Nvidia is discussing ways to support financing for OpenAI’s massive artificial intelligence data centre project, fuelling warnings about a circular funding structure in the AI investment ecosystem.
CNBC reported on July 27 that Jim Cramer (짐 크레이머) said the way the recent AI boom is unfolding brings back memories of the dotcom bubble’s overheating. "I lived through 2000. I don't want the sequel," he said.
The issue is that Nvidia is discussing a $250 billion guarantee to back OpenAI’s plan for a 10-gigawatt AI data centre campus in Ohio. The guarantee is known to be structured to support leasing and construction debt for the project, not purchases of Nvidia chips to be installed in the data centre. Nvidia declined to comment, and Nvidia shares fell more than 4 percent on the day, dragging down semiconductor stocks broadly.
The talks show how large the structure has become in which suppliers and buyers in the AI industry are intertwined again through funding. Nvidia has invested in multiple companies that are also major chip customers. It invested $30 billion in OpenAI in March and $10 billion in Anthropic last year. It also provided funding to so-called new cloud companies that lease Nvidia-based computing resources. Nvidia has said such investments help the AI ecosystem grow and support long-term profits.
Cramer said the structure resembles the late 1990s, when telecommunications equipment makers provided funding to help customers make large purchases. It helped expand revenue at the time, he said, but both suppliers and investors suffered large losses when cash-strapped customers could not afford payments.
Cramer’s core concern is a structure in which a supplier takes on the funding lifeline of customers that buy its products. "What we learned in 2000 is you shouldn’t lend money to a company that buys your merchandise," he said. While acknowledging Nvidia is a strong company, he said market confidence can break quickly in a structure where customers’ large spending relies on raising outside capital.
OpenAI’s funding situation is cited as a key variable in the talks. "In this case, if the buyer, OpenAI, can actually afford to pay for these chips, for example if it succeeds with an IPO, Nvidia is in a very good position," Cramer said. "If the buyer can’t pay, it’s a different story," he said. OpenAI filed confidentially for an initial public offering in June, but has not yet disclosed timing for a listing. OpenAI was valued at more than $800 billion by private investors in March.
The concern is not limited to Nvidia alone. As more companies in AI infrastructure investment are counting on performance, there is a view that the ripple effects could be broad if the flow of data centre investment cools. "Right now too many companies are betting their performance on data centres," Cramer said. "If the market no longer wants to fund data centres, and the companies themselves don’t have money or can’t collect payments, we will go back to 2000 again," he said.
In Cramer’s view, Nvidia’s ample financial capacity is not enough to offset the risk. "Nvidia should not make this kind of guarantee no matter how much money there is in the world. The reason is history, period," he said.
The talks show that the race to expand AI infrastructure is growing beyond a simple phase of facility investment into a structure in which supplier funding support and customers’ capital raising are intertwined. Market attention is expected to focus on whether Nvidia’s guarantee talks are actually concluded, whether OpenAI secures a stable funding channel through an IPO and other means, and whether the enthusiasm for AI data centre investment can continue.