The Uber ATG Sale and What It Signaled for the Industry

Building a self-driving car turned out to be more expensive than running a rideshare app, and Uber learned that lesson the hard way. In December 2020, the company sold its Advanced Technologies Group, the internal unit responsible for its self-driving development, to Aurora Innovation. Uber took an equity stake in Aurora as part of the deal rather than walking away with cash alone, which softened the exit without changing what it signaled. Selling a costly business unit while keeping a minority stake in the buyer is a common way to step back from a bet without fully giving up on its future value.

The choice of buyer wasn't incidental. Aurora had already absorbed engineering talent and technology from Toyota's self-driving effort and had built a partnership with Volvo aimed specifically at trucking, a market ATG had also been developing toward alongside its passenger-vehicle work. Combining ATG's testing miles, sensor development, and engineering staff with Aurora's existing trucking relationships produced a company with a broader base of resources than either had running independently, which is a large part of why Aurora, rather than some other acquirer, ended up on the other side of the transaction. A merger of that size also meant Aurora inherited a stronger case for pursuing both trucking and passenger-vehicle autonomy under one roof, rather than choosing between the two the way many smaller, less capitalized competitors eventually had to.

The sale did not happen in isolation. Two years earlier, an Uber ATG test vehicle struck and killed a pedestrian in Tempe, Arizona, a collision that had already forced Uber to scale back its on-road testing well before the Aurora deal was announced. The unit spent much of the intervening period rebuilding trust with regulators and the public rather than accelerating toward deployment. Trust doesn't rebuild on a deadline.

That is not the trajectory a capital-intensive research program can sustain indefinitely without a parent company losing patience.

Uber's own approach to autonomous vehicles didn't end with the ATG sale; it just changed shape. Rather than building a self-driving stack in-house, the company shifted toward integrating third-party AV developers directly into its ride-hailing app, letting a rider request a partner company's autonomous vehicle through the same interface used to hail a human-driven car. That model shifts the technology risk onto specialist developers and leaves Uber doing what a ride-hailing platform is built to do: matching demand with available vehicles, autonomous or not, rather than engineering the vehicles itself.

Uber was not a struggling company casting off a failing side project when it finally sold. It was one of the better-capitalized players in the rideshare business, and it still concluded that running a full in-house self-driving stack alongside a global ride-hailing operation cost more than it was worth to keep separate. Ride-hailing itself already runs on thin margins and heavy competition; layering a capital-intensive robotics research program with no revenue of its own on top of that business is a different kind of bet than the core app ever was. Consolidation among specialist developers looked like the more sustainable path after that, for reasons that overlap with why a wave of autonomous-vehicle companies chose to go public via SPAC merger instead of the slower traditional route. A ride-hailing marketplace succeeds by getting incrementally better at matching supply and demand; a self-driving research program succeeds, if it ever does, in one discontinuous leap from not-quite-ready to deployable at scale. Running both inside a single corporate structure means funding two entirely different kinds of risk out of the same balance sheet, and Uber ultimately decided that wasn't a combination worth preserving. Aurora, notably, ended up on the other side of both transactions.