Customers Internalizing Valuable Tasks

Diving deeper into

XDOF

Company Report
Large customers could internalize valuable tasks and leave suppliers competing for less differentiated collection work.
Analyzed 11 sources

The strategic risk is that XDOF is selling into customers who can learn the workflow and then pull the highest value layers in house. XDOF already spans demonstrations, collection equipment, and evaluation services, and its founders helped create GELLO, a teleoperation system now commonly used in the field. That makes it useful as an early partner, but it also teaches capable labs how to run capture, grading, and improvement loops themselves, especially once collection hardware and open datasets become easier to copy.1 2 3 4 5

  • XDOF has openly published ABC-130K as a 130K plus episode teleoperation dataset and says GELLO is commonly deployed. When a supplier shares both data and tooling, the scarcest asset shifts away from raw collection toward task design, model debugging, and evaluation, which are exactly the parts sophisticated customers are most likely to absorb internally.2 3 5
  • The collection stack is increasingly reproducible. Public robot data systems like DROID document the exact rig, cameras, gripper, headset, and control laptop. Camera vendors like Stereolabs sell synchronization, multi camera fusion, and rig building tools off the shelf. That lowers the barrier for a well funded robotics team to stand up its own pipeline.6 7 8 9
  • If insourcing happens, outside vendors get pushed toward lower margin overflow work. Scale moved into robotics data collection with dedicated robots and human demonstrations, and Mecka sells task specific collection, evaluation, and full data programs at a reported $100M revenue run rate by June 2026. That points to a crowded market where undifferentiated collection can become capacity driven.4 10 11

The next phase of this market favors suppliers that own the hard to copy layer, not just labor and hardware. The winners are likely to be the firms that turn collection into a closed loop, where they help decide what data to gather next, score whether it improved the policy, and transfer that workflow across new robots faster than customers can rebuild it themselves.1 2 4 10