Uber Launches 'AV Labs': Going After the 'One Thing' Over 20 Partners Want

The news that Uber has launched 'AV Labs' is not just about a new department. It is a declaration by the ride-hailing platform itself that the winning strategy in the robotaxi race is shifting from 'vehicles' to 'data'.

1. What is 'AV Labs'—Why Uber is Competing Without Building Its Own 'Robotaxis'

According to TechCrunch, Uber has established a new organization called Uber AV Labs, noting that while they have over 20 robotaxi/autonomous driving partners, what they all have in common is a need for 'data'.

The key point here is that Uber is not 'developing autonomous vehicles again,' but rather arming its own vehicles with sensors and sending them out into the city to collect driving data for its partners. The article touches on the history of them offloading their autonomous driving division after a past accident (in 2018), while clearly stating their stance of 'not going back'.

Uber Newsroom also explains that the goal of AV Labs is not 'winner-takes-all,' but to create a data flywheel that addresses the biggest bottleneck to autonomous driving progress: the real-world long tail (rare and difficult situations).

2. Why 'Data' Now—Autonomous Driving Shifts from 'Rules' to 'Learning'

TechCrunch observes that as autonomous driving shifts from rule-based to reinforcement learning, the value of real-world driving data is skyrocketing.
In other words, you can no longer win with just 'pristine test tracks' or 'well-crafted simulations.' What is needed are 'real-world logs' full of the exceptions that occur on actual roads.

What Uber Newsroom calls 'rare, messy, real-world scenarios' is the very fuel for learning-based autonomous driving.

3. What exactly will they do? Sensor vehicles + a design that 'doesn't hand over raw data'

TechCrunch describes AV Labs as being in the stage of 'literally installing' lidar, radar, cameras, and more, starting with a single Hyundai Ioniq 5.
It is also mentioned that the policy is not to hand over raw data to partners as-is, but to provide a 'semantic understanding' layer that is ready for use.

Even more interesting is 'shadow mode'. The idea is that while Uber drives the collection vehicles, they simultaneously run the partner's autonomous driving software in the background, flagging and delivering the moments where there is a discrepancy between 'human judgment' and 'AI judgment'. The article notes that this mechanism is effective not only for discovering software weaknesses but also for learning how to 'drive more like a human'.

There is a symbolic statement here. CTO (Praveen Neppalli Naga) stated, 'We want to democratize data first. The value of the data is greater than the money we could make from it,' suggesting that they may not charge for it for the time being.

4. What this means for the industrial structure—Uber is aiming for the 'robotaxi OS layer'

Once AV Labs is fully operational, Uber will become strong in a different area than 'robotaxi companies that own cars.' In short, it is a data supply hub for a multi-company autonomous driving ecosystem. TechCrunch cites Waymo, Waabi, and Lucid Motors as examples (though no contracts have been signed yet).

This move is a continuation of the Lucid x Nuro x Uber robotaxi plan announced at CES 2026 (expected to launch in the Bay Area within the year). Uber is not 'building everything themselves,' but is trying to take hegemony on the network side by dividing labor between vehicles, autonomous driving stacks, and the ride-hailing experience.

Furthermore, in 2025, NVIDIA mentioned 'expanding Uber's L4 ecosystem' and a 'data factory' concept. AV Labs also appears to be a piece that brings that flow down to 'on-the-ground driving data'.

5. Expectations and Issues—'Scale Barrier,' 'Privacy,' and 'Who Benefits'

TechCrunch points out that while Uber's approach is similar to Tesla's learning approach, they do not have the 'millions of vehicles' scale that Tesla has.
On the other hand, Uber emphasizes its mobility, saying they can 'choose from 600 cities and deploy in a targeted manner to the cities partners want'.

From an investment and business perspective, the issues are also clear.

• Privacy/Data Governance: The design of providing processed data rather than raw data is rational, but 'how much it is anonymized and at what level it is shared' will determine future trust (details are limited at this time).

• Economies of Scale vs. Targeted Collection: Since they cannot beat the constant collection of millions of vehicles, can they increase value density through shadow mode and city selection?

• Profit Sharing: While Uber talks about 'democratization first,' the company that ultimately stands at the 'critical junction of data supply' will hold the bargaining power. For partners, this is a shortcut to data acquisition, but it also creates a risk of dependency.

The essence of AV Labs is a declaration that Uber is not positioning itself as a 'robotaxi operator,' but as the 'data infrastructure' required to run robotaxis.
What to watch next: (1) How many vehicles will actually be deployed in how many cities, (2) the terms of the data provision contracts (conditions for moving from free to paid), and (3) which partner produces 'results' first. When these pieces fall into place, AV Labs will cease to be just news and become the 'OS layer of the industry.'

Recommended Articles

Next Big Wave (Growth Stocks, Seeds of Ideas, Deep Dives into Trends)