Figure AI launches Index with 16 million videos already uploaded

Signals Inbox·August 26, 2026·Humanoid Robotics

Figure came out of stealth with Index, a global robot-data network that says it has already collected 16 million videos. The striking part is the pace: 30 minutes of new footage per second, plus a stated commitment to spend more than $1 billion on data and compute over the next year.

Q1What was actually announced?

Figure founder Brett Adcock introduced Index directly, calling it the largest and most diverse robot dataset in the world. He reported 16 million video uploads, 264,000 downloads, $15 million paid to contributors, and 30 minutes of video arriving every second.

Q2How big is the signal?

At 30 minutes per second, Index receives about 720 hours of video every day, or roughly 30 days of human activity per hour. Figure also says it has paid contributors $15 million already and plans more than $1 billion of spending on data and compute over the coming 12 months. This is no longer a small research collection.

Q3Why does it matter now?

Humanoid models need examples of how people manipulate objects across many homes, jobs, camera angles, and failure cases. Factory robot logs alone are narrow. A contributor network can broaden the long tail of tasks and give Figure a training-data engine that improves while hardware deployments are still scaling.

Q4What is the catch?

Sixteen million uploaded videos are not automatically 16 million useful robot demonstrations. Quality, task diversity, labels, camera geometry, consent, and how much of the footage maps cleanly to robot actions all matter. The performance claims and dataset statistics are also company-reported, so downstream model gains are the real test.

Q5What should we watch next?

Watch whether Index creates measurable jumps in Figure Helix task success, how much data comes from paid human contributors versus deployed robots, and whether Figure opens any benchmarks. The strategic question is whether data network effects become a stronger moat than the robot hardware itself.