Mercor Revenue Linked to Training Cycles

Mercor is less a steady software business than a labor marketplace sitting on top of a few giant AI training budgets. When a frontier lab starts a new reasoning push, Mercor can ramp thousands of doctors, lawyers, or PhDs onto hourly projects and gross revenue jumps fast. When that lab pauses a training run, changes model priorities, or reprices the work, a large share of billed volume can disappear just as quickly.

• The work itself is tied to specific model building phases. Labs use Mercor to source expert contractors for RLHF and reasoning data, often paying $50 to $100 per hour for specialized workers, so demand rises and falls with active training roadmaps rather than recurring seat based software usage.
  1 sacra 2 sacra
• Reported revenue in this category behaves more like payments volume than pure software revenue. In adjacent data labeling models, 60% to 70% of gross revenue can pass through to contractors, which means a single 5,000 person project can create a huge top line spike without the durability or margin profile of SaaS.
  3 sacra 4 sacra
• This concentration risk is common across the sector. Invisible scaled on large contracts with Microsoft, Cohere, AI21, Mistral, and Perplexity, while Handshake reignited growth through Handshake AI. The winners are the marketplaces that can keep shifting labor supply toward the newest high value model evaluation jobs.
  3 sacra 4 sacra

The next phase is a race to make this revenue base less project by project and more embedded in customer workflows. Mercor will likely push deeper into recurring evaluation, benchmarking, and full time recruiting, where demand persists between major training cycles and dependence on a handful of capex driven labs declines.