Uber has a far-reaching goal that extends well past transporting riders: the firm ultimately intends to equip its drivers’ vehicles with sensors to gather real-world information for autonomous vehicle (AV) enterprises — and possibly other businesses training AI models in real-world contexts.

In a discussion at TechCrunch’s StrictlyVC event in San Francisco on Thursday night, Praveen Neppalli Naga, Uber’s chief technology officer, unveiled the initiative, characterizing it as a logical progression of a fledgling program the company introduced in late January known as AV Labs.

“That is the route we hope to pursue eventually,” Naga mentioned regarding outfitting human drivers’ automobiles. “But first, we must comprehend the sensor kits and their operations. There are certain regulations — we need to ensure every state has [clarity on] what sensors signify, and what sharing entails.”

At this time, AV Labs depends on a specialized, small fleet of sensor-laden vehicles that Uber manages independently, apart from its driver network. However, the vision is evidently much broader. Uber has millions of drivers worldwide, and if merely a small percentage of those automobiles could be converted into moving data-collection units, the extent of what Uber could provide the AV sector would surpass what any single AV company could create on its own.

The insight propelling the initiative, Naga stated, is that the critical limitation for AV progress is no longer the foundational technology. “The bottleneck is data,” he noted. “[Companies like Waymo] must gather data, capturing various scenarios. You might be able to specify: in San Francisco, ‘At this school intersection, I need data at this particular time of day to train my models.’ The challenge for all these companies is accessing that data, as they lack the capital to deploy vehicles and gather all this information.”

Becoming the data layer for the entire AV landscape is a rather clever strategy, especially considering Uber years ago set aside its aspirations to create self-driving cars (a decision that co-founder Travis Kalanick has openly regretted as a significant error). Indeed, many industry watchers have speculated that, absent its own self-driving vehicles, Uber could potentially become irrelevant as AVs proliferate worldwide.

The company presently collaborates with 25 AV enterprises — including Wayve, which operates in London — and is developing what Naga termed an “AV cloud”: a repository of labeled sensor data that partner firms can access and utilize to train their models. Partners, which Uber intends to invest in more actively, can also leverage the system to test their trained models in “shadow mode” against actual Uber journeys, simulating how an AV would have functioned without actually deploying one on the roads.

“Our objective is not to profit from this data,” Naga remarked. “We aspire to make it accessible to everyone.”

Considering the evident commercial potential of what Uber is developing, that stance may not be sustainable for long. The company has already invested equity in several AV players, and its capability to provide proprietary training data at scale could grant it considerable influence over a sector that currently relies on Uber’s ride marketplace to reach clients.