Google and OpenAI Cut Scale Ties

The Meta deal turned Scale from a neutral arms dealer into infrastructure partly owned by one of the labs it serves. That matters because Scale’s biggest customers do not buy generic labor, they send sensitive model prompts, failure cases, evaluation rubrics, and post training workflows. Once Meta bought 49% and brought Alexandr Wang into its AI effort, rivals had a strong reason to move that work elsewhere, even if Scale kept operating independently.1 2 3 4

• Google was reported to be Scale’s largest customer and planned to cut ties right after the June 2025 deal. Reuters also reported that OpenAI had already been pulling back, and later reporting said OpenAI was phasing out the relationship while shifting specialized work to other providers.1 2
• This is especially damaging because Scale’s LLM era growth was concentrated in frontier labs. Scale’s ARR was estimated at $760M in 2023 and $1.5B in 2024 as model builders bought RLHF, evals, and expert labeling, so losing a few large labs can remove a meaningful chunk of demand very quickly.4 5
• Competitors were positioned to absorb the spend. Prolific said customers stepped back from Scale over independence concerns and that work flowed back into the market, benefiting Prolific, Handshake, and Surge. Invisible also shows the broader buyer pattern, frontier labs spread work across vendors when they need trusted human feedback at scale.3 6

Going forward, the winners in human data for AI are likely to be the providers that look most independent and least conflicted. Scale can still grow with Meta, defense, and enterprise work, but the market is moving toward a clearer split between captive data suppliers tied to one lab and neutral vendors that every lab feels safe using.3 4 6