked at why passengers came back. He noted that some women particularly valued having the cabin to themselves on late-night trips.
Pony.ai had 2,768 Robotaxis worldwide at the end of September. On the earnings call, Leo linked higher fare revenue to specific additions to the service, including broader urban coverage in Guangzhou and rides to and from Bao’an International Airport and Shekou Cruise Port in Shenzhen.
By November, Pony.ai had connected its central Guangzhou and Nansha service areas, giving passengers a single operating area of more than 1,000 square kilometers. On the earnings call, our CEO James Peng said cross-district trips had helped increase the average trip distance and average fare per ride. Pony.ai’s third-quarter Robotaxi fare-charging revenue rose nearly 400% year over year. Total Robotaxi services revenue, which also benefited from higher vehicle deliveries under the joint deployment model, grew 165.4% y/y.
What It Costs to Run a Robotaxi
Leo pointed to Guangzhou and Shenzhen as examples of fleets that had reached breakeven at the vehicle level. At Web Summit, he highlighted the cost of autonomous driving hardware. Pony.ai reduced the bill-of-materials cost of its Gen-7 autonomous driving kit by 70% compared with Gen-6. A lower hardware cost reduces the amount an operator needs to recover over the life of each vehicle.

On the earnings call, Leo said both fleets had achieved positive unit economics and explained how Pony.ai calculates that measure. The company compares fare revenue after discounts and refunds with seven cost categories. Vehicle depreciation, including the battery and autonomous driving hardware, accounts for about half the cost. The rest covers charging, remote assistance and on-the-ground support staff, cleaning and maintenance, insurance, parking and network connectivity.
Insurance is one example of how operating experience can affect those costs. Per-vehicle premiums for the Gen-7 fleet fell 19% year over year at renewal. Leo attributed the more favorable terms to Pony.ai’s safety record.
Bringing the Service to More Cities with Local Partners
In the Bloomberg fireside chat, Leo drew a connection between the economics Pony.ai had demonstrated in Guangzhou and Shenzhen and the opportunity he sees in Europe. Achieving positive unit economics in competitive ride-hailing markets gave Pony.ai confidence in the commercial potential of its Robotaxis elsewhere. Zagreb gave the team and its partners another reason for confidence. They had seen the system adapt to the streets of a European city, including its historic districts. Each new city still has traffic patterns and local operating conditions of its own.
Leo described the joint deployment model as a way to make expansion “a win-win situation” by giving local companies a role in the Robotaxi business. In Zagreb, Pony.ai supplies the autonomous driving technology, Verne leads service operations and Uber connects riders to the service through its platform. The mix can vary by market. In Luxembourg, Pony.ai is testing with mobility platform Bolt and automaker Stellantis, which brings vehicle engineering and manufacturing expertise.
Local operating capability also matters as fleets grow. In a separate interview with Susan Li at Web Summit, Leo pointed to how quickly an operator can send help when a vehicle needs on-site assistance. Local authorities can look at how an operator handles those situations when assessing a wider deployment.
Pony.ai is advancing deployments in London and Manchester. In his France 24 interview in Lisbon, Leo said the company was also working with partners to explore how it could take part in potential Robotaxi pilots in France.
The partnerships may look different in each city. The commercial test is whether the partners can build a service that puts enough cars on the road, picks people up when they need a ride and gives them a reason to book again.