Meta 49% stake triggers Scale exodus

The Meta deal turned Scale from a neutral arms dealer into a supplier tied to one of the labs it was supposed to serve. That matters because Scale’s biggest customers were not buying generic labor, they were buying a trusted external partner to run sensitive post training workflows, route model outputs to raters, and manage proprietary evaluation data. Once Meta owned 49% and pulled Alexandr Wang into its AI effort, rivals had a strong reason to move work elsewhere.1 2

• Scale’s surge came from frontier labs spending heavily on RLHF and evaluation. By 2023 it had jumped to $760M ARR, then to an estimated $2B of revenue in 2025, with Google alone spending about $150M in 2024 and planning about $200M in 2025 before cutting ties after the Meta transaction.
• The work that moved was portable because competitors already offered similar human feedback pipes. Invisible built RLHF programs for Microsoft, Cohere, AI21, Mistral, and Perplexity, while Prolific says labs use external vendors for second opinions, specialist participant pools, and independent validation even when they have internal annotators.2 4
• The customer flight also exposed a deeper shift in the market. Human data vendors are no longer judged only on scale of contractor supply. They are judged on neutrality, speed, and quality in specialized tasks like red teaming, safety review, cultural nuance, and expert evaluation, which favored vendors like Prolific after the reshuffle.4 5

From here, the market moves toward more specialized and more independent human data providers. Scale can still grow in government, robotics, and Meta adjacent workloads, but the broader category is being rebuilt around vendors that can plug into frontier labs without ownership conflicts and handle narrower, higher trust evaluation work.4 5