Can AI data companies sustain margins beyond labor payout ratios?

Brendan Foody's Mercor reported more than $2 billion in gross annualized revenue in June 2026, double its pace from earlier in the year, according to The Information's reporting as relayed by RuntimeWire. The excerpt frames this against Amazon putting Mechanical Turk, "the old marketplace for internet-scale human microtasks," into maintenance mode: "AI labs and enterprises are no longer just buying cheap, interchangeable task completion. They are paying for credentialed human judgment that can train, evaluate and tune increasingly capable models." Mercor pays contractors with domain expertise in fields such as physics and finance to answer questions and build specialized training data, and the person cited by The Information said Mercor is profitable on a free cash flow basis, though Mercor has not released audited financials and the figure is a single month annualized.

The excerpt is careful to separate the headline number from durable revenue: Mercor pays contractors 60% to 70% of gross revenue, which "would put annualized net revenue at roughly $600 million to $800 million" if that ratio still held at the June run rate. That gap is why, in the article's account, "AI data companies can show unusually large revenue run rates while still being evaluated by investors on whether they can turn contractor supply into durable margin." The excerpt frames Mercor's strategic bet as whether it can move past being a broker for expert labor — where the payout ratio caps the business — into packaging that labor into products: APEX, an AI Productivity Index launched October 2025 to test models against expert knowledge work, and Mercor Enterprise AI, launched March 2026 to turn workplace context into "agent behavior specs, evaluations and quality guardrails." A rival, Handshake, is cited at roughly $1 billion gross and $450 million net (per Sacra), facing the same question of whether data-labeling marketplaces can become "a stickier data infrastructure layer" rather than staying commodity supply.

The excerpt's commodity/expert split runs in the opposite direction from a finding already in the library: Did ChatGPT's release reduce freelance writing work and pay? measures falling demand and earnings for freelancers doing easily-substituted, copy-and-paste-able writing work — the kind of generic task Mechanical Turk represented. Mercor's reported growth sits on the other side of that same substitution line: work credentialed enough that AI labs pay a premium for it rather than handing it to a model. The excerpt's account of Mercor's own origin also complicates Are recruiters and job seekers really adopting AI in hiring?: Mercor itself began as an AI recruiting platform automating "resume screening, matching, interviews and payroll management" before pivoting to the more lucrative expert-data business, so LinkedIn's figures on AI screening candidates describe a different transaction than Mercor's current one of paying candidates directly for their expertise. Against How close are frontier AI models to expert work quality?, which measures frontier models closing the gap with expert human output, Mercor's revenue growth reads as the opposite market signal: buyers still paying a premium specifically for human expert judgment, at least in the domains Mercor serves.

The excerpt does not establish durable margin: the $600-800 million net figure is the article's own conversion from the gross number using a payout ratio it says "would" apply, not a disclosed company figure, and the free-cash-flow claim rests on an unnamed person "with direct knowledge" rather than audited financials. Nor does it establish that expert-labor demand is general: Mercor's growth is concentrated among AI labs and Fortune 500 customers fine-tuning models, a narrow and well-funded buyer set, not evidence about demand for expert human input across the economy. The implication the excerpt supports is narrower than "human expertise becomes more valuable as AI gets stronger" — it is that a specific, concentrated set of AI developers currently pays well for credentialed human judgment as a training input, while the commodity end of human task labor that Mechanical Turk represented keeps fading.