AI investment is a high-stakes loop, but investors and industry watchers in this space remain largely unfazed — for now — about whether and when it might break.
OpenAI recently closed its latest funding round to the tune of $122 billion, with usual suspects Amazon, Nvidia, Microsoft and SoftBank continuing their backing.
Nvidia is not only a backer of OpenAI; it also sells the AI company chips needed to advance its technology. Such arrangements have led to criticism that the AI sector is something of a money pit, sustained by investor capital while companies are still trying to figure out profitability. That might be typical for startups, but the expectations on AI’s shoulders mean failure could have widespread repercussions.
For CIOs, the question is whether this self-reinforcing funding cycle — where investors back companies that in turn become customers — can sustain the vendor ecosystem and pricing models enterprises are relying on to push their AI initiatives from pilots to production.
Further, public pushback against the buildup of large data centers intended to support AI raises questions about the continued cost and progress of the technology. Municipalities in Tennessee, Missouri, Indiana, New Jersey and other states saw residents contest plans to build or expand data centers in their communities. Maine recently advanced legislation for a temporary moratorium on large data center construction across the entire state. That bill has yet to be signed into law.
The funding cycle that fuels AI reminds Craig Everett, assistant professor of finance at Pepperdine Graziadio Business School, of the fiber-optic buildout in the 1990s. At the time, the telecom industry was “going gangbusters” to connect the world with fiber-optic cables, he said — so much so that they overbuilt.
“They weren’t doing equity investments in each other; they were doing what are called capacity swaps, which was really kind of dishonorable,” Everett said. He is also director of the Pepperdine Private Capital Markets Project.
Everett said some telecom companies were doing in-kind purchases of each other’s capacity. For the companies involved, there was a net-zero effect on actual expenses, but on the books, both companies’ revenue would be boosted because the in-kind deal was recorded as revenue. “That was kind of shady,” he said.
Keeping the money honest
The current dealmaking and funding of AI may also raise eyebrows, but Everett said the way it is being handled appears to be above board. “Definitely, it’s a funding merry-go-round … You’re investing in a company that then buys your product. That will tend to have an upward spiral effect, of course, until the music stops,” he said.
Despite surface-level appearances, he said these seem to be legitimate investments. “The fact that they’re also a customer is a nice side effect.”
AI is often framed as a tool CIOs can deploy for efficiency or internal creativity, but not every idea spawned by AI companies has legs. Even well-funded bets can falter: OpenAI will shut down its Sora generative video app later this month, with the API to follow in September, underscoring how quickly expensive AI initiatives can be reevaluated. With Sora’s demise, so too went a $1 billion licensing deal with Disney. The cost of running Sora, along with copyright challenges, seems to have outweighed its near-term returns.
And while the pursuit of military contracts could be a revenue source for AI players, such relationships have been dicey. Anthropic’s insistence on guardrails for its AI, if used by the military, ran afoul of the Department of Defense, which banned the company from its contracts. OpenAI has also sought to refine its defense contract to prevent its tech from being used for surveillance and other purposes.
Is there a revenue flow?
Does that leave AI to survive largely on its funding, rather than real revenue? Daniel Docter, managing director at Dell Technologies Capital , said similar questions surfaced in earlier tech cycles, including telecom in the early 2000s. He cited the revelations of fraud at Enron and WorldCom, which both imploded in bankruptcy. “Isn’t the money going here just to turn it around and buy equipment and fiber and put it back here? Hey, something’s going on. Clearly, there was something going on,” he said.
What Docter sees as different this time is the underlying demand for AI, which he said has yet to show signs of letting up. “The important word is yet,” he said. “I haven’t seen anything yet.”
Docter said the multitude of companies in the AI sector are needed to do the heavy lifting to build the infrastructure — chips, computers, networking and data centers — with new capacity devoured as soon as it is online. “It is instantly consumed. It’s like, ‘It’s now ready. Raise your hand if you want it,'” he said.
Rethinking what it takes to fund innovation
The funding cycle for AI might be misunderstood or misdiagnosed, according to Steven Waterhouse, founder and general partner of Nazaré Ventures. He has been building in the technology and internet sectors since before web browsers. Recalling the rise of Yahoo and other dot-coms that went public, Waterhouse said there were questions about the money that went into those companies and the revenue they generated. “In any period of rapid expansion from a new technology, you will see some strange funding,” he said.
Deals such as Nvidia’s investment in OpenAI, or Microsoft putting money into Anthropic, may take up the spotlight, but there are other AI players and investors across a broader ecosystem that continues to grow, supported by what he said is real revenue. “We have now 16 companies in our portfolio globally, across Europe and the US. This isn’t just a Silicon Valley phenomenon that I’m talking about,” Waterhouse said.
In particular, he said he sees an acceleration from proofs of concept toward production revenue, with companies planning longer-term contracts, either in compute or applications and agentic workflows.
Despite that potential, the cost of building up AI capacity remains a tangible issue, said Greg Zorella, a principal analyst at Forrester. “There is constrained supply for things like data centers to support scaling AI use cases across enterprises,” he said.
Moreover, the cost of AI could rise in the near term as more enterprises shift from proofs of concept to scale by the middle or later this year. Limited supply naturally means enterprises may need to dig deeper into their pockets. “If the capacity to handle an exponential increase in AI deployments isn’t there, then somebody’s going to be paying more to deploy theirs,” Zorella said.
The other shoe that may drop
Companies might not have factored in the very complex economics around how much AI really costs them, he warned, especially as market dynamics may push prices up.
It remains to be seen how long investors are willing to burn money in the AI sector as costs continue to be significant for all parties involved. Even after end-user companies figure out what their cost models look like, they must also figure out what those costs might look like two to three years from now, Zorella said.
“How much does it cost me to turn on an agent, given that I’ve got cloud fees, I’ve got LLM fees, I’ve got all these other kinds of fees out there that I might not have thought of,” Zorella said.
