Scale's AV Reliance and Pivot

This decline showed how concentrated Scale’s early engine really was, a usage based labeling business tied to a narrow set of self driving programs that could vanish when funding dried up. Scale had built strong economics around huge sensor datasets from AV teams, but by 2022 and 2023 major buyers either shut programs, sold units, or slowed deployment, which hit the steady flow of images, video, and 3D mapping work that had powered the company’s first phase of growth.

• Scale’s original advantage was operational, not just technical. It sold an API and managed workflow for labeling camera, LiDAR, and map data from autonomous fleets, and that specialization helped it reach better gross margins than broader outsourcing peers like Appen and TELUS.
  2 sacra
• The downturn was driven by real customer pullback. Lyft sold its Level 5 unit to Toyota in April 2021. Uber sold ATG to Aurora in December 2020. Argo AI was wound down in October 2022. Cruise paused U.S. operations in October 2023, and GM said in early 2024 it would slow Cruise investment.
  1 sacra 5 lyft 6 uber 7 gm
• What mattered strategically was that Scale’s labeling system was reusable. The same workflow for routing tasks to human annotators could be repointed from drawing boxes around pedestrians and lane lines to ranking model outputs and writing preference data for LLM training, which is why revenue reaccelerated so sharply in 2023.
  1 sacra 2 sacra 3 sacra

Going forward, human data spend is moving away from one giant AV category and toward a wider mix of model training, evaluation, safety, and specialized expert tasks. That favors platforms like Scale that can shift the same labor orchestration engine across use cases, but the winning work will increasingly be higher judgment tasks rather than pure high volume sensor labeling.