A $750 million bet on agent reliability
Mercor’s latest raise is not just another oversized AI financing round. The San Francisco company announced a $750 million Series D valuing it at $20 billion, with the round meant to expand its evaluation and post-training infrastructure for companies that want to build AI systems on their own data and expertise . The announcement confirms a valuation double from Mercor’s $10 billion Series C in October 2025, when it raised $350 million .
The investor list is also a signal. Mercor said participants included Benchmark, Felicis, Fundomo, funds managed by affiliates of Apollo, G Squared, General Catalyst, Glade Brook Capital, NVIDIA, Salesforce, Sapphire Ventures, StepStone Group, XTX Ventures and Zoom . Reuters separately reported participation from Benchmark, Felicis, Fundomo, Nvidia and Apollo-affiliated funds, framing the deal as part of a wider surge in private AI funding .
For enterprises, the practical question is no longer whether a model can produce a plausible answer in a demo. It is whether an AI agent can perform a job inside real workflows, using company-specific rules, systems, exceptions and quality bars. Mercor’s pitch is that companies need independent evidence of agent performance before they let autonomous systems touch business-critical work.
From expert data to enterprise evaluation
Mercor began as an AI-powered hiring platform, but its current business is better understood as a marketplace and infrastructure layer for human expertise. Reuters described the company as connecting technology companies with subject-matter experts whose knowledge is used to train, test and improve AI models and agents . Mercor says it has built a network of more than five million experts who are paid to train frontier AI models using their specialized knowledge .
That matters because the next wave of enterprise AI will be evaluated less by generic benchmark scores and more by job-specific outcomes. A legal agent must understand a firm’s drafting standards. A medical agent must respect clinical protocols. A finance agent must handle internal models, assumptions and review processes. In Mercor’s framing, professional work depends both on knowing how a task is done and knowing how a particular company wants that task done .
The company says it works with enterprises and domain experts to build custom evaluations based on actual workflows, using the results to measure whether models meet internal requirements and where more training or realistic practice is needed . This is the deeper meaning of the round: Mercor is selling evaluation as a system of record for AI work, not merely as a checklist at the end of model development.
Why the Deeptune acquisition matters
The Series D also puts fresh attention on Mercor’s acquisition of Deeptune earlier this year. Mercor said Deeptune’s team spent two years recreating hundreds of enterprise applications where AI agents can practice professional workflows . SiliconANGLE reported that part of Mercor’s post-training runs on Deeptune software and that the new money will fund research, evaluation infrastructure and more such environments .
That is an important shift. If AI agents are supposed to execute multi-step work, companies need more than static tests. They need simulated workplaces where agents can try tasks, fail safely, receive feedback and improve. In other words, enterprise AI evaluation is starting to look like a blend of exam room, simulator and training ground.
This is especially relevant as companies move from copilots to agents. A copilot suggests. An agent acts. Once systems act on behalf of employees, evaluation must cover cost, reliability, compliance, handoffs, escalation and edge cases. Mercor’s financing suggests investors believe the testing layer around agents can become a large business in its own right.
The enterprise turn
Mercor’s own language makes the strategy clear. The company said it is bringing capabilities built for frontier labs to enterprises “across the economy,” helping them evaluate what agents can do, what they cost and how to improve performance . The company also said the new funding will accelerate investment in evals, post-training research and environments where agents learn complex workflows .
The Next Web reported that Mercor’s main buyers so far have been frontier AI labs, and that the new money is meant to sell the same service to other companies . SiliconANGLE similarly noted that frontier labs have been the main buyers of expert-driven training data so far, while investors expect demand to broaden as open models become good enough for enterprise customization .
That explains why the round includes strategic names such as NVIDIA, Salesforce and Zoom . Each is exposed to the enterprise AI stack from a different angle: compute, business software and workplace communications. Their presence does not automatically validate Mercor’s economics, but it does show that major AI ecosystem players see value in infrastructure that can measure and improve agent performance.
A valuation that prices in a new category
A $20 billion valuation puts Mercor in rare company among private AI startups. Reuters cited Crunchbase data showing that AI-driven startups raised $102 billion, or 64% of global venture capital, in the third quarter of 2026 . In that environment, Mercor’s Series D reads as both a company milestone and a market thesis: the bottleneck in AI adoption may be less about model access and more about trustworthy deployment.
Tech Funding News reported that Mercor raised the round at double its last valuation and is pushing into enterprise AI evaluation, including competition with Scale AI and Surge AI . It also reported that Mercor announced an annualized revenue run rate above $2 billion in July, while noting that this figure represented gross billings and that contractors kept a large share of it . That distinction matters because a huge run rate does not automatically translate into equally huge net revenue or margins.
The financing also included an employee tender, according to Mercor’s blog post, but the company did not break out how much of the $750 million represented primary capital versus secondary share sales . SiliconANGLE highlighted the same point, noting that Mercor had not disclosed the split between fresh capital and the tender offer .
The risk behind the excitement
Mercor’s opportunity is obvious: every enterprise considering AI agents needs some way to test them against real work. But the risks are equally clear. The company must prove that demand from enterprises can grow beyond the foundation-model labs that helped define its early market. It also has to show that expert networks, custom benchmarks and simulated environments can scale without becoming services-heavy.
There is another tension. Mercor’s advantage comes from human expertise, but its customers are buying automation. The company’s long-term margins will depend on how efficiently it can turn expert input into reusable evaluation systems, post-training workflows and agent environments. If each customer requires bespoke, labor-intensive work, the business could look more like consulting than software. If the infrastructure becomes repeatable, the valuation becomes easier to understand.
For now, investors have made their bet. Mercor’s $750 million Series D says that quality assurance for AI agents is no longer a side feature of model development. It is becoming a standalone layer in the enterprise AI stack. As more companies pay agents to do work, they will also pay someone to prove that the work is reliable.
Comments
Be the first to comment.