Uber has spent a year adding things that look like nice extras: hotel bookings through Expedia, boat rentals in Europe, a grocery assistant that builds your cart from a voice command. The company's chief product officer calls travel "the third leg of the stool" after rides and food. That framing is accurate but incomplete. The more important shift is happening underneath.
Uber launched a division called AV Labs at the start of 2026. It deploys cars fitted with cameras, radar, and other sensors onto real roads to collect driving data for self-driving car companies. The target is 500 such vehicles globally this year, capable of gathering roughly 2 million miles of recorded driving per month. That data is what self-driving software needs to handle unusual situations: a car stopped at an odd angle, a cyclist behaving unexpectedly, a construction zone that was not on any map.
The timing matters. Self-driving companies have hit a well-known wall. The common situations are solved. What remains are the rare ones, and you cannot encounter rare situations without driving an enormous number of miles across dozens of cities. Uber has those miles already, and now it has the sensors to turn them into sellable product.
There is also a subtler angle here. Uber has partnerships with over 18 self-driving car companies globally, including Waymo, and holds equity stakes in several of them. Waymo robotaxis dispatched through the Uber app in Austin were rated 4.9 stars on average by riders, and Waymo's vehicles were reportedly more productive than 99 percent of human Uber drivers in those cities. That is an impressive number, and it also explains why Uber would want a data layer it owns independently. If you are both a partner and a potential competitor to the same companies, controlling the data supply is a natural hedge.
On the driver side, Uber has started selling something different: the attention of its workforce during idle time. Drivers who are not on a trip can take short digital tasks inside the Uber app: recording voice samples, uploading images, transcribing audio. These feed directly into AI training pipelines for Uber's enterprise clients. Uber acquired a Belgian startup called Segments.ai to sharpen its sensor data labeling capabilities and formally named this business Uber AI Solutions. The AI data labeling market is projected to reach over five billion dollars by 2030, and Uber is positioning its driver network as a ready-made, geographically spread workforce for it.
The membership side of the business reinforces all of this. Uber One has reached 50 million paying members, up from 30 million at the end of 2024. Members now drive more than half of all bookings across rides and delivery, and they spend three times more than non-members. A business where half your revenue comes from subscribers is a different kind of business than one that relies on whoever opens the app and picks the cheapest option that day.
What this means practically: Uber is building three revenue streams that did not exist three years ago. One is data for self-driving companies. One is AI training tasks sold to tech companies using its driver workforce. One is a subscription layer that converts occasional users into habitual ones. None of these require Uber to build self-driving cars itself.
For businesses that rely on moving people or goods, the pattern here is worth watching. Uber is not trying to be everything; it is trying to be the platform that everything else runs on top of. That is a different kind of ambition, and a more durable one.