Mercor raises $750 million to train and test AI agents for enterprise work

Mercor is taking the infrastructure it built for leading AI labs into businesses that need agents to handle their own workflows. The company announced a $750 million Series D on November 10, 2026, at a $20 billion valuation, with funding earmarked for testing agents, improving models and building realistic places for them to practice.

Participants included Benchmark, Felicis, General Catalyst, Nvidia, Salesforce and Zoom, alongside other investors. Mercor said the round also included an employee tender—a share sale by employees. It did not disclose how much of the total goes to that sale rather than fresh operating capital.

Learning the job, then learning the company

CEO and co-founder Brendan Foody argues that knowing a field is not enough for an agent to perform professional work reliably. Mercor divides the required knowledge into two parts:

• How the task gets done: the professional knowledge needed to carry out the work itself.
• How this company does it: the systems, policies and quality standards that shape an acceptable result inside a specific business.

Mercor says it works with companies and experts to build custom evaluations: tests based on the company’s actual workflows. The results help identify where a model needs additional training or more practice in realistic environments. The service is intended to show not only what agents can do, but also what their work costs.

The next step is post-training—further training that adapts an already capable model. Foody argues that enterprises can start with an open model and tailor it using a fraction of the resources needed to build a leading model from scratch. Model builders and computing providers supply the foundation; Mercor supplies data and evaluations for making it fit the business.

Experts and simulated workplaces

Mercor began as an AI-powered hiring platform, according to PYMNTS. Its current training business draws on what the company describes as a network of more than five million experts paid to help train leading AI models.

Practice environments are another part of the system. Mercor said it acquired Deeptune earlier in 2026 because its team had spent two years recreating hundreds of enterprise applications. Those replicas give agents places to practice professional workflows. The new financing will support more environments for learning complex work.

We already spend 3x as much on inference as on employee salaries, and it pays for itself by making our team better at their jobs.

Brendan Foody, Mercor co-founder and CEO

Foody offers Mercor’s own spending as an example of the economics he expects elsewhere. The claimed return is his assessment, not a measured productivity gain disclosed in the announcement.