Surge AI reliant on 12 frontier labs

Surge looks less like a broad based software business and more like a critical supplier to a very small set of customers spending enormous sums to push model quality. The work is concentrated because frontier labs are the ones running the biggest RLHF, eval, and training data programs, and they increasingly need doctors, lawyers, coders, and researchers rather than generic clickworkers. That creates huge revenue fast, but it also means a handful of labs effectively set the pace of Surge’s growth.1 2 3 4 5

• The underlying workflow is expensive and specialized. Earlier RLHF could use low cost overseas contractors to rank outputs, but reasoning models pushed labs toward $50 to $100 per hour experts in medicine, law, science, and software, which sharply increased spend per project and favored firms that could source and manage elite talent at scale.2 4
• This customer pattern shows up across the category. Mercor and Micro1 are also described as selling mainly to labs like OpenAI, Anthropic, Meta, and other frontier model builders, with both companies warning that revenue can be concentrated among a limited number of big AI buyers. Surge’s own connected company set sits squarely in that same RLHF and expert data ecosystem.3 4 5
• Surge has moved beyond plain labeling into higher value research workflows. External reporting says it generated about $1.2B from labs including OpenAI, Google, Anthropic, and Meta, and then built a dedicated RL environments group. That matters because labs now buy not just labeled answers, but whole test worlds where models can practice coding, research, and agent tasks.1 5 6

The next step is a shift from annotation vendor to core training infrastructure. As frontier labs spend more on reasoning, agents, and domain specific evals, the winners in this market will be the firms that own expert supply, secure delivery, and reusable environments, making their revenue even larger per lab, and even more tied to a small circle of frontier buyers.3 4 5 6