Mercor, an AI upstart founded by 3 post-00s, raises $350 million in funding, valuation breaks $10 billion

What is the concept of dropping out of school at 22 to start a business and doing it at 24 to create a unicorn with a valuation of tens of billions of dollars?

Today, the US AI recruitment unicornMercorofficial announcement of the takeoverUS$250 million (roughly Rs. 1.8 billion) in new financingThe valuation is up toUS$10 billion (equivalent to about 71 billion yuan), which was valued at $2 billion (roughly Rs. 14.2 billion) in February of this year, the5 times.

The AI startup, which was founded in 2023, now has a combined funding ofUS$350 million (equivalent to about RMB 2.5 billion)The following are some of the most important issues that have been raised in the past.OpenAI, Anthropic and other top 5 AI labs in the world included in client list, 17-month revenue run rate growth from $1 toUS$500 million (equivalent to about RMB 3.6 billion).

And it was none other thanThree sophomore dropouts in the 00s: Adarsh Hiremath, CTO; Brendan Foody, CEO; and Surya Midha, COO. They dropped out of Harvard and Georgetown University, respectively, in 2023 to start their joint venture.

▲ CTO Shiremas, CEO Fudi, COO Midha (from left to right)

The business that helped them make their first bucket of gold was AI Recruiting, which uses AI to screen resumes and quickly match candidates to jobs. In February of this year, based on this vast network of professionals, theMercor opens up data labeling, big model evaluation business, which is to say, it contracts with existing expert talent to help the big modeling company with data annotation and provide professional feedback in a short period of time. Today, the total number of experts under its management has reached 30,000, and the daily pay for all experts totalsMore than $1.5 million (equivalent to about $10.65 million).

In February of this year, Mercor's annual recurring revenues have beenUS$70 million (equivalent to about 497 million yuan)With its new business in large model evaluation, the startup has a “hidden gold mine” in the large model evaluation track.

Mercor's new funding was led by venture capital firm Felicis, with participation from Benchmark, General Catalyst and Robinhood Ventures. The new financing will be used in three key areas of focus: expanding the company's talent network, advancing the matching system and training opportunities among experts, and providing faster delivery.

It is worth mentioning that Scale AI, which was previously acquired by Meta to buy shares and poach the CEO, is a strong competitor of Mercor, but after the storm Scale AI's employees and customers have turned to Mercor, which also contributed to the doubling of its revenue.

One, sophomore dropout targets AI recruiting, inadvertently creating a huge network of high-quality talent

Mercor's three founders are conspicuously labeled: post-00s, sophomore dropouts.

Shiremas, Foudy, and Midha were high school classmates who all attended Bellarmine Preparatory School in San Jose, met on the school's debate team, and formed a team that won the U.S. Policy Debate Championship.

It's worth noting that Foudy has been in business since 2021. He founded Serosin with the goal of building the next generation of personal computer infrastructure in the cloud, and managed to reduce the cost of using high-performance computers by 90%.

In 2023, Shiremas, who was a sophomore at Harvard University, and Foudy and Midha, who were sophomores at Georgetown University, chose to drop out of school to focus on entrepreneurship, and Mercor was founded that same year. At the time, Shiremas was a computer science major, and Foudy and Midha were economics and diplomacy majors, respectively.

In the early days of its establishment, Mercor wasUse AI technology to screen resumes and match candidates with the best fit, and qualifying candidates, which is geared mostly towards software engineers, math-related technical positions.

Mercor's enterprise customers use natural language to describe the position, the desired candidate, e.g., “full-time Python developer with computer vision experience,” etc., and its AI tools can perform deep semantic searches in seconds on hundreds of thousands of resumes, portfolio sites, social media platforms, X, AI interview transcripts, and GitHub to find the best matches. GitHub to find the best matches with deep semantic search queries.

Clients can then instantly watch AI interviews with candidates and add matches to their company with a single click.

▲Mercor homepage job postings

The official website of the startup shows that in January 2024, Mercor had reached the million-dollar level of annual recurring revenue and had built a talent pool containing 100,000 users in 25 countries and territories. Then, to meet talent acquisition needs, Mercor continued to expand the talent pool, helping HR teams assess the468,000 applicantsIndia is its largest source of talent, followed by the United States, and the European and South American talent pools are growing rapidly.

By February of this year, its push into AI resume screening revealed that Mercor had inadvertently woven a large network of specialized talent, something that major AI firms crave, and which they hope to leverage to train increasingly complex, large models to improve competitiveness.

This is because as the model's capabilities increase, it requires talent in specialized fields to evaluate it in a short period of time, which requiresAI companies quickly find the right talent and offer temporary positions.

Observing this trend, Mercor scaled up rapidly, expanding its business to include big model evaluation and data labeling. On the one hand, Mercor began hiring contractors who could assess the quality of chatbot answers, and also poached Uber's former chief product officer, Sundeep Jain, to serve as its first president; on the other hand, it continued to expand the size of its talent network, expanding its involvement in the field of job screening to include lawyers, doctors, journalists, and many other industries.

Second, part-time experts work 20 hours a week, and 30,000 experts can earn 16 million dollars a day

Today, Mercor's business system for evaluating the capabilities of large models is maturing.

Mercor currently manages the worldwide30,000 expertsThese experts are responsible for labeling images, writing sentences, and providing professional feedback to help chatbots master human-like thinking and expression, and they earn more than $1.5 million (10.65 million yuan) per day in total.

Among other things, doctors moonlighting as data taggers are tasked with evaluating AI's medically relevant answers and reviewing AI-generated medical research, according to a list of the company's contracts obtained by The Wall Street Journal.Hourly income can reach $170 (equivalent to about 1207 yuan), in a six-week contractWork at least 20 hours per week. Based on a five-day workday, the average expert would need to work more than four hours per day, which means that theDoctors can earn at least $680 per day on a part-time basis (equivalent to about 4,828 yuan).

In addition.Mercor will retain approximately 301 TP4T to 351 TP4T if the customer pays Mercor a data tagging labor rate of $100 (equivalent to approximately $710) per hour.The remainder is passed on to the contractors, whose contracts have an average hourly rate of about $85 per hour (equivalent to about $603).

Earlier this month, Mercor officially announced its first-of-its-kind AI Productivity Index (APEX), which allows AI models to be evaluated based on their ability to perform economically valuable knowledge work. Currently, the APEX contains tasks representing four occupational jobs: investment banking assistant, large legal assistant, strategy consulting assistant and general practitioner (MD).

APEX v1.0 consists of 200 cases distributed evenly across investment banking, legal, consulting, and healthcare. Each case consists of a prompt (a description of the task), a source (the information needed to complete the task), and a scoring criterion (a criterion for scoring the model response).

Its construction consists of five steps: assembling a team of about 100 experts with top-level experience across four specialties; experts generate task descriptions or prompts that describe common workflows in each area; experts generate source documents containing relevant evidence needed to respond to the prompts; experts generate scoring rubrics specific to the criteria of the prompts; and, after the experts have generated the prompts, sources, and scoring rubrics, they are reviewed by a separate expert to ensure quality control. After the expert generates the prompts, sources, and scoring criteria, they are reviewed by a separate expert to ensure quality control.

Its blog mentions that professionals take between one and eight hours to complete tasks in APEX, with an average of 3.5 hours.

In May of this year, OpenAI released HealthBench, a collection of healthcare big model test evaluations, which also uses this APEX system. Based on the APEX evaluation results, GPT-5 received the highest score of 64.21 TP4T, and the best performing open source model was Qwen3, which ranked 7th with 59.81 TP4T.

Third, Scale AI fiasco fuels Mercor surge into commercial litigation

In addition to the revenue generated by its vast network of talent, the fiasco with data labeling startup Scale AI some time ago has caused Mercor's revenues to soar a handful of times.

In June this year, Meta acquired a 49% stake in Scale AI for $14 billion (roughly Rs. 99.4 billion), pushing Scale's valuation to a staggering $29 billion (roughly Rs. 205.9 billion). The company's co-founder and CEO Alexandr Wang then moved to Meta to lead its AI efforts as part of the deal.

This has led some of Scale AI's customers and competitors, to raise concerns about its ability to remain neutral and protect customer data following Meta's investment.

As a result, the deal has in turn led to growth in Mercor's revenue, which has quadrupled since Meta invested in Scale, according to people familiar with the matter cited by the Wall Street Journal.

Meanwhile, Mercor has recruited a number of former Scale employees. Last month, Scale also sued and accused Mercor of allegedly stealing trade secrets and sued former Scale employee Eugene Ling for breach of contract, with the lawsuit revealing that the employee tried to sell Mercor to one of Scale's biggest customers before officially leaving Scale. the lawsuit is still pending, though.

There's also a big debate around Mercor about the potential for AI advances to accelerate the loss of hiring jobs. Foudy argues, however, that Mercor is not replacing labor, but rather automating much of the economy, making labor more valuable in areas where they are still needed.

He told foreign media outlet TechCrunch, “If AI automates 901 TP4T of the economy, then humans become the bottleneck for the remaining 101 TP4T. So every unit of economic output contributed by humans has 10x leverage because the rest has been automated, which means that the way people work is changing as we move to more piecemeal, odd-job-like work patterns.More and more companies today are hiring specialists for short-term projects rather than relying on full-time employees."

Conclusion: Accumulating a Huge Talent Pool with AI Recruiting to Fill the Big Model Assessment Gap

Mercor automates resume screening and candidate matching, and provides AI-driven interviewing and payroll management. Enterprises upload job descriptions in natural language and the system will recommend the best candidates. The large pool of high-quality talent deposited by this model has made Mercor an unexpected “invisible winner” in the large model assessment track.