Nvidia’s OpenAI Backstop Plan Revives Concerns Over ‘AI Circular Financing’

“Guarantee for $250 billion of data-center lease payments, plus $350 billion in financing for GPU purchases”

SK project valued at $500 billion also draws scrutiny over ‘AI circular financing’

Nvidia Corp.’s talks to provide massive financial guarantees to OpenAI are reigniting concerns in the market over so-called AI circular financing. Anxiety has grown after reports that Nvidia is considering a $250 billion payment guarantee tied to OpenAI’s data-center computing use, even though the company remains deeply unprofitable.

AI skeptics have argued for years that Nvidia’s investments in AI companies such as OpenAI and Anthropic, as well as data-center operators including CoreWeave and Nebius, artificially inflate industry demand and company valuations.

The Wall Street Journal and Bloomberg reported on July 27 that Nvidia has discussed guaranteeing as much as $250 billion in lease payments so OpenAI can use a 10-gigawatt data center SoftBank is building in Ohio. Nvidia is also considering providing $350 billion in financing for OpenAI to purchase Nvidia GPUs.

People familiar with the matter said the talks are at an early stage and could collapse, or the financing terms could change.

“The fact that Nvidia is providing additional guarantees on OpenAI’s data-center debt is a signal of demand for AI buildout, but it also serves as a reminder of the funding strain,” said Billy Leung, an investment strategist at Global X Management.

Funding from Nvidia could help ease creditors’ concerns about lending to an unprofitable company to support OpenAI’s relentless need for computing power. But it could also increase the burden on Nvidia.

Much of SoftBank Group Corp.’s upside depends on a higher OpenAI valuation and the prospect of a large initial public offering in the future. SoftBank has agreed to invest about $65 billion in OpenAI by October. It also signed a $40 billion bridge loan to fund that investment, one of the largest such financings in Asia-Pacific. But the more SoftBank expands its exposure to OpenAI, where it has limited management control, the more uneasy its investors may become.

Investors are particularly concerned about a structure in which Nvidia lends money to OpenAI and others, and that money is then used to buy Nvidia chips. For Nvidia, loans, structured financing and equity investments can all translate into higher GPU demand. Investors say that if AI companies are buying GPUs with Nvidia-backed funding rather than their own cash, demand may be overstated.

Critics including Michael Burry have argued that the arrangement is circular because companies Nvidia finances or backs with equity typically go on to buy or use Nvidia chips. They say such deals could magnify losses if demand for artificial intelligence falls short of lofty expectations.

Nvidia also announced a $500 billion AI project with SK Group last Friday.

The company said it would work with SK Group, the parent of SK Hynix Inc., to build more than 2 gigawatts of AI data centers in South Korea. Chief Executive Officer Jensen Huang said the $500 billion figure includes both Nvidia’s purchases of high-bandwidth memory from SK Hynix and SK Group’s purchases of Nvidia GPUs and supercomputers. In other words, the figure reflects the value of long-term transactions rather than a direct cash investment.

Nvidia also agreed to invest $1 billion in Naver, which is building AI data centers. Those facilities will also use Nvidia’s AI computing hardware.

“This is Korea’s golden age,” Huang said in an interview with Bloomberg TV. Korea has the potential to help the world build AI infrastructure, he added.

Nvidia has been active in acquisitions and investments across the broader ecosystem in recent years. It has taken stakes not only in developers such as OpenAI, but also in companies including Marvell Technology Inc.

Huang has pushed back against claims that deals with major buyers of Nvidia chips are inherently circular. Referring to the company’s investment in CoreWeave in January, he said it was “ultimately a very small part of the amount they had to raise,” dismissing the criticism.

Nvidia is not alone in offering financing or payment guarantees to prospective customers.

Alphabet Inc.’s Google agreed to guarantee lease payments for five Anthropic data centers. That allowed Anthropic to secure about $35 billion in loans.

Such deals have tightly bound many AI companies to one another, raising concerns that the industry has become vulnerable to a systemic shock.

One of the main concerns is the industry’s rising debt load. Since last year, many AI companies have increased borrowing to finance data-center and chip projects. OpenAI, SoftBank, CoreWeave, Nebius and Oracle Corp. all use significant leverage. Even Google, Amazon.com Inc. and Meta Platforms Inc., which still generate strong cash flow, have issued large amounts of debt this year and reached unprecedented borrowing levels.

Kim Jung-a, Guest Reporter at Hankyung.com, kja@hankyung.com