Will risky data center buildouts slow the growth of AI?

AI is big business, and so is development of the digital infrastructure needed to power it. Goldman Sachs estimates that the AI buildout will require $7.6 trillion of capex between 2026 and 2031 across compute, data centers, and power. Data centers alone will require $2.15 trillion of that spend.

There’s uncertainty, though, about whether data center developers will be able to handle the surge in demand for these facilities used to train and operate AI models.

Where’s the risk? There’s no shortage of challenges awaiting a data center project. Experts cited power generation, materials backlogs, and growing community opposition as some of the big ones.

To say the stakes are high in building out data center networks is putting it too lightly for experts like Mike Mathews, global digital infrastructure practice leader at Marsh, who told CFO Brew, “AI isn’t a new technology as much as it is a new utility,” like electricity. The digital infrastructure serves more than AI, he added. Everything from tap-to-pay technology to military drones relies on networks of data centers and fiber-optic lines.

Power “is the bottleneck in sort of getting these projects online,” according to Olivia Wang, research analyst at market intelligence firm Sightline Climate. Wang told us back in March that her firm observed median wait times of about five years to get sites powered, with that queue stretching as long as seven years for high-demand areas like the state of Virginia. Developers may also find themselves waiting a long time—as long as two years—for certain electrical equipment like transformers, she said.

Bitcoin company pivots. CleanSpark, founded in 2014, was a bitcoin mining company until it made a strategic expansion into AI data centers shortly after company co-founder Matt Schultz took over as CEO in August 2025, a role he held previously.

Gary Vecchiarelli, president and CFO of CleanSpark, said his company is in serious talks with a hyperscaler to build a data center at a site the company owns in Sandersville, Georgia. CleanSpark currently uses the site for bitcoin mining—the hardware intensive process of adding transactions to the digital ledger and introducing new bitcoins into circulation. It also recently announced the acquisition of a 271-acre site with 285 megawatts of power near Houston.

CleanSpark is trying to land its first lease and publicly disclosed that it’s in “deep negotiations with a hyperscaler tenant,” Vecchiarelli said.

“We just need to make sure that that first deal is foremost a good deal, and second of all, that we’re able to execute on that deal, because these AI data center leases can be rather punitive, so it’s more than just doing financial modeling,” he said. “It’s making sure that you have the people, process, and system behind that to make sure that you’re able to fully execute and see that contract through.”

Risk management advantage. Good risk management of these projects has its financial benefits.

Developers with “strong risk and capital management tend to be more likely to secure better financing terms and lower borrowing costs,” Marsh experts noted in a recent report on digital infrastructure risks. Insurance is “a key enabler” for projects, and financial tools including surety guarantees “can free up cash for growth rather than locking it in collateral.”

Overall, CleanSpark has a leg up because of its bitcoin mining operations, according to Vecchiarelli. For one, its existing operations are on sites with readily available power—no need to wait for new infrastructure before it can turn the lights on.

He added that his company’s proposals look attractive to utilities because they can guarantee constant use of the full power they’re allotted, making them a more certain revenue stream. Whatever power isn’t being used at the data center can go toward bitcoin mining.

“We’re in an environment now where companies that have access to power basically have keys to the castle,” Vecchiarelli said. “We’re seeing that not just the lease rates or the revenue that lessors or developers, such as us, are getting for their power assets, but we’re also getting really good financing rates.”

Other roadblocks. What about the communities hesitant to welcome a data center? Wang said transparency and communication are differentiators. She said that residents don’t like the secrecy shrouding some of the proposals, which are often developed under an anonymous SPV or LLC, with the end-user a mystery until much later in the process.

“The community, they don’t want to be blindsided,” she said.

Growing public concern about data centers and their environmental impacts—in particular, the vast amounts of water and electricity they require—are a substantial risk to developers. A number of US states have introduced legislation or have filed to restrict or ban data center developments. The New York State legislature is currently weighing a one-year moratorium on large-scale data centers, according to Politico.

The uncertainty around data center projects even extends to the developers. “There’s a lot of speculative announcements being made,” Wang explained. “A lot of developers will announce a lot of projects and see which ones land at the end of the day. It’s hard for anyone to make sense of which one’s signal, [and] which one’s actually going to get built.”