Nvidia Plans to Invest in AI Data Labeling Firm Mercor, Potentially Valued at $20 Billion
Nashnova编辑部
Nvidia is negotiating a stake in AI data-labeling company Mercor at a valuation that could double to $20 billion — marking its first investment push from chip customers into the upstream data supply chain.
Who is Mercor, and why is it worth $20 billion?
Mercor recruits domain experts — in science, law, finance, GPU programming — to evaluate how AI models perform on specialist tasks. In plain terms = it is the "exam grader for AI," and model developers buy that grading data to improve their systems.
Valued at $10 billion last October, this round would push it to $20 billion, surpassing Scale AI (roughly $14 billion when Meta took a 49% stake) as the highest-valued company in the AI data-labeling space.
Existing backer General Catalyst is negotiating to lead the round. Mercor is roughly three years old and has raised about $520 million to date.
How strong are the financials?
First-half gross revenue hit $614 million, up 70% from last year's full-year total. Full-year gross revenue is projected at $2 billion.
Because Mercor pays contractors roughly 60–70% of gross revenue, first-half net revenue is estimated at $180–250 million. This means → the business model is labor-intensive with a platform take-rate; gross margins are not especially thick.
The profit trajectory is more striking: operating profit turned slightly positive last year, is targeted at $226 million this year, and is projected to jump to $1.7 billion next year. This reflects a data-labeling market in an explosive growth phase.
How deep is the Nvidia–Mercor relationship already?
Nvidia has used Mercor-supplied data to train its two latest Nemotron open-source models and paid Mercor tens of millions of dollars in data-service fees last quarter.
Several Mercor employees devote nearly all their time to serving Nvidia. Put simply = this is closer to an embedded partnership than a standard procurement deal.
Mercor's main clients have historically been closed-source model builders — OpenAI, Google, Anthropic. Nvidia's rapid growth as a customer signals an expansion toward the open-source model camp.
Why would Nvidia invest in a company that doesn't buy chips?
Nvidia's deal pace has surged: in the quarter ending April, it deployed $18.6 billion into private companies and infrastructure funds — more than its full-year 2025 total. Earlier stakes include $2 billion each in CoreWeave and Nebius, plus investments in laser suppliers Lumentum and Coherent.
Mercor is different from those targets — it is not a chip buyer but an upstream data supplier for Nvidia's own AI model development. This means → Nvidia is trying to use equity ties to build a moat in the data supply chain.
Nvidia is not betting on a single supplier, though: it also uses Turing, Scale, and other data providers, runs an internal data team, and relies heavily on synthetic data for training. Whether an equity stake can truly create a durable advantage remains to be seen.
Content is for reference only, not financial advice.