A growing sector within the artificial intelligence (AI) industry is transforming highly specialized professional skills into gig work, supplying critical training data for AI models. Start-ups such as Mercor, founded in 2023, are hiring tens of thousands of contractors—ranging from voice actors fluent in Hebrew to Ph.D.-level physicists and medical professionals with experience in specific healthcare systems—to annotate and vet data for use by AI companies.

These firms operate as intermediaries in a rapidly expanding marketplace that caters to leading AI developers like OpenAI and Anthropic. Unlike earlier stages of AI training, which depended heavily on lower-paid workers performing repetitive tasks, today’s data-labeling work increasingly calls for expert knowledge to refine complex AI models. For instance, mathematicians annotate scientific proofs, lawyers review legal documents, and professors grade academic essays. This shift represents a move up the “value chain,” with start-ups such as Mercor, Scale AI, and Handshake growing rapidly in response to the surging demand.

Mercor’s recent funding round placed its valuation at $10 billion, and the company has reportedly entered discussions to double that figure. Similar ventures are also posting impressive revenue gains; Handshake reported its annualized revenue more than doubling to over $1 billion since early 2026, after pivoting to focus on data training.

While profitable, the model faces inherent tension. These start-ups depend on AI systems remaining imperfect enough to require continual human intervention but improving enough to provide value to their clients. Once models master a given task, demand for additional training data in that area may decline significantly, potentially jeopardizing the businesses and the contractors they employ.

To sustain growth and diversify, some companies are expanding their offerings toward more complex simulations. Mercor recently acquired Deeptune, a start-up that creates simulated operational environments of widely used workplace software such as Slack and Salesforce. These environments allow AI models to observe interactions within organizations like investment banks, potentially enabling AI to understand multifaceted workflows and business processes.

Gig workers attracted to these roles often seek them for supplemental income, career transitions, or to gain experience with AI technology. However, contractors have reported challenges with inconsistent work, unclear evaluation criteria, and demanding schedules. For example, Amanda Brown, an assistant professor of biology, described early enthusiasm for data-labeling gigs evaporating amid mandatory meetings, late hours, and disappointing compensation relative to effort. Several contractors also express concern over the increasingly intricate nature of tasks as AI systems improve, making it harder to identify errors or gaps for correction.

Labor disputes have emerged, with some contractors filing lawsuits alleging misclassification, unfair pay, and data breaches. A data breach at Mercor last spring resulted in multiple lawsuits from contractors claiming personal information was exposed. Despite these challenges, the sector continues to attract highly credentialed individuals, including those with advanced degrees and professional pedigrees, who might otherwise face limited job opportunities amid tightening labor markets.

Executives in this space predict that white-collar jobs will increasingly involve working alongside AI tools and overseeing their outputs—a hybrid role that blends professional expertise with model training. Handshake’s chief operating officer, Jonathan Stull, characterized this dynamic as an inevitable evolution, with many professionals effectively automating routine aspects of their work while focusing on guiding AI systems.

As AI development advances, these data-labeling companies play a crucial role in transferring specialized human knowledge into machine learning models. Collaboration with major AI labs like Google and OpenAI underscores the strategic importance of this work amid intensifying competition to refine AI capabilities. Nonetheless, the long-term viability of the gig-based model for sourcing expert training data remains uncertain, contingent on how quickly AI achieves proficiency and the ability of start-ups to adapt to changing demands.