Scale Reaches $760M ARR via LLMs
Scale at $760M ARR
Scale’s 2023 jump shows that the company did not simply survive the collapse of self driving labeling demand, it found a much larger and faster moving market in LLM post training. The core workflow changed from labeling camera and lidar data for cars to having humans rank model outputs, write ideal answers, and test safety edge cases for labs building foundation models and AI products. That swap reopened hypergrowth even as the old AV business slowed.
-
Before the LLM boom, Scale was heavily tied to autonomous driving and similar world of atoms workloads. It sold usage based labeling for images, video, and 3D maps from customers like Cruise, and benefited from the huge volume of sensor data those programs produced. When AV R&D pulled back in 2022 and 2023, that demand shrank.
-
LLMs created a different kind of labeling market. Instead of drawing boxes around pedestrians, contractors reviewed chatbot answers, ranked competing responses, flagged unsafe outputs, and later supplied higher skill domain expertise for reasoning tasks. That made human labor central again, but now around RLHF and evaluation for labs like OpenAI, Anthropic, and Cohere and the apps built on top of them.
-
The broader market quickly filled with new specialists. Mercor built a marketplace for doctors, lawyers, and PhDs, while Invisible used a managed workforce model to win RLHF work from Microsoft and Cohere. Scale still had the biggest revenue base, moving from $215M ARR in 2022 to $760M in 2023 and then to $1.5B in 2024, but competition shifted from commodity crowdwork to expert labor and workflow orchestration.
The next phase is less about basic annotation volume and more about owning the full human feedback stack for frontier models. Scale is best positioned when customers need managed workflows, secure handling of sensitive data, and large pools of specialized reviewers. As AI labs push further into reasoning, safety, and enterprise deployment, the winners will be the platforms that turn scarce expert judgment into a repeatable product.