GM Quit Cruise’s Taxis. Uber Offers Fleet Support

After eight years and $10 billion, General Motors walked away from Cruise’s taxi service while Waymo kept cars on Austin streets without human monitors. Both had worked on removing the driver. Only one still wanted to run the service left behind.

For commercial autonomous vehicles, the scarce capability is no longer just a driving system. It is the ability to coordinate vehicle supply, fleet capital, remote assistance, maintenance, insurance, passenger demand and permission to operate on particular streets—and to decide which of those obligations to own. Cruise’s exit and Waymo’s continued expansion are different answers to that operating question, not opposing verdicts on autonomous driving.

Cruise built a service that its parent no longer wanted to run

GM’s original design joined the carmaker to the fleet operator. In 2017, the coverage record described Cruise’s employee ride-hailing service running in San Francisco more than 16 hours a day, with plans to add over 100 cars. Cruise could develop the vehicle and driving system, then learn from operating both in a service it controlled. The experiment answered a question car sales could not: what happens when the manufacturer, rather than the buyer, remains responsible for every trip?

Cruise could run an employee service in 2017 but missed its planned 2019 commercial San Francisco launch, citing performance and safety concerns. The company later acquired Voyage, whose low-speed fleets served retirement communities—a way to work within a more constrained driving environment. In October 2023, GM, Cruise and Honda were planning a Japan ride service using the purpose-built Cruise Origin. Each plan assumed that driving capability could mature alongside a repeatable fleet business.

After the October 2023 pedestrian crash, GM slowed Cruise’s expansion and cut its planned spending. Cruise restarted testing in Dallas, Houston and Phoenix in 2024, but GM suspended development of the Origin. Then GM ended the robotaxi experiment after eight years and $10 billion, saying Cruise would leave the taxi business while its technical teams joined GM’s work on autonomy and driver-safety systems for future cars. GM kept a use for the engineering; it declined to keep underwriting the dedicated operator.

The decision reached beyond GM. Honda planned to dissolve its self-driving partnership after GM’s announcement, despite having invested $852 million in Cruise. A manufacturer can redirect engineers toward products it sells. Its fleet partner cannot preserve a joint service plan when the lead operator stops building the fleet.

A city can take back the part of autonomy that matters most

Cruise’s retreat cannot be reduced to expensive vehicles and depots. California suspended its driverless deployment after a Cruise vehicle struck and dragged a pedestrian in October 2023. In November 2024, Cruise admitted submitting a false report intended to influence the crash investigation and agreed to pay a $500,000 fine. The driving failure was grave; the reporting failure also damaged the institution asking the public to trust its account of what had happened.

Safety governance has a direct operational address: the permission attached to a fleet in a particular jurisdiction. NHTSA closed a preliminary investigation into hard-braking and immobilization problems affecting 1,194 Cruise vehicles after the company made updates. That closure matters. Cruise could remediate identified problems; GM’s subsequent withdrawal was a strategic decision, not proof that a robotaxi service was technically impossible.

Other operators face the same local boundary. Wayve obtained London private-hire licenses for supervised robotaxis, and Uber’s London service using Wayve technology began with human safety drivers. A license for supervised operation is not permission to remove those drivers everywhere. Under deployment accountability, an operator must make its incident response, oversight and service boundaries legible to the authority that can restrict them. The related transition in autonomous delivery makes the distinction plain: authorization for a pilot does not by itself supply the governance required for a larger service.

Uber can sell fleet functions without building the driver

Before GM withdrew, Uber and Cruise agreed to put Cruise vehicles in front of Uber riders beginning in 2025. Cruise would have supplied the cars and driving operation; Uber would have supplied passenger demand and dispatch. GM’s exit overtook that proposed arrangement, but the division of work outlived it. Uber and Lyft had already left their own driverless-taxi development efforts and turned to autonomous-vehicle operators, including Waymo and May Mobility, as partners.

Uber has since made its role more explicit. Uber Autonomous Solutions offers insurance, roadside assistance, mission-control tools and fleet financing. Those services once sat inside the imagined all-in-one robotaxi company. Separating them gives an autonomy supplier a route to riders without requiring it to build every dispatch and support function. It also creates interfaces that someone must govern: the party sending a car to a passenger is not necessarily the party maintaining it or deciding how to recover it.

Waymo uses the Honk app to summon towing and roadside help. It has also trained about 25,000 people to assist a fleet of roughly 3,000 robotaxis; that is a trained support pool, not a count of people working every shift. The driverless trip still encounters disabled vehicles, roadside recovery and exceptions that require human work. Autonomy changes where people enter the process rather than removing them from it.

Passenger miles reveal more than a launch photograph

Tesla’s Austin debut showed how small an initial operating domain can be. Its June 2025 launch involved about 10 cars, geofenced away from challenging intersections and backed by teleoperators. In December, an Austin comparison put Waymo at roughly 200 robotaxis without human monitors and Tesla at an estimated 30 with safety drivers. By June 2026, Bloomberg counted 59 Tesla robotaxis across three Texas cities. Those observations describe different dates and operating conditions, not a fixed limit on Tesla’s technology.

A city name alone says little about service depth. Operators still have to specify the streets, the conditions, the presence of a safety driver and the assistance available when a vehicle cannot complete a trip unaided. California’s decline in autonomous test mileage in 2024 likewise cannot be read simply as Waymo retreating: Waymo had become a robotaxi operator there, rather than only a tester.

Utilization supplies a different measure. A study of 13.8 million Waymo trips from August 2023 through December 2025 found that the share of miles driven with a passenger aboard rose from 36% to 56%. More of the fleet’s movement was doing the work riders paid for. The study does not establish profitability, but it measures an improvement that a driving demonstration cannot: how effectively an operator turns vehicle time into occupied service.

The fleet still needs a balance sheet

Waymo offers the strongest case against assuming that every operator must split itself into separate companies. Its Austin fleet and improving utilization show what a well-resourced, deeply involved operator can accomplish. That involvement also requires financing: sources reported that Waymo enlarged its inaugural debt raise from more than $3 billion to $5 billion while expanding its fleet. The reported raise was a financing effort, not evidence that the service already earns enough to fund expansion from its rides.

Other companies have chosen narrower positions. Pony AI and Toyota began commercial production of an autonomous bZ4X, with Pony AI targeting more than 3,000 robotaxis by the end of 2026; that fleet size remains a target. Volvo and Aurora unveiled a self-driving truck for freight hauling, pairing an autonomy developer with a vehicle manufacturer and a different commercial use. Chinese self-driving companies facing losses and thinner funding have also moved toward less advanced smart-driving products that they could sell sooner. None of those choices settles which model wins. Each identifies a different obligation its backers are willing to finance.

GM could keep Cruise’s engineers and leave its taxi operation. Waymo could keep operating and seek more capital. Uber could supply dispatch and the roadside network. At the curb, those decisions meet in one object: a car with no driver, waiting on a licensed street for a tow truck that still has to arrive.