Ten cars. That is the whole robotaxi fleet carrying paying passengers in Europe on this partnership right now, a handful of Arcfox SUVs doing flat 1.99 euro rides around Zagreb with a safety operator sat behind the wheel. On 14 August, Pony.ai and Uber said they plan to put more than 2,000 robotaxis across Europe, growing from that Zagreb service into four more cities. Which cities? Not disclosed. Starting when? Also not disclosed. We have read enough of these to recognise the shape, where the number is the announcement and the number happens to be two hundred times the fleet that exists.
The short answer
Pony.ai and Uber said on 14 August 2026 that they plan to deploy more than 2,000 robotaxis across Europe, expanding from the Zagreb service to four more cities announced in phases. Pony.ai brings the Level 4 stack, Uber brings the app, a local partner owns the fleet. No city names, no dates, no per city vehicle counts and no financial terms. The Zagreb service it builds on went live on 8 April with 10 cars, a flat 1.99 euro fare and a safety operator in every seat.
What was actually signed
Read the release carefully and the committed part is thinner than the headline. Pony.ai and Uber expanded a partnership that has existed since May 2025. Pony.ai supplies its Level 4 autonomous driving technology and the operational know-how it built in China. Uber supplies the demand side, so booking, payment and customer service through the app. A local company owns the cars and does the unglamorous work of cleaning, charging and maintaining them.
That structure is the genuinely interesting bit, and it is easy to skip past. Nobody in this deal is proposing to own a European fleet from Guangzhou or San Francisco. The capital and the depot sit with a local operator, which in Zagreb means Verne, spun out of the hypercar maker Rimac.
Sarfraz Maredia, who runs autonomous mobility and delivery at Uber, put it as moving “from individual launches to repeatable commercial scale”. Pony.ai founder James Peng talked about bringing services “to more European cities”. Neither sentence contains a city or a date, and I do not think that is an accident.
The service it scales from is ten cars
Here is where the story gets concrete, because Zagreb is unusually well documented.
Europe’s first commercial robotaxi service went live on 8 April 2026. Ten Arcfox Alpha T5 electric SUVs, built by BAIC, running Pony.ai’s seventh generation autonomous driving hardware. A flat fare of 1.99 euro no matter how far you go. Roughly 90 square kilometres of coverage across the city centre and some southern districts, running 07:00 to 21:00. Around 300 active riders, with a waiting list reported in the thousands.
And a trained safety operator sitting behind the wheel of every single one.
Image: Verne, from its announcement of Europe’s first commercial robotaxi service. Company promotional photography, which is worth remembering when you look at how empty that road is.
Verne has said it wants to drop the safety operator in Zagreb before the end of 2026, subject to regulatory approval. Worth being precise about what that sentence is. It is a company goal with a regulatory condition attached, and the condition has not been met. Until it is, the European end of this partnership carries a person in the driver’s seat on every trip, which is the same position Wayve and Uber are in on the London side, where TfL licensed up to 15 cars on 5 August with a human driver legally required.
Why the missing city list matters more than the missing dates
Vague timing in an announcement is normal. Vague geography is not, and in Europe it is the whole ballgame.
Autonomous vehicle permission on this continent is granted nation by nation, sometimes city by city, under regimes that mostly were not written with driverless cars in mind. Germany has a framework. France has one. The UK built a new one and switched on its passenger service permits in May 2026. They do not recognise each other, and none of them recognises a Chinese driverless permit. So a plan to enter four European cities is really four separate regulatory projects, on four timelines, with four sets of local politics about taxi licensing and jobs.
Which is exactly why the cities are not named yet. I would guess the list is not settled because the permits are not settled, and naming a city you then fail to enter is a worse headline than naming none.
The part that is genuinely strong
I have been sceptical for six paragraphs, so let me put the other side properly, because it is stronger than the sceptical reflex suggests.
Pony.ai is not a demo company. It runs paid, fully driverless robotaxi services in China’s four tier-one cities, and it says it has reached city-wide breakeven unit economics in multiple markets. That claim is the company’s own and nobody outside has audited it, so hold it loosely. But operating driverless at fare-paying scale is a thing very few firms on earth have done at all, and it is a completely different achievement from a pilot with a safety driver.
The joint-deployment model is also smarter than a straight fleet play. Splitting the stack from the platform from the local operator means the hard local problems, which are depots, permits, cleaning, charging and municipal relations, land on a company that already lives there. That is a structure that can actually repeat across borders, which is more than most robotaxi expansion plans can say.
So the plan is credible in a way that, say, a pure announcement from a startup with no revenue would not be. The gap between credible and scheduled is just wider than the headline number implies.
What to watch instead of the 2,000
If you want to know whether this is happening, ignore the fleet target. Three things will tell you first, and none of them will arrive as a press release with a big number in it.
Watch whether Verne gets its Zagreb driverless approval before the end of the year, because that is the first European regulator to actually say yes to an empty driver’s seat on this stack. Watch which city gets named next, since the country tells you which regulatory regime the companies think they can clear fastest. Then watch whether Verne’s own purpose-built vehicle, the two-seater with no steering wheel it plans to build near Zagreb, reaches production, because the Arcfox fleet was always described as a bridge.
The thing about robotaxi announcements is that they are cheap and the permits are expensive. Cities that took the enthusiasm at face value are the ones now arguing with their robotaxi operators in court about where the cars park overnight. Two thousand cars is a real ambition from a company that has earned some benefit of the doubt. It is also, today, ten cars and a waiting list.
Sources: the expansion, the 2,000 vehicle figure, the four additional cities, the Middle East mention, the division of labour and both executive quotes are in the joint Pony.ai and Uber press release (14 August 2026), which also carries the China driverless and breakeven unit economics claims. Coverage and the absence of named cities or dates from TechCrunch, heise online and CnEVPost. The Zagreb launch date, the Arcfox Alpha T5 vehicles, the safety operator requirement and Verne’s driverless target are reported by The Next Web, with the 10 vehicle fleet size, the flat 1.99 euro fare, the roughly 90 square kilometre zone and the 07:00 to 21:00 hours in this Zagreb service report. Pony.ai’s own launch announcement covers the April start. Image: Verne.
Frequently asked questions
What did Pony.ai and Uber actually announce on 14 August 2026?
An expansion of an existing partnership, with plans to deploy more than 2,000 Pony.ai robotaxis across Europe. The service in Zagreb is the starting point and it grows to four additional European cities, which the companies said would be announced in phases. The Middle East is mentioned as a further deployment target. Pony.ai provides the Level 4 autonomous driving technology and operational expertise, Uber provides customer access through its platform, and a local operator owns and runs the fleet.
Which four cities are getting the robotaxis?
Nobody has said. That is the single biggest gap in the announcement. The companies stated only that details would be released in phases, and no country list exists either. Zagreb is the fifth and only named city, and it is the one already running. Verne, the Croatian operator, has separately said it has permitting discussions underway with a number of cities across the EU, UK and Middle East, but that is a different company talking about its own pipeline rather than a confirmed list for this deal.
Are these cars driverless?
Not in Europe. Every Zagreb vehicle carries a trained safety operator behind the wheel, and the operator Verne has said it aims to move to fully driverless running before the end of 2026 subject to regulatory approval. That is a stated goal, not a granted permission. Pony.ai does run paid, fully driverless services in China tier-one cities, which is a genuine track record, but a Chinese driverless permit carries no weight with a European regulator.
How big is the Zagreb service today?
Small and precisely documented, which is what makes it useful. It launched on 8 April 2026 with 10 Arcfox Alpha T5 electric SUVs running Pony.ai seventh generation autonomous driving hardware. The fare is a flat 1.99 euro regardless of distance, the operating zone is roughly 90 square kilometres covering the city centre and selected southern districts, and service runs from 07:00 to 21:00. Verne is the fleet owner and operator, and it was spun out of the Croatian hypercar maker Rimac.
How does this compare with Uber other European robotaxi deal?
Uber says it has partnered with more than 30 autonomous vehicle companies, so it places a lot of these bets. The instructive contrast is London, where Transport for London licensed Wayve vehicles on 5 August 2026 for up to 15 cars on a trial basis, with a human driver legally required on every trip. London has a small number attached to an actual permission. This announcement has a large number attached to an intention. Both are real, and they are not the same kind of thing.