Mercor raises $750 million to build AI evaluation for enterprises
The San Francisco-based data company is expanding its platform to help businesses train and test autonomous agents on specific professional workflows.
The short version
- Mercor raised $750 million in Series D funding at a $20 billion valuation.
- The company operates a network of over five million experts who provide professional knowledge for AI training.
- Mercor helps businesses create custom benchmarks rather than relying on public tests.
- The company recently acquired Deeptune to help simulate real-world business environments for AI practice.
- Investors include major names like NVIDIA, Salesforce, and General Catalyst.
Why it matters
As companies move from using general chatbots to specialized AI agents, they need a way to prove these systems actually work on their specific data. Mercor provides the infrastructure to test if an AI can follow a company's unique rules and handle its specific edge cases.
Securing a $20 billion valuation
Mercor, a San Francisco-based artificial intelligence data company, has successfully raised $750 million in Series D financing. This funding round values the company at $20 billion and marks a significant milestone for the startup founded in 2023. The capital will be used to expand the company's platform for training and evaluating AI models and autonomous agents.
A diverse group of prominent investors participated in this latest round of funding. The group includes major technology players and venture capital firms such as NVIDIA, Salesforce, Benchmark, and General Catalyst. Other participants include Fundomo, G Squared, Glade Brook Capital, Sapphire Ventures, StepStone Group, XTX Ventures, and Zoom.
While Mercor has already established itself as a provider for leading AI model developers, this new capital is intended for a different goal. The company plans to use the funds to accelerate its expansion into the broader enterprise market. This shift focuses on providing specialized tools for companies that need to improve AI systems for industry-specific tasks.
Moving beyond public benchmarks
One of the primary challenges for businesses today is determining if an AI model actually works for their specific needs. Adarsh Hiremath, co-founder and co-CEO of Mercor, noted that a model might score well on a public benchmark but still fail on the specific policies and edge cases of a business. Because every company operates differently, standardized public tests are often insufficient for professional deployment.
To solve this, Mercor develops customized evaluation benchmarks that assess AI performance against the unique requirements of individual organizations. These benchmarks allow a company to measure how reliably an AI agent follows internal protocols or handles specific exceptions. This process ensures that the AI can complete complex workflows at an acceptable cost before it is ever put into production.
These custom benchmarks also serve as a foundation for post-training programs. By using examples and feedback drawn from a company's own professional workflows, Mercor helps businesses refine their AI systems. This allows for a more tailored approach where the AI learns from the specific work the company actually performs every day.
Leveraging a massive expert network
At the core of Mercor's infrastructure is a massive network of over five million experts. These individuals contribute professional knowledge to various stages of the AI lifecycle, including training, evaluation, and post-training programs. This human-in-the-loop approach ensures that the data used to train models is grounded in real-world expertise rather than just internet text.
The company's strategy is built on the idea that every job in the economy will eventually require its own AI agents. Co-founder and co-CEO Brendan Foody explained that every company will need a way to train and evaluate those agents on their own specific work. Mercor spent the last two years building this infrastructure with frontier labs and is now moving that technology to the wider economy.
By combining this expert network with its training infrastructure, Mercor helps enterprises establish clear performance standards. This allows organizations to maintain greater control over the reliability and operating costs of their AI systems. As autonomous agents become more common in professional functions, having a way to verify their performance becomes a critical requirement for any business.
Simulating real-world environments
To make AI training more effective, Mercor is investing in research and simulated enterprise environments. These simulations allow AI agents to practice complex professional tasks in a safe, controlled space. This mimics the real-world business processes that the agents will eventually encounter in a live production environment.
A key part of this capability came from Mercor's acquisition of Deeptune earlier in 2026. This acquisition added a team that had spent two years recreating hundreds of enterprise software applications to serve as training environments. These environments allow AI systems to practice tasks and be evaluated before they are deployed to handle actual company data.
Mercor is also building on its APEX series, which is a public benchmark designed to measure AI performance on economically valuable professional work. By combining these public benchmarks with private, simulated environments, Mercor provides a comprehensive toolkit for enterprises. This helps companies ensure their AI agents can handle the nuances of professional work without risking errors in the real world.
The shift toward enterprise AI
The current market is seeing a shift in who is buying expert-driven training data. While frontier labs were the primary buyers in the past, that dynamic is expected to change as open models become more capable. Jai Das, a partner at Sapphire Ventures, noted that as open models become good enough to build on, every enterprise will want to evaluate and train agents on their own work.
This trend reflects a growing demand for infrastructure that helps businesses adapt existing AI models to specialized applications. Instead of every company trying to build their own foundational models from scratch, many are looking for ways to customize existing ones. Mercor is positioning itself to be the primary infrastructure provider for this customization process.
The new financing will support continued investment in the research and infrastructure needed to make this possible. Mercor aims to provide organizations with the tools to manage the reliability and costs of autonomous agents. As these agents become more widely deployed across professional functions, the ability to train and evaluate them on specific work will become a standard business requirement.
What to watch
- How many more enterprise software applications Mercor will recreate as training environments.
- The adoption rate of custom benchmarks versus public benchmarks among Fortune 500 companies.
- How the acquisition of Deeptune impacts the speed of AI deployment for Mercor's clients.
Try it
Look for your company's internal AI policies and consider if a standard public benchmark (like those used to test ChatGPT) would actually be enough to prove the AI is safe for your specific office workflows.
Sources and links
- Reportingcitybiz: the article we sourced
First reported by citybiz. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.