A three-year-old startup you’ve probably never heard of is quietly processing more than $2 million in daily payouts to contractors who teach AI models how to think. Mercor, a San Francisco-based AI recruiting and data platform, hit $614 million in gross revenue during the first half of 2026, with an annualized run rate now touching $2 billion.
More than 90% of that revenue comes from just three customers: OpenAI, Anthropic, and Google DeepMind.
The numbers behind the hype
Mercor’s growth trajectory is genuinely staggering. The company reached a $1 billion annualized revenue run rate, then scaled to $2 billion just four months later. That $2 billion figure represents a 70% increase compared to its entire revenue for 2025.
The platform connects roughly 30,000 domain experts and contractors with AI research labs. These workers handle data labeling, model evaluation, and reinforcement learning from human feedback, commonly known as RLHF.
Contractors earn an average of $105 per hour. The company has raised $486 million in funding and currently carries a $10 billion valuation. Mercor was founded around 2023 by Brendan Foody, Adarsh Hiremath, and Surya Midha.
Why crypto cares about human feedback pipelines
Mercor isn’t a crypto company. It doesn’t issue tokens, run a blockchain, or have a Discord full of people asking “wen moon.” But its business model sits squarely at the intersection of a problem decentralized networks have been trying to solve for years: how do you coordinate distributed human labor at scale without centralized intermediaries extracting massive margins?
Several crypto-native protocols are building marketplaces for data labeling, model evaluation, and compute resources that aim to replace centralized middlemen with token-incentivized contributor networks. The thesis is straightforward: if you can coordinate 30,000 contractors through a startup, you can theoretically coordinate them through a protocol, with lower fees, transparent pricing, and permissionless access.
What this means for investors
Mercor’s financials serve as a useful benchmark for anyone evaluating decentralized AI projects. A $2 billion annualized run rate from essentially connecting humans to AI labs tells you that the market for human-in-the-loop AI services is not speculative. It’s real, it’s scaling fast, and the major labs are willing to spend aggressively to secure access.
The $105 average hourly rate for contractors also provides a pricing anchor. Decentralized labor protocols that can deliver comparable quality at lower rates, by cutting out the platform’s margin, could attract both supply-side workers and demand-side AI labs looking to optimize costs. At daily payouts exceeding $2 million, even modest fee compression translates to meaningful savings for buyers.
The projects worth watching are those building verifiable quality assurance into their protocols, because that’s the real moat. Not the coordination layer, not the payment rails, but the ability to guarantee that the human feedback flowing into a frontier AI model actually meets the standard those labs require. Mercor’s $10 billion valuation is ultimately a bet on that quality guarantee.