The AI training startup Mercor is pursuing a valuation of more than $10 billion as it nears an annualized revenue run rate of about $450 million, people familiar with the funding said. The company is receiving several term sheets for a Series C and has had discussions with existing backer Felicis about increasing its investment, the people said.
When closed near the target, one of the year’s largest up-rounds for a provider of AI infrastructure built around human-in-the-loop workflows. Some offers have already valued Mercor close to $10 billion, The Information reported earlier.
A Rapid Rise From Series B
Mercor last reported a $100 million Series B at a $2 billion valuation, and has since told investors it is on track to hit $500 million ARR quicker than a number of high-profile counterparts.
The firm’s speed is enabled by a unique combination of service and software supporting core AI development cycles – especially reinforcement learning with human feedback (RLHF) and expert evaluation. That formula is driving out revenue forward while maintaining burn, a growth investor profile that is sexy as sin in a chopp town lates-stage market.
How Mercor Earns Its Money
For example, Mercor leverages subject matter experts, who are typically doctors, lawyers, scientists, and other credentialed roles, to label data, validate model output, and scaffold and verify reinforcement learning.
The company claims to provide contractors to leading AI labs and model builders, including a number of Big Tech players and marquee research organizations. A focus group of core accounts, familiar sources say, contribute outsized revenue, including one foundation model leader—the type of opportunity and concentration risk that is typically seen among companies at the forefront of AI.
Why RL and Expert Feedback Are Thriving
And even as models scale, systems remain reliant on high-quality human judgment to help ensure that outputs meet safety, legal, and brand standards. The OpenAI and Anthropic studies report that expert feedback can have substantial effect on factuality and harmfulness, but it is manually-intensive and domain specific.
Industry research from McKinsey and Gartner identified data preparation and curation as the biggest hidden costs in enterprise AI initiatives. For industries like healthcare, finance, government, credentialed reviewers and auditable pipelines are becoming a purchasing requirement, not a good-to-have. On that front, there is a tailwind for providers that can deliver vetted experts at scale with repeatable quality controls.
Software Expansion and Marketplace Ambitions
To offset the pure services revenue driven model, Mercor is adding software around the RL workflows—assignment routing, consensus scoring, disagreement analysis, reward model management—intended to enhance throughput/efficiency and margins. The roadmap also involves an AI-powered recruitment marketplace that connects pre-vetted experts with model builders on a near real-time basis.
Execution capacity is the emphasis: the company recently hired veteran operator Sundeep Jain, previously a product executive at Uber, as its first president, reported Forbes. The hire is a sign of pushing more toward this kind of scaled operation, with standard SLAs and enterprise-level compliance.
Crowded Field, Shifting Moats
Competition is intensifying. Category pioneer Scale AI has pushed further into RLHF and also into eval. And Surge AI, the advocate for the digital personal assistant, is rumored to be raising at a valuation well past earlier marks. Turing and other boutique firms are competing for many of the same expert labor pools. Elsewhere, a new hiring platform announced by a leading model provider suggests that customers might vertically integrate some targeted expert feedback flow.
The right question is, where does the defensible moat form? +:+ Proprietary expert networks; + superior quality systems; or, + software that drives out cycle times and unit costs. Late-stage AI infra valuations range massively according to private-market trackers at PitchBook and CB Insights; companies with credible software leverage and durable enterprise contracts will often generate higher revenue multiples than pure labor marketplaces.
Investor Lens: Multiples, Margins, Risk
$10bn price x ~$450mm run rate = Mercor looks like a forward multiple that is betting on a mix shift to SW and stable expansion at largest accounts
Bears are concerned about customer concentration, the risk of compressed margins as more entrants pile into RLHF services, and that better synthetic data and automated underwrite engines could spoil the growth of human-in-the-loop demand. Shorter-term, though, most big-model builders are ramping up, not down, investment in high-skill feedback and red-teaming.
For now, investors seem willing to shell out for a rare asset: a scaled, profitable AI training outfit with line of sight to a half-billion in ARR and a plausible avenue to yet more software income.