Nvidia Corp is working on a new round AI infrastructure deals that could be worth more than $750 billion. The latest deals extend Nvidia’s aggressive AI investment strategy and raise concerns that the company could be helping create an AI bubble. Critics say Nvidia’s financing model may artificially inflate demand for AI chips and push the company’s valuation higher than it should be.
Nvidia announced a new AI partnership with SK Group, parent company of South Korean memory chip maker SK Hynix. Nvidia said that this agreement with SK Group is worth more than $500 billion. Nvidia is also discussing another big deal with OpenAI, according to people familiar with the matter, Bloomberg reports.
OpenAI deal and AI funding
Under discussions, Nvidia could guarantee up to $250 billion to help OpenAI lease computing power from a giant US data center project. Nvidia is also discussing $350 billion in financing for OpenAI’s purchase of Nvidia chips for that U.S. project, according to a person familiar with the matter. If completed, it would become one of Nvidia’s largest financing deals with a customer.
What is circular financing?
Critics have warned for months that Nvidia’s investments constitute “circular financing.” Circular financing means that Nvidia invests in or finances companies, and then those same companies spend the money buying Nvidia’s AI chips or services. Experts say this structure could create wrong business incentives, encourage risky spending and increase losses if demand for AI slows.
According to Bloomberg, Gary Tan, portfolio manager at Allspring Global Investments, said, “Although Nvidia’s investment and partnership reinforces confidence in the long-term AI buildout, investors remain concerned about circular financing.” “The capital is increasingly being used to finance the deployment of future AI customers and infrastructure,” Tan said.
Also read: OpenAI AI agent Hugging Face hacked, went undetected for days: Report
Nvidia isn’t the only company using this approach. Google recently agreed to support lease payments for five data center sites used by AI startup Anthropic, helping it secure a $35 billion loan. Another concern is the rising debt among AI companies as they borrow heavily to build data centers and buy expensive AI chips.
OpenAI talks still ongoing
Bloomberg reported that the proposed OpenAI deal includes developing a $500 billion AI hub in Ohio. According to people familiar with the matter, discussions between Nvidia and OpenAI are still in the early stages.
Talks could still fail or the terms of the financing could change, the people said. “Nvidia is guaranteeing more “OpenAI’s data center loan deepens vendor financing that is already under scrutiny,” said Billy Leung, investment strategist at Global X Management. “This is as much a reminder of the funding tensions in AI buildouts as it is a demand signal,” Leung said, according to a Bloomberg report.
According to Bloomberg, Nvidia’s closer relationship with SK Group will also improve its access to high-bandwidth memory (HBM) chips. Jensen Huang said the deal, valued at more than $500 billion, includes Nvidia buying memory chips from SK Group, as well as SK Group buying Nvidia’s AI supercomputers. As Bloomberg noted, Huang said in the interview, “So between us, we’re going to have half a trillion dollars of trade.”
Jensen Huang defends the strategy
Jensen Huang has said that Nvidia’s investments in companies like OpenAI and Anthropic are meant to generate returns as well as strengthen the AI ecosystem. Nvidia has also invested in AI infrastructure companies including IREN Ltd., CoreWave Inc. and Nebius Group NV. Nvidia has announced AI-related deals worth more than $540 billion this year alone, excluding the proposed OpenAI agreement.
Huang has rejected claims that Nvidia’s financing model creates circular demand. “This is a small percentage of the amount of money they ultimately have to raise,” Huang said in January, discussing Nvidia’s investment in CoreWave. “The idea that this is circular — that’s ridiculous,” Huang said, according to Bloomberg.