Mercor’s $750M, $20B bet: Why enterprises can’t trust AI agents without human benchmarks

Mercor has raised $750 million at a $20 billion valuation, doubling its size in a single round to expand its human-powered AI evaluation infrastructure beyond frontier labs into enterprises. The funding, announced today, includes corporate backers like Nvidia, Salesforce, and Zoom, alongside returning investors such as Benchmark and Felicis. The company’s co-CEOs, Brendan Foody and Adarsh Hiremath, argue that AI models—even those performing well on standard benchmarks—may struggle in business environments where unique workflows, edge cases, and internal policies create distinct challenges.

Mercor’s model relies on a network of over 5 million domain experts who assess AI performance against real-world scenarios. While the company’s revenue has grown rapidly—Foody previously noted it had surpassed $2 billion annualized—its focus is now shifting toward enterprise adoption. Enterprises, Mercor contends, require tailored evaluation methods to determine whether AI agents can handle their specific systems, rather than relying on generic assessments. The company builds these custom frameworks for each client, leveraging infrastructure developed during its two-year collaboration with AI labs.

A significant component of this infrastructure comes from Deeptune, a New York-based startup Mercor acquired in July. Deeptune’s team had previously recreated hundreds of enterprise applications as simulated training environments for AI agents—a step Mercor believes is essential for ensuring models function reliably in production. The new funding will accelerate this work, expanding evaluation environments and research capabilities. The round also included a tender offer for employee shares, though the company has not disclosed how much of the capital is allocated to equity buybacks versus fresh investment.

The funding reflects a broader industry trend: while AI labs have historically driven demand for high-quality training data, enterprises are now prioritizing deployment but lack reliable ways to validate AI agents. Sapphire Ventures’ Jai Das suggests Mercor has built an advantage by developing infrastructure through its lab partnerships that enterprises may struggle to replicate internally. The company’s valuation has surged from $10 billion in October 2025 to $20 billion today, signaling strong investor confidence in addressing this gap.

However, challenges remain. Additionally, can Mercor scale its expert network quickly enough to meet growing demand? The company’s emphasis on post-training evaluation, rather than just data labeling, distinguishes it from peers, but the broader market for AI trust infrastructure remains in its early stages. Competitors like Infisical, which raised $40 million last month to secure AI agent access in enterprise systems, are tackling related but distinct problems, hinting at potential fragmentation in how companies approach AI validation.

For now, Mercor’s funding signals a bet that enterprises will only adopt AI agents if they can verify their effectiveness. The $20 billion valuation suggests investors share this view. What remains unclear is whether Mercor’s infrastructure can scale as adoption accelerates—and whether companies will see it as essential or an optional expense.

Sources: siliconangle.com