Twenty billion dollars. That is the valuation AI training-data startup Mercor is discussing with new investors, according to a Bloomberg report cited by TechCrunch, doubling the $10 billion mark it reached in October on a $350 million Series C. The talks remain early, but TechCrunch reported that Mercor has already told prospective backers it holds a term sheet at the new price, a detail that moves this from rumor toward a live negotiation. No round has closed.
Mercor pays contract workers, many of them credentialed professionals, to generate the labeled data and reasoning traces that foundation labs use to train and fine-tune large language models. Founder-CEO Brendan Foody said on X that the company’s annualized revenue run rate crossed $2 billion, a figure he described as double what it was four months earlier. That growth claim, unverified outside the company’s own post, is the strongest argument investors have for pricing Mercor at roughly double its nine-month-old mark.
The funding talks follow Thursday’s announcement that Mercor is acquiring Deeptune, whose team helps build training pipelines for AI agents, with the entire staff joining Mercor as part of the deal. Paired with the revenue disclosure, the acquisition reads as a deliberate signal timed to the same week as the valuation story: a growing top line, an expanding product surface, and fresh talent absorbed just as investors weigh a new price.
The backdrop matters here. Mercor spent early 2026 dealing with a data breach and lawsuits filed by several of its contract workers, as Business Insider reported at the time. A $20 billion conversation this soon after those problems suggests investors are pricing the growth numbers well above the governance concerns. Whether that trade holds depends on whether the legal issues stay contained through diligence.
The larger question is why human-data infrastructure commands multiples like this at all. Pretraining on scraped web text has diminishing returns for frontier labs, and the current bottleneck in model improvement sits in post-training: reinforcement learning against verified, expert-checked outputs. That work needs people who can write and grade chain-of-thought solutions in law, medicine, and software engineering, not click-through crowd labor. Labs are paying a premium and signing recurring contracts for that expertise, which is what converts a staffing-adjacent business into one investors price like infrastructure rather than services.
That premium is the wager behind Mercor’s talks, and it is the same wager behind every other data-labeling startup fundraising at outsized multiples this year: that reinforcement learning demand for domain experts keeps compounding for years, not quarters. Operators evaluating vendors in this category should treat a $20 billion price tag as a signal of where labs expect to keep spending, not confirmation that any single vendor has a durable moat. If Mercor’s round closes near that number, it becomes the reference price the next data-labeling raise gets measured against.
TechCrunch reported the Mercor valuation talks on July 9, 2026, citing an earlier Bloomberg report on the funding discussions.
