What changes if this closes: Nvidia stops being just Mercor’s biggest buyer of human-expert training data and becomes its owner-adjacent backer. Right now the relationship is transactional — Nvidia paid Mercor tens of millions of dollars last quarter to source legal, financial and scientific expertise for its Nemotron models, according to Yahoo Finance and BigGo, with a dedicated Mercor team working almost full-time on Nvidia projects. An equity stake at a reported $20 billion valuation would lock in that supply line the same way Nvidia’s 2024 investment in Scale AI (valued then at $14 billion) locked in another vendor — except this time Nvidia would be buying into a company doubling its valuation in under a year on the strength of one customer’s appetite.
Nvidia’s data-supply chain, before and after Mercor equity would be nvidia-owned instead of nvidia-rented, giving the chipmaker leverage over pricing and access as expert-labeled data becomes a scarcer input than compute itself. The numbers make the urgency legible: Mercor’s annualized gross revenue hit $2 billion in June, doubling from earlier in the year per BigGo, while Nvidia poured $18.6 billion into private enterprise and infrastructure bets in the quarter ending April — more than its entire prior fiscal year, according to Yahoo Finance and TradingView. That capital has gone mostly to compute plays like CoreWeave and Nebius; a Mercor stake would mark a rarer bet on the labor-intensive data layer beneath the model itself.
Locking up a customer’s biggest supplier only looks clever until the supplier’s legal exposure becomes the buyer’s problem.
Treat the $20 billion figure skeptically until someone confirms deal size or terms — every outlet here sources it to The Information’s anonymous sourcing, and neither Nvidia nor Mercor has commented, per PYMNTS. There’s also a complication none of the valuation chatter addresses: Mercor was hit with at least seven class-action suits this spring alleging a data breach exposed contractor interview recordings, biometric data and screenshots, which PYMNTS notes the company disputes. For a data-labeling vendor whose entire pitch is trustworthy human expertise, that’s the kind of liability an equity partner inherits along with the upside. Watch Nvidia’s August 26 fiscal-Q2 earnings for any hint of confirmed terms, and watch whether Mercor’s revenue concentration in Nvidia work — already enough to warrant a dedicated internal team — starts to look less like diversified growth and more like dependency.
At Mercor, our vast talent network trains frontier AI models in the same way teachers train students: by sharing knowledge, experience and context that can't be captured in code alone. Each project expands what models understand.
