The Robotaxi Fleet on Saadiyat Just Changed Passports

"Autonomous driving can't guarantee 100 percent safety. But it could be ten times safer than human drivers within the decade." Tony Han, WeRide's founder and chief executive, told Fortune that in October 2025. Nine months later Han's company runs the largest commercial driverless fleet in Dubai. But its Chinese peers just got retired from the fleet operating one causeway east of Sheikh Zayed Road.

On 4 August 2026, Autonomous A2Z, a South Korean autonomous-driving firm founded in 2018 by four former Hyundai Motor autonomous-driving engineers, signed an 11 billion won ($7.7 million) contract to supply 19 vehicles to TXAI, the ride-hailing operation of Abu Dhabi's Space42. Eight of the 19 are the ROii, a Level 4 roboshuttle with no driver's seat. Five are built on the Kia PV5 commercial-EV skateboard. Five more sit on the Kia Carnival platform. One is a MAN-based autonomous bus. The Korean vehicles will run on Saadiyat and Yas Islands. According to both the Korea Herald report and the Space42 joint-venture announcement that preceded it, they fully replace the Chinese-made autonomous cars that had been operating there.

That last sentence carries more of the story than the price tag does.

What TXAI is, in operational terms

TXAI is not a pilot. It has been in continuous operation since 2021, initially under G42's Bayanat subsidiary and now under Space42, the merged AI-and-space entity majority-owned by G42. The service has completed roughly 600,000 km of autonomous driving and 20,000 passenger trips across Saadiyat, Yas, Al Maryah and Al Reem Islands, plus routes to and from Abu Dhabi Airport.

For an operator running that kind of service, the vehicles are the disposable component. What you build once and refuse to hand to anyone else is the mapping stack, the fleet-dispatch software, the passenger app, the V2X (vehicle-to-everything) plumbing, and the regulatory relationship with the Abu Dhabi mobility authority. Space42 has been assembling all of these in parallel with e& and, for high-definition maps, with Japan's Dynamic Map Platform.

The vehicles, in that architecture, are commodity endpoints. Which is why the swap can happen cleanly. You are not rebuilding TXAI; you are re-plugging its cars.

Why the pivot is not only about the cars

Set this alongside a piece of news CnEVPost carried out of Dubai in early March 2026: Chinese robotaxi firms including WeRide and Pony halted commercial operations in Dubai when Iranian missile strikes reached Middle Eastern airspace and vehicle operations became national-security-adjacent. The pause turned out to be brief. What it demonstrated was less brief: a foreign-controlled autonomous fleet is, at bottom, an asset whose remote command-and-control leg lives somewhere else, and can go dark on somebody else's political calendar.

An Emirati sovereign fund operating a passenger service on its own islands has to price that risk. Dubai's Roads and Transport Authority priced it and stayed with Chinese partners: 1,200 robotaxis committed with WeRide and Uber by 2027, a separate Baidu Apollo Go pilot underway. Abu Dhabi priced it and diversified. Both are defensible readings. They are not the same reading.

The Korean stack (a2z, Kia platforms, MAN chassis) comes with a different political calculus. Seoul is not currently negotiating tariff and export-control terms with either the UAE or its neighbours. The a2z engineers left Hyundai to build a Level 4 stack that now runs the largest driverless fleet in South Korea, 62 vehicles with roughly 740,000 km of cumulative urban driving. That is a smaller record than WeRide's. It is a much smaller record than Waymo's. It is not a smaller record than most Chinese peers' actual published safety reporting, which is thinner than the marketing material would lead you to expect.

What is happening across the region, in one table

Four cities. Five deployments. Three distinct AV stacks: Chinese, Korean, and the emerging European-Chinese hybrid on the Mercedes S-Class. No American robotaxi presence at commercial scale in the Gulf. Waymo does not run here. Cruise no longer exists. Tesla's Robotaxi has, as of mid-March 2026, a single unsupervised car in Austin and fifteen NHTSA-filed crash reports since June 2025: not a Gulf-ready fleet.

The mileage question

Compare the operating numbers. Waymo has reported more than 170 million rider-only miles as of 19 March 2026, with the Waymo Driver involved in 92 percent fewer serious-injury crashes than human drivers in the same conditions. An independent IIHS analysis published in July 2026 confirmed the broad direction of that number. TXAI's public record is 600,000 km. WeRide has more than 200 vehicles operating in the Middle East and does not publish per-vehicle crash rates for Gulf jurisdictions.

The counterthesis you will hear from the Chinese-stack partisans is that WeRide's Middle East operational footprint (200-plus vehicles now, 1,200 committed by 2027, presence across three Gulf capitals) is the actual industrial fact, and Abu Dhabi's 19-vehicle Korean swap is a headline in a footnote. It is not a stupid argument. It also is not the argument the operator making the swap decided to take seriously.

The Gulf's operating environment adds specifically demanding conditions the American mileage does not test for. Ambient temperatures well above 45 degrees Celsius. Sandstorms. Camel-crossing edge cases. Prayer-time traffic surges. Multilingual signage in scripts most Western training sets under-sample. Pony AI has argued that its own hardware is calibrated for these conditions; that claim is not yet backed by a public multi-year Gulf crash record from any operator. Anyone selling a "Level 4 anywhere" story has not spent August in Abu Dhabi.

Han's Fortune quote (10x safer than humans within the decade) is a decade projection, not a 2026 delivery. Space42 knows this. TXAI still runs on relatively short, geofenced island loops precisely because that is what the current honest performance envelope allows.

A note from a Riyadh summer

I spent stretches of the summer of 2017 in Riyadh working with an Aramco engineering team, and I drove the King Fahd Road at odd hours more times than I would like to remember. The heat coming off the tarmac at 2 a.m. is a physical fact, not a metaphor. The sensor rigs that survive it are not the ones spec'd for a Palo Alto climate. The a2z engineers who trained on Chungju summers and Seoul winters are, on balance, better calibrated for that reality than a Silicon Valley team on its first regional posting. Environmental match matters as much as raw stack sophistication, and Gulf operators know this by 2 a.m. tarmac experience, not slide deck.

Somewhere in the fleet-selection meetings at Space42, that match was one of the arguments made. The political one was the other. Both hold. Neither is the whole story.

What the swap tells us

The 4 August contract, valued at a rounding-error $7.7 million against Gulf infrastructure spending, is not a big commercial event in itself. What it is, is a legible signal about how a specific class of operator — an Emirati sovereign-fund-owned mobility company running a service under its own national aviation-and-mobility regulator — is now thinking about supplier concentration in autonomous vehicles. When the cars carrying your passengers can be paused by a decision made in a foreign capital, and when the sensors have to survive your climate rather than the one they were designed in, the vehicle-supplier choice migrates from the procurement line into the operational-risk register. Abu Dhabi wrote that entry down. Dubai did not. Riyadh has not yet.

The last time the Gulf changed the composition of its car fleet at this level of intent was in the late 1970s, when regional oil wealth first made it possible to buy Mercedes and BMW at scale rather than the American V8s that had dominated the 1960s. That transition took a decade and a half and reshaped both regional taste and regional supplier relationships. The autonomous transition is running faster because the vehicles are commodity by design. But the same instinct is at work: a Gulf state deciding who gets to sell it its next fleet, and refusing to let that decision default to whoever showed up first. It is the same instinct. The vehicles have simply learned to drive themselves.

Tarry Singh is the founder and CEO of Real AI, an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan, an Energy AI startup, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.