This post was originally published on this site.
The AI boom is sparking a boom in circular, creative financing and purchasing schemes, which are making some investors nervous.
The boom in AI stocks is routinely described as “staggering,” and with reason. Nvidia alone boasted a market cap, as of January 2, exceeding the GDP of every country in the world save the US and China.
But the collection of marquee names powering the AI industry is comparatively small, and the word frequently attached to their relationships—much to the annoyance of their CEOs—is “circular.” Investors and analysts are asking questions about the latticework of supply-chain commitments these highflying companies are weaving with each other, and the direct investments that bolster them.
As 2025 ended, the market was still waiting for Nvidia’s landmark deal with OpenAI, which called for the chipmaker to invest $100 billion in the AI developer over 10 years and for OpenAI to purchase 10 gigawatts (GW) of AI infrastructure, to be finalized. In November, Nvidia’s quarterly report stated that “there is no assurance that we will enter into definitive agreements” with its previously announced partner.
The news did not significantly dent Nvidia’s stock price, but it contributed to an already growing wave of skepticism about the funding and future of enterprises built around an expected surge in AI tool adoption. In December, Michael Burry, the hedge fund manager made famous by his successful 2008 bet against the US housing market, revealed that he is shorting Nvidia and another high-flying tech stock, Palantir. He cited specifically Nvidia’s strategy of supporting customers’ purchases of its products, which he compared to Enron’s practices before the dot com bubble collapsed in March of 2000.
Other, similarly circular deals were announced last year. Between them, Microsoft and Nvidia invested approximately $15 billion in the AI startup Anthropic, which will spend approximately $30 billion on Microsoft’s cloud services and Nvidia’s chips. Advanced Micro Devices (AMD) granted OpenAI warrants that allow it to buy AMD stock at a nominal price; AMD will power its operations with AMD chips.
In July, OpenAI and Oracle inked a $300 billion partnership to develop up to 4.5 GW of capacity for Stargate, OpenAI’s AI infrastructure platform. Two months later, Oracle signed a deal to supply $300 billion of compute capacity to OpenAI over the next five years.
CoreWeave, a nine-year-old company that provides cloud-based infrastructure to AI developers and enterprises, is 5%-owned by Nvidia, which sells chips to CoreWeave. OpenAI, which uses CoreWeave’s cloud service, owns an approximately $350 million stake in the company. Nvidia also has its hand in with Elon Musk’s xAI. The two companies are part of a consortium that bought Alligned Data Centers in the second half of 2025 for $40 billion, alongside which Nvidia struck a $20 billion lease-to-own deal with xAI for AI chips, partially funded through a special purpose vehicle in which Nvidia itself has an investment.
Some of the companies playing prominent roles in the evolving drama, like Nvidia—which reportedly made more than 50 deals last year—and Microsoft, have deep pockets and can afford to lose their investment if AI falls short of the hype, notes Andrew Odlyzko, emeritus professor of mathematics at the University of Minnesota, who studies financial bubbles. But others are giving investors the jitters. Oracle’s stock plunged 30% in the third quarter amid worries about its ability to deliver on its five-year commitment to OpenAI and money-losing OpenAI’s ability to pay for the compute capacity Oracle aims to supply.
“While the top four technology companies—Microsoft, Alphabet, Amazon, and Meta—generated $451 billion in operating cash flow in 2024, providing apparent financial stability, their interconnected AI investments mean trouble for one could rapidly cascade to others,” research analyst Gregory Blotnick said in a Columbia Graduate School of Business online forum last year. “If Microsoft’s AI monetization disappoints, it could reduce Azure spending, impacting Nvidia’s revenue, which would affect CoreWeave’s valuation, ultimately circling back to OpenAI’s funding capacity.”
This Time, It’s Different?
Such concerns are understandable, given that AI accounted for roughly half of all US venture capital commitments in the first seven months of 2025, according to PitchBook, while a McKinsey & Co. report published last April estimated that meeting worldwide demand for AI will require $5.2 trillion to be invested in data centers by 2030.
In an October 2025 blogpost, Mihir Kshirsagar, director of Princeton University’s Technology Policy Clinic, warned of an “accounting mismatch” regarding the critical element in the chain of codependency: “The chips at the heart of the infrastructure buildout have a useful lifespan of one to three years due to rapid technological obsolescence and physical wear, but companies depreciate them over five to six years. In other words, they spread out the cost of their massive capital investments over a longer period than the facts warrant.” Adding to the perplexity, most tech companies combine the costs of their chips and data centers with their other construction projections, making depreciation difficult to estimate.
With any new technology or business, observes Bryan Routledge, associate professor of finance at Carnegie Mellon University’s Tepper School of Business, there are two key questions: Is the investment worthwhile? And if so, how do you fund it? In the case of AI, he argues, the second may be less important than the first.
Compared to the late 1990s, when the dot com and telecom stocks were soaring, capital is much more abundant today, he points out: “The venture capital market and private equity are much bigger now. There are a bunch of places [AI companies] can go to raise money.” And the suppliers who furnish them with chips and data center capacity may be better places to go than most, “since it’s easier to explain the technology to them and the deals are not just financial, but about physical operations.”
That said, the value of AI as an investment and the question of how to fund it are closely related. “What’s extra-complicated,” says Routledge, “is where all the pieces in the supply chain fall in.” Chips, data centers, energy supply, and other components are all being built out at the same time, and all must be ready and available when they are needed. Complicating the timing problem, regulatory oversight of certain aspects of the AI supply chain is complex and evolving.
So is AI itself, and this poses a further issue, critics say. Chip technology is developing rapidly and could move in quite different directions in the coming years. So could the development of AI programming. Could the purchase-and-investment deals companies make today leave them with too little flexibility tomorrow? OpenAI’s 10-year deal with Nvidia comes to mind.
“This is definitely a risk all across the AI ecosystem,” says Blotnick. As industry titans shift their strategies to adapt to new developments, the competitive landscape could become increasingly detrimental for smaller players. “The titans have the brainpower and the capital to insource all of this themselves. So for a second- or third-tier player who is relying on Microsoft or Google as a customer, they could absolutely get cut off and have serious issues.”
Even for the titans, the risks are not negligible, the University of Minnesota’s Odlyzko warns: “The big danger is if people wake up one morning and say that large language models are not the holy grail, and growth is slower than expected. Nvidia, I’m sure, will survive as a company, but their valuation may drop severely as their order book collapses.”
That, he notes, is what happened to some high-flying telecom companies in the early 2000s; if it happens again, some of the circular funding deals spilling out of the AI space now could aggravate the adjustment. “Creative finance then becomes a danger; we saw it again in 2008.”
Other observers, however, counsel a step back and a refocus on the question of whether AI is a fundamentally worthwhile investment: that is, whether it will ultimately be profitable.
“That’s the question on everyone’s mind,” says Blotnick. “No one knows the return on capex, and no one knows the profitability of this entire arms race. The titans—Google, Microsoft, Meta, etc.—clearly view this as winner-take-all, get-the-market-share-first-and-worry-about-profits-later.” Which is how much of the tech industry has long operated.
As for the risks AI companies face in their circular financing deals—particularly the long-term infrastructure and equipment purchases included in these—Routledge notes that the risks could well run in the opposite direction. While the deals are complex, they are largely being made in the open: “If you buy Microsoft, you know you’re buying a little bit of OpenAI as part of it. The risk is not trivial, but if you’re in this industry and it’s causing you stress, this is what you signed up for.”