While bullish sentiment naturally follows surging digitalization efforts and the race among tech companies to win dominance in AI, the frenzy may belie some of the chokepoints in this market, including limits on power and land, consultants, analysts and digital infrastructure investors tell me.

To start, data centers need massive amounts of electricity to support their round-the-clock operations and to maintain a stable ecosystem for servers, network equipment and other gear. The facilities designed to train and run AI models have a more voracious demand for power than those supporting traditional cloud software providers—the largest can require more than 100MW of power, or enough to power tens of thousands of homes.

To fulfill that demand, hyperscale tech companies and developers plan to build data centers that require a gigawatt of power. In September, data center-as-a-service company ECL announced its plan to build an AI data center supporting a gigawatt of power capacity east of Houston.

Who’s got the power?

Satisfying AI’s hunger for power will require the industry to rapidly expand its access to energy resources. By 2030, data centers could consume up to 12% of total US power, whereas they account for 3% to 4% of total US power demand today, McKinsey & Company noted in a September report.

But some established markets are already jammed. In hot spots like Northern Virginia, it can take more than three years to connect new data centers to the power grid, according to the report.

Developers are increasingly drawn to alternative markets such as Dallas and other cities in the region known as the Texas Triangle, or Atlanta, due to their availability of space, power and tax incentives, said Patrick Lynch, global head of data center solutions at CBRE, and the firm’s head of Americas data center research, Gordon Dolven.

However, access is not always readily given. Local governments weigh data centers against other development projects—for instance, manufacturing plants that might provide more jobs—and don’t always prioritize the construction of data centers.

Additionally, despite the economic benefits touted by developers and tech companies, construction projects have encountered fierce opposition from residents in places including Fort Worth, Texas, Burns Harbor, Indiana, and Fayette County, Georgia, who are worried about noise pollution and strains on power and water resources, according to an October report from the Washington Post

On another note, the push to meet rising electricity needs could also create challenges to decarbonize the power grid. Goldman Sachs estimates that the carbon dioxide emissions produced by data centers may more than double from 2022 to 2030.

Supply chain stumper

Besides power and land resources, there is another complication to consider: constraints on the supply chain.

The data center market currently faces a massive equipment shortage, as rising demand for new data centers has stressed the supply chain. Orders for many devices, from generators to electrical equipment, may take years to fulfill.

“If you call the manufacturers, it’s quite likely they’ll tell you that we’ll put you in the queue, and that queue is going to be 36 months or more out,” Lynch said.

“It is a full-cycle challenge for the industry,” he added.

President-elect Donald Trump’s plan to levy steep tariffs may pose further challenges to the US market.

A fair portion of parts needed to build and operate data centers—such as switching gears, network gears and batteries—are produced offshore, said Jon Mauck, senior managing director and global head of DigitalBridge’s data center investment strategy. Raising tariffs could increase input costs and delay the delivery of essential components to US developers, others told me.

Expected demand shifts

Some cautioned that while the frantic surge in demand for data centers is substantially pushing up lease rates, this phenomenon is unlikely to last forever.

The advancement of technology and increasing efficiency may eventually decrease the demand for data centers. And operators won’t be able to extend the leases at the rates and terms so favorable to landlords today, said Angelo Sabatelle, who manages Moody’s project finance team for the Americas.

“You can’t close your eyes to the fact that technology changes quickly,” he said. “Now, there’s all the rage over data centers. I don’t know what about 15 years from now. I’m sure there will be a more efficient way of dealing with data centers.”

https://pitchbook.com/news/articles/sticking-points-in-the-ai-boom