Uber to buy Seattle-based Mighty AI, which helps train computer vision models, to strengthen its self-driving car unit; Crunchbase: Mighty AI had raised $27.3M
Amir Efrati / The Information :
Context & Ripple Effects
Uber's self-driving push entered 2019 with fresh capital and a hiring problem it chose to solve by acquisition: months after the unit raised $1B from SoftBank Vision Fund, Toyota, and Denso at a $7.25B valuation, Uber is buying Mighty AI, a Seattle startup whose computer-vision training tools address the labeled-data bottleneck every autonomous program faces.
The purchase looks small next to that fundraise, but its afterlife is the interesting part: when Uber later offloaded the whole unit in the $4B sale of Advanced Technologies Group to Aurora, it did not abandon the data-labeling function — six years on, it acquired Segments.ai for lidar data labeling under Uber AI Solutions, suggesting Mighty AI seeded a capability Uber now treats as core infrastructure rather than an AV-unit asset.
First-order effects
- Mighty AI's team and tooling fold into Uber's Advanced Technologies Group, giving the freshly capitalized unit in-house control over the computer-vision training pipeline instead of relying on external vendors.
- Mighty AI's backers exit a company that had raised $27.3M per Crunchbase, adding another modest training-data tuck-in to the era's acquisition ledger.
Second-order effects
- Rival autonomous programs face the same labeled-data bottleneck, pushing them toward their own data-tooling acquisitions or vendor contracts and lifting valuations across the training-data segment.
- Once ATG went to Aurora, the labeling capability Uber retained became a service layer it could apply beyond self-driving — a path that culminated in the Segments.ai deal expanding Uber's lidar-labeling offering.
Third-order effects
- If the pattern holds, training-data tooling proves more durable than any single autonomy program: Uber kept building the labeling stack long after selling the cars-and-models business, splitting the AV value chain into model developers like Aurora and data-infrastructure owners like Uber.
- That stratification mirrors the broader tooling economy visible in Applied Intuition's rise as a simulation-software supplier with hundreds of millions in ARR — the picks-and-shovels layer compounding independently of which automaker wins.
The trend: Autonomous-vehicle development is stratifying into model builders and data-infrastructure owners, with training-data capabilities outliving the self-driving units they were built to serve.