Mercor, a hot human-in-the-loop AI training provider, said it was likely to be valued at more than $10 billion on an annualized revenue run rate of around $450 million, the people said of the fundraising talks. Several investors have floated offers to the company, and the final terms may still change as negotiations proceed.
The target represents a black-and-white step-up from Mercor’s February financing — a $100 million Series B at a $2 billion value led by Felicis — it highlights how fast revenue has scaled in a niche that is at the nexus of the AI boom.
Felicis declined to discuss the new round. The Information had previously reported that some firms had been holding out for spurious $10 billion valuations.
The expert-training AI business
A network of domain specialists — scientists, doctors, lawyers, and other subject-matter experts — who oversee models and make them better through data curation, evaluation, and reinforcement learning workflows, is Mercor’s main product.\xa0 The company takes a mix of hourly finder’s fees and matching rates on these projects, and considers itself a “managed marketplace” for high-skill feedback rather than commoditized labeling.
The company has reportedly told partners that it provides contractors to five of the biggest AI labs: Amazon, Google, Meta (previously Facebook), Microsoft and OpenAI, in addition to Nvidia. People familiar with the business say a disproportionate share of revenue comes from a subset of those accounts, among them OpenAI — a concentration that fuels near-term growth but introduces platform risk should big buyers insource critical training functions.
Growth rates, margins, and the 20 times question
Internally, Mercor has told investors it expects to hit the $500 million ARR faster than Anysphere, the creator of the Cursor coding assistant that reached $500 million in ARR around its one-year launch anniversary, according to a person familiar with the investor materials. Unlike many peers that are still in high-burn mode, Mercor made an estimated $6 million in profit in the first half of the year, according to Forbes.
A $10 billion sticker on a ~$450 million run rate suggests a revenue multiple of north of 20x — rich even for high-growth AI infrastructure! Bulls claim this premium partly prices in structural scarcity: aligned high quality human feedback is a gating function to model performance, safety, and enterprise adoption. If Mercor can put software on top of services, it could make software-like margins on top of a capacity-constrained expert network.
From services to software: the margin expansion play
To diversify from labor revenue, Mercor is building tooling for reinforcement learning workflows — the systems that check or challenge the outputs of models, collect detailed feedback signals and standardize quality assurance across projects. That infrastructure can make feedback loops tighter, cut down variance across annotators and transform what is largely custom work now into something repeatable and measurable.
The company has also mentioned an AI-driven recruiting marketplace for expert evaluators. If it works, this might drive down customer acquisition costs, increase utilization for the best contributors and provide Mercor with proprietary data on the quality of talent — a defensible asset as the competition gets fiercer.
Omnidirectional competitive pressure
Mercor competes in a crowded and rapidly changing sector. Scale AI, which last raised at a valuation north of $13.8 billion, has grown from labeling into model evaluation and reinforcement learning. Surge AI is said to be weighing a new round at a roughly $25 billion valuation. Specialist firms like Turing Labs are also continuing to dive further into RL and safety evaluation.
There is also the possibility of vertical integration for customers. OpenAI created an employment system recently, which has led to many rumors that the company may combine talent acquisition with its proprietary RL training services. For suppliers like Mercor, the pushback is differentiation on expertise density, reliability and end-to-end workflow software, not just access to humans, but with measured results and compliance at scale.
Leadership, heritage and the risk of execution
Mercor was co-founded by Thiel Fellows and Harvard dropouts Brendan Foody (CEO), Adarsh Hiremath (CTO), and Surya Midha (COO), who are all in their early twenties. To deepen its ranks, the company hired its first president, Sundeep Jain, a former chief product officer at Uber, as a way to help run the company, Forbes reported. The recruit is an indicator that the company wants to formalize processes for quality, security and client delivery as it grows.
Execution will depend on diversifying revenue away from a few giant customers; proving that software can lift margins without sacrificing accuracy; and negotiating any turn toward in-house RLHF at large labs. On the financing side, there is strong investor demand, but the exact size, structure and valuation of the new round are still in flux, investors said.
What to watch
Key takeaways were breaking $500 million in ARR, software vs services revenue mix, customer concentration, and depth of partnerships beyond flagship labs.
For now, Mercor’s pitch is straightforward: in the arms race for smarter models, the scarce input isn’t compute — it’s trusted human judgment, and the packaging of that judgment with discipline and data.
