Zhidx (WeChat official account: zhidxcom)
Compiled by | Sun Xiaowen
Edited by | Cheng Qian
Zhidx reported on November 11 that last night, Mercor, a U.S. AI training data unicorn, announcedit had raised $750 million (approximately RMB 5.038 billion)in a Series D financing, with Nvidia, Zoom, Benchmark, and others participating. Following the round, its valuationreached $20 billion(approximately RMB 134.334 billion), double the $10 billion (approximately RMB 67.167 billion) valuation following its Series C financing last October. The deal also included an employee share sale arrangement, according to reports.
Mercor said the funds will be used formodel evaluation, post-training infrastructure, and related research, helping clients test whether models can perform their jobs and improve performance through further training. The company also plans to build more environments in which agents can practice complex workflows, expanding the training and evaluation services it has previously provided to frontier AI labs to more companies.
In a funding blog post published simultaneously, Mercor co-founder and co-CEO Brendan Foody revealed that the company’sinference spendingfor running AI models has reached three times its staff payroll, butthe gains from AI improving teams’ work performance are sufficient to offset the expense.
Founded in 2023 and headquartered in San Francisco, Mercor connects experts from various fields to provide training data and model evaluation services to technology companies, applying professionals’ expertise to train, test, and improve models and agents. According to Mercor, its expert network covers more than five million people, and the world’s top five AI labs, including OpenAI and Anthropic, are all its clients.
These experts provide models with experience and judgment from their actual work. According to Mercor, doctors can help AI identify early warning signs that might otherwise be overlooked, banking professionals can train agents to conduct financial analysis, and lawyers can help models improve their reasoning over legal precedents.
Mercor was founded in 2023 by three co-founders born after 2000—CEO Brendan Foody, CTO Adarsh Hiremath, and COO Surya Midha—whodropped out together to found the company. The three wereonly 19 years old when they started the business. They became friends on their high school debate team; Hiremath attended Harvard University, while Foody and Midha attended Georgetown University.
▲CTO Hiremath, CEO Foody, and COO Midha (from left to right)
Before this round, Mercorhad raised approximately $484 million (approximately RMB 3.248 billion) in total. In October 2025, the company closed a $350 million (approximately RMB 2.351 billion) Series D financing at a valuation of $10 billion (approximately RMB 67.167 billion); in February of the same year, it closed a $100 million (approximately RMB 672 million) Series B financing, with Felicis leading both rounds. In September 2024, Mercor said it had completed a $30 million (approximately RMB 202 million) Series A financing led by Benchmark; in January of the same year, it disclosed $3.6 million (approximately RMB 24.18 million) in financing led by General Catalyst.
On November 10, Sapphire Ventures disclosed in its investment announcement for the current funding round that Mercor had annualized total revenue of more than $2 billion (approximately RMB 13.433 billion) and had generated positive free cash flow.
According to Sapphire, Mercor initially used AI to help companies recruit talent before shifting its focus to providing expert training data for AI models. The company currently still uses AI-led interviews to screen experts.
Mercor said it is currently building custom evaluations around companies’ actual work to determine whether agents can comply with internal rules and complete specific tasks, and then using the results to conduct post-training and simulated-environment exercises. Foody believes every industry will need AI agents capable of performing specific jobs, and companies will also need to use their own business tasks to train and evaluate them.
Conclusion: AI startups attracted nearly RMB 700 billion in the third quarter, while data remains scarce in fields such as law and finance
According to Crunchbase, AI startups raised a total of $102 billion in the third quarter of 2026 (approximately RMB 685.103 billion), accounting for 64% of total global venture capital investment. In this funding boom, supporting services such as training data and model evaluation also received capital support. Mercor is targeting the use of agents in enterprises, providing models with training data and evaluations for specialized tasks. Companies require AI to master industry knowledge, use internal systems, and complete work according to business rules, creating demand for specialized data services.
Mercor and its investors believe public web data cannot fully cover work experience in fields such as law and finance, and companies’ internal operating procedures are also lacking. Professionals can use this to design tasks, create data, and evaluate model performance. However, when companies purchase these services, what ultimately matters is how well AI can perform the work. For Mercor, training and evaluation must not only help clients improve task-completion quality and reduce human rework, but also help clients consider how the returns generated by AI can cover their investment.
Source: Mercor, The Information
