Key Takeaways
- Mercor secured $750 million in Series D funding at a $20 billion valuation, roughly double its October 2025 valuation (source).
- Mercor plans to invest in AI-agent evaluation, post-training infrastructure, and realistic professional workflow environments.
- Participation from NVIDIA, Salesforce, Zoom, Benchmark, Felicis, and Apollo-affiliated funds reflects broad interest in specialist AI-training data.
Mercor has raised $750 million in Series D financing at a $20 billion valuation, marking a sharp increase from its roughly $10 billion valuation in October 2025 (source). The financing gives Mercor more capital to build infrastructure for evaluating and improving increasingly capable AI models and agents.
The valuation illustrates how investor attention is moving beyond foundation-model developers. Mercor operates in the supporting layer, connecting technology companies with subject-matter experts who train, test, and refine AI systems. Those experts can assess whether model outputs are accurate, useful, and appropriate for specialized professional tasks.
According to AI Weekly, NVIDIA has already used Mercor data in two recent Nemotron models. That connection offers a practical example of how specialist human input can feed directly into foundation-model development rather than remaining a peripheral quality-control exercise.
Mercor says its network includes more than five million experts paid to help train frontier AI models. As DHRMap reported, the latest funding is intended to support AI evaluations, post-training work, and enterprise adoption. Mercor also plans to develop realistic environments in which agents can be tested on complex professional workflows.
A model may perform well on a static benchmark yet struggle when asked to navigate a multistep assignment involving ambiguous instructions, domain-specific judgment, and changing information. Testing agents in simulated work environments can reveal whether they recover from errors, use tools appropriately, and produce results that a knowledgeable professional would accept.
As models become more capable, evaluating them can become more difficult. Simple preference rankings and broad data-labeling projects may not adequately measure performance in law, finance, engineering, medicine, or other fields where context changes the answer. Who evaluates an agent that is expected to complete work normally reviewed by an experienced specialist? Increasingly, the answer involves structured networks of those same specialists.
Mercor is not alone in pursuing this market. Scale AI and Surge AI also operate across data labeling, expert sourcing, model evaluation, and post-training. Competitive differentiation may increasingly depend on the quality of expert networks, the realism of evaluation environments, and the ability to produce repeatable evidence about model behavior.
The investor group reflects that widening strategic importance. Benchmark, Felicis, Apollo-affiliated funds, NVIDIA, Salesforce, and Zoom participated in the financing. Venture investors gain exposure to a rapidly expanding infrastructure category, while strategic investors can gain insight into methods for testing AI systems used in products, enterprise applications, and model-development pipelines.
Still, scale brings operational questions. Managing millions of experts can introduce challenges around identity verification, qualification, consistency, confidentiality, and compensation. Enterprise customers may also want clearer records showing who evaluated a model, what instructions they received, which data they accessed, and how disagreements were resolved.
Those concerns connect the AI-training market with formal governance programs. The NIST AI Risk Management Framework addresses characteristics including validity, reliability, transparency, and accountability. ISO/IEC 42001:2023 similarly provides a management-system approach for governing AI development and use. Mercor's evaluation infrastructure could help customers gather evidence relevant to those programs, although the usefulness of that evidence will depend on documentation, controls, and implementation.
That said, Mercor's $20 billion valuation sets a demanding bar. Mercor will need to show that specialist evaluation is not merely a temporary response to current model weaknesses, but a durable part of how AI systems are developed and monitored. The stronger case is that human expertise changes form rather than disappearing: experts shift from producing every output themselves to defining quality, identifying failure modes, and judging whether agents can handle real work. Mercor's new funding is a sizable bet on that transition.
⬇️