The training data that powers Nvidia’s open-source model ambitions may soon have a new financial backer: Nvidia itself.
The chip maker is in talks to invest in Mercor, an AI data supplier that helps it develop open-source models, according to people familiar with the matter. The round is expected to value Mercor at about $200 billion… at about $20 billion, and existing investor General Catalyst is in discussions to lead the financing. The talks were reported by The Information and confirmed by people close to the companies.
Mercor sits at a specific point in the AI supply chain. The company provides high-quality data labeling, curation and a network of specialized workers and contractors used to prepare training data for large language models. For most of its life, its revenue has come from the closed-model labs: OpenAI, Google and Anthropic have been its largest customers, paying for the data work that precedes model training. The balance has been shifting, though. As Nvidia has pushed its Nemotron line of open-source models, built to compete with the world’s best open-weight systems, its purchases from Mercor have grown.
The investment would extend Nvidia’s reach from silicon into the data layer. Nvidia has long described itself as an infrastructure company, selling chips, networking and software to everyone building AI. A stake in a data supplier would give it a window into, and a claim on, one of the scarcest inputs in the business: clean, well-labeled data at scale. Analysts said the logic mirrors Nvidia’s earlier investments in cloud providers and energy startups, positioning the company across the stack rather than only at the chip layer.
For Mercor, the pairing is strategic on both sides of the table. A round led by General Catalyst with Nvidia participating would bring in the world’s most valuable chip company as an investor and, increasingly, a customer. Mercor’s revenue mix would continue to tilt toward Nvidia’s open-model program, which the chip maker funds in part to broaden the ecosystem around its hardware. Open models that run well on Nvidia chips, trained on data Mercor supplies, reinforce the flywheel at the center of Nvidia’s strategy.
The data business has become one of AI’s most contested battlegrounds. Model developers have spent heavily on data acquisition, licensing and labeling, and a handful of companies have emerged as critical intermediaries between raw internet content and usable training sets. Mercor’s model, which combines software tooling with a global contractor network, has drawn comparisons to both a data platform and a labor marketplace. Its valuation, if the round closes at $20 billion, would place it among the most valuable companies in the AI data industry.
The talks come as the open-source model market heats up. Meta, Alibaba and a roster of startups have released competitive open-weight models, and Nvidia’s Nemotron line is its answer to that trend, giving developers a way to run frontier-class models on its accelerators without depending on closed labs. Data quality is the difference between a model that is merely open and one that is actually useful, and Nvidia’s willingness to invest in suppliers reflects the stakes.
The exact size of the round, the final valuation and whether Nvidia commits at all remain unresolved, people familiar with the matter said. Mercor’s reliance on closed-model labs for the majority of its revenue also raises a question about incentives: OpenAI, Google and Anthropic are Nvidia’s customers, but they are also competitors in the model business, and their willingness to keep paying a supplier backed by the chip giant is untested. Analysts said the arrangement could hand Nvidia outsized influence in those relationships, a consideration that may complicate the deal’s reception.
For investors, the talks are another data point in Nvidia’s transformation from a hardware vendor into a broad AI holding company. The firm has invested in cloud providers, robotics startups and energy companies, and a data supplier would fit that pattern. The practical effect, if the deal closes, is that Nvidia gains a direct interest in the quality and cost of the data feeding both its own models and the broader ecosystem. People close to the discussions said the companies are working through terms and that no agreement is guaranteed.
The broader lesson, analysts said, is that the AI industry’s value chain is consolidating around the companies that control scarce inputs. Whoever owns the data pipeline shapes what models can learn, and the chip giant’s move into that territory is a sign of how central data quality has become to competitive advantage. Chips, power, data and talent are the four bottlenecks, and Nvidia now has its hand in three of them. Mercor, for its part, would gain a patron with the deepest pockets in the industry and a reason to see its data business thrive.
