Companies are so desperate for data centers they're leasing them before they’re even built

Data center construction levels are at an all-time high. And more than ever, companies that need them have already called dibs.

In the first quarter of 2024, what amounts to about half of the existing supply of data center megawattage in the US is under construction, according to real estate services firm CBRE. And 84% of that is already leased. Typically that rate had been about 50% the last few years — already notably higher than other real estate classes.

“I’m astonished and impressed by the demand for facilities yet to be fully constructed,” CBRE Data Center Research Director Gordon Dolven told Sherwood.

That advanced interest means that despite the huge amount of construction, there’s still going to be a shortage of data centers to meet demand. In other words, data center vacancy rates are staying low and rents high.

Nationwide the vacancy rates are near record lows of 3.7% and average asking rent for data centers was up 19% year over year, according to CBRE. It was up 42% in Northern Virginia, where many data centers are located. These sorts of price jumps are “unprecedented” compared with other types of real estate. For comparison, rents for industrial and logistics real estate, another hot asset class used in e-commerce, is expected to go up 8% this year.

As such, data center REITs like Digital Realty, Equinix, and Iron Mountain have been performing really well.

The backlog of demand isn’t going away anytime soon. It takes a really long time to build data centers because it doesn’t just require building data centers. Data centers need updated power grids, fiber lines, and water to come online, and often those connections have to be built as well. And those are subject to supply chains, permitting, easements, and other roadblocks.

As Dolven put it, “Instead of a dirt road, they need 6-lane highways.”

Getting all that squared away adds an additional 1-3 years at least on top of 1-3 years it already takes to build the typical data center, Dolven said.

And our need for data centers and the infrastructure they rely on is only going up. Not only are the applications we use daily collecting, generating and parsing tons tons of data, but the rise of generative AI is going to turbo charge these needs even more.

As my colleague Matt Phillips pointed out, data centers were already sucking up vast amounts of power before the AI boom. Indeed, data center electricity demands are running up against power companies’ ability to generate it.