Uber turns its gig workforce into an AI labeling army

Your next Uber driver might be training AI models between rides.

Bloomberg reports that Uber is expanding its army of gig workers into the AI data-labeling business. Through a new division called Scaled Solutions, the ridehailing giant is deploying independent contractors to handle data labeling, testing, and localization for third-party tech companies.

The initiative grew out of an internal Uber team based in the US and India, which originally handled internal operational chores like testing app features and digitizing restaurant menus into Uber Eats listings. Now, Uber is packaging that workforce as a paid B2B service.

From street corners to token rankers

Frontier AI training still leans heavily on human labor for tedious tasks, from ranking chatbot responses to frame-by-frame pedestrian bounding in autonomous driving footage.

Uber has already lined up notable clients, including autonomous driving company Aurora, generative video startup Luma AI, and AR studio Niantic.

To supply the demand, Uber is actively recruiting workers across Canada, India, Poland, Nicaragua, and the US. The company is leaning on diverse cultural backgrounds to help models localize better. Compensation varies by task and is paid out monthly. One Indian engineer told Bloomberg they earn 200 rupees, or about $2.37, per set for reviewing and rating AI-generated solutions to complex coding problems.

This is hardly Uber's first brush with AI, though past attempts were far more capital-intensive. The company previously spent billions developing autonomous vehicles before shutting the unit down following a fatal pedestrian crash, and it acquired an AI research lab co-founded by Gary Marcus in 2016. This time around, Uber is skipping the moonshot research and leaning directly into its core competency: dispatching flexible global labor.

Why it matters

Uber entering data labeling makes total business sense, but it comes with clear execution risks.

For AI startups and labs, Uber presents an attractive alternative to incumbent data vendors like Scale AI. Uber already owns the plumbing for global contractor payouts, fraud detection, and workforce dispatching. That infrastructure gives it an immediate cost advantage for high-volume, repetitive tasks like bounding boxes or image localization.

However, the bottleneck in AI today has shifted from sheer volume to high-quality reasoning data. Evaluating complex code or domain-specific logic requires vetted specialists, not casual gig workers picking up a task between delivery runs. Paying roughly $2.37 for complex code grading suggests Uber may struggle with expert retention and data quality compared to specialized human-data shops.

If Uber sticks to mid-tier localization and physical-world sensor annotation where its driver base excels, it could carve out a massive business. But if it tries to commoditize high-end intellectual labor through low-rate gig contracts, the quality hits will show up quickly in downstream models.