US Experts Flood AI Training Platforms with Hourly Rates Up to $400
According to Bloomberg, practicing radiologists, lawyers, and engineers in the U.S. are joining the gig economy for AI model training, with some positions offering hourly rates as high as $400, breaking the long-held perception that data labeling is linked to low-paying crowdsourcing jobs.
Radiologist Anu Brixey earns an hourly income from annotating medical images equivalent to several hundred dollars that a typical data worker makes in a week; recruiters are seeking software engineers at $120 per hour, condensed matter experts at $170, and civil rights lawyers at $400 to provide professional judgment for the next generation of models.
Amazon Mechanical Turk, operational for 21 years, officially closed on September 30, 2026, with task submissions ending and the remaining approval window extended to October 30; meanwhile, the Mercor platform, focusing on white-collar experts, has an average hourly rate of $110 and is discussing a new funding round at a $20 billion valuation.
LinkedIn lists data annotators as the fourth fastest-growing job role in the U.S. this year; platforms like Mercor require experts to write real case studies, provide reference answers, score model outputs, and identify hallucinations or errors, shifting tasks from simple bounding boxes to complex reasoning validation.
General data workers like Brook Hansen can only earn a few hundred dollars a week after losing their jobs, while higher-level positions are highly competitive and hours are unstable; platforms screen candidates through AI interviews, with some medical positions requiring current practice, licensure, and independent practice experience.
Capital is shifting from low-cost general labeling to scarce professional judgment: AI labs are buying doctors' reading logic and lawyers' review standards by the hour, with intermediaries like Mercor charging matching and management fees; experts with domain knowledge are temporarily earning high hourly rates, but as public datasets become exhausted, the demand for implicit knowledge continues to drive up expert premiums.
Some platforms pay contractors between $1.5 million and $4 million daily, with expert networks reaching tens of thousands; as basic labeling is automated, professional validation becomes a key input for models in vertical fields like healthcare and law.
Source: Public Information
ABAB AI Insight
Mercor was founded by three young entrepreneurs, initially focusing on AI recruitment matching, later shifting to supplying experts such as doctors, lawyers, and PhDs for RLHF and evaluation data to labs like OpenAI. In 2025, it completed a $350 million Series C round at a $10 billion valuation, with discussions of a valuation rising to $20 billion thereafter, similar to Scale AI's early reliance on manual labeling before being backed by large firms.
Funding is primarily directed towards expert hourly rates and platform matching systems, with labs purchasing "golden answers" and scoring rules by the hour rather than buying public datasets outright; the motivation is that after the saturation of publicly available internet data, high-value implicit knowledge such as medical imaging reasoning and contract review cannot be obtained through scraping and must be rented at professional rates.
This contrasts with the early Mechanical Turk model of breaking down simple tasks into low-cost HITs and resembles consulting firms selling senior consultants' time to clients and then refining methodologies; the current phase represents a transition from general crowdsourcing to a high-premium professional validation market, with expert supply becoming a bottleneck.
Essentially, this is a form of technological substitution: as models lower execution layer costs, value concentrates on experts who can provide verifiable reasoning traces, attracting existing practitioners to sell their judgments at high short-term hourly rates, while in the long term, personal moats are transformed into replicable model capabilities, making it harder for new entrants to compete against stronger automation.
ABAB News · Cognitive Laws
- Short-term high hourly rates = long-term zero marginal cost competitors
- After public data is exhausted, implicit knowledge is priced by the hour
- Execution can be outsourced, but once judgment is sold, it is no longer scarce.