The investment review shows Nvidia’s push to link the AI data supply chain with capital, beyond chip sales. [Photo: Shutterstock]

Nvidia is moving to invest in Mercury, a provider of training data for its artificial intelligence (AI) models. If the deal goes through, Mercury could be valued at $20 billion.

On Aug. 19 local time, blockchain outlet Cryptopolitan reported that Mercury sells training data needed to build AI models. Such data is seen as an essential input for Nvidia’s customers to use chip performance in real services. General Catalyst is reported to be discussing a plan to lead the funding round.

Mercury’s performance growth is also steep. Its annual revenue surpassed $2 billion in June 2026. First-half revenue was $614 million. Mercury was valued at $10 billion right after its Series C investment in October 2025. If this deal reaches $20 billion, its valuation would double in nine months.

Its customer base is already centered on major AI companies. Mercury’s clients include OpenAI, Google and Anthropic. Nvidia also pays Mercury and receives support for developing its open-source AI model, Nemotron. The investment is drawing attention because it is seen as less a simple financial investment than an effort to bundle data and chips within the AI supply chain.

Markets are also seeing renewed controversy over recycling finance. Recycling finance refers to a structure in which a company invests in or lends money to its customers or suppliers, and that funding returns to the company through purchases of its products. In such cases, there is concern that sales can appear to come from real demand, but may in fact be the result of the company’s own money returning.

The Bank for International Settlements (BIS) named recycling finance as one of the top 3 risks threatening global financial stability in its 2026 annual report. The Bank of England also said the pace of AI investment is at a "historically unprecedented level" and warned that if companies fail to generate profits, excessive debt financing could strain credit markets.

Nvidia has already invested in cloud operators such as CoreWeave, Nebius and Nscale. These companies have a structure in which they buy Nvidia chips again based on Nvidia funding. OpenAI also signed a $100 billion contract with Nvidia to build data centers that use Nvidia chips, and Nvidia is providing guarantees for some loans.

Nvidia has also signed an agreement with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize up to $500 billion in funding for AI infrastructure. In the past, it paid for the right to use technology from chip startup Groq and later joined a $350 million investment round. After restructuring at the time, Groq was valued at $3.5 billion, about half its peak in September 2025.

Nvidia has countered that its strategy does not amount to recycling finance. It has argued that its chips are "fungible assets," meaning it can sell them to other sources of demand even if a specific customer leaves. If an investment in Mercury materializes, however, the interests of Nvidia and its supplier would become more directly intertwined.

These circumstances are also affecting recent market indicators. Nvidia shares fell, and credit default swaps hit a record high in July 2026. As a result, the talks on the Mercury investment are increasingly likely to serve as both an expansion strategy for the AI supply chain and a market test of Nvidia’s funding structure and how it generates demand.

Keyword

#Nvidia #Mercury #General Catalyst #Bank for International Settlements #OpenAI
Copyright © DigitalToday. All rights reserved. Unauthorized reproduction and redistribution are prohibited.