Power and energy supply were key issues at this year’s Data Center World event in Washington, D.C. Industry experts emphasized that power availability has become the primary factor driving data center site selection decisions, often outweighing traditional considerations like proximity to urban centers.
“New data centers need to have all their ducks in a row,” said Colby Cox, managing director Americas of analyst firm DC Byte. “That means power availability as well as land.”
According to DC Byte numbers, a boom in construction is ongoing. Colocation providers and hyperscalers added 122 GW of new capacity in 2024. A good portion of that is in prime markets. But a growing percentage is in more remote and often cooler northerly places. Many of them have two factors in common: they are large-scale and involve the delivery of AI.
Generative AI workloads are typically looking for far more power than the average data center. They are being built on a grander scale. According to Cox, as many as 35 GW-scale projects were announced in 2024 compared to three in the previous year. Another 15 projects broke ground that were slightly below 1 GW in size, indicating the extent of the ongoing AI boom.
As few established metro areas have a spare gigawatt or two available, this new generation of so-called ‘AI factories’ are being situated in non-traditional data center locales like North Dakota, West Texas, and the province of Alberta in Canada.
Cox also listed Northwest Indiana, Pennsylvania, Kentucky, Kansas City metro, West Virginia, and several North Carolina metros as areas experiencing explosive growth to serve AI workloads. Many of these places can also offer fiber connectivity and the availability of inexpensive land as well as plenty of power.
Grid Vs. Onsite Power
Some of the new data center boom towns and boom states have plenty of power available on the grid, but many don’t. Instead, they have stranded power – energy that is available for generation but is geographically isolated. Examples include major natural gas deposits in West Texas and Alberta – and in many other parts of the U.S. – and large wind farms in North Dakota.
In North Dakota, Applied Digital is tapping into a huge wind farm that is far distant from load centers. And in Wonder Valley, Alberta, a massive natural gas basin is the source of power. The developer of what could be an 8 GW data center is buying 10 gas turbines to generate power onsite for the facility. The first 1.5 GW should be completed by 2027.
“Everything comes from the availability of power in abundance,” said Kevin O’Leary of Shark Tank fame, whose company is involved in the project.
Bring Your Own Power (BYOP) is becoming a necessity for developers. It is up to them to bring power to the party – especially when they want to open an AI factory.
“If you want to attract investors, you must be able to get a data center project from start to finish within 24 months,” said O’Leary. “The only way to do that for AI factories today is to go off-grid and build the power yourself.”
AI Workloads Impact Data Center Siting
“The location where high-density workloads need to be placed is changing dramatically,” said Phill Lawson-Shanks, chief innovation officer at Aligned Data Centers, during this year’s Data Center World.
Traditionally, Aligned engaged in land banking – the practice of purchasing land that might one day be the site of a new data center. With the land in place, it then arranges power from a nearby utility.
“These days, if you ask for 1 GW, you often hear that you can only get 28 MW in three years,” said Lawson-Shanks. “Understandably, utilities are worried about overextending themselves as they overbuilt during the dotcom boom and were left with assets they couldn’t sell when the bubble burst.”
One Aligned Data Centers site serves as an example of the level of creativity involved. In Australia, the company found an old mill that already had 1 GW of power available. That said, the company still prefers to be the provider of land rather than power.
Location, Location, Location (and Power)
Location has always been vital in data center placement. It affects many factors: the cost of construction, leasing rates, proximity to IT staff and customers, cost of power, tax incentives, geographic risk, and geopolitical risk factors.
There is immense competition for sites, construction materials, power, equipment, CPUs/GPUs, and cooling. But the rise of AI-based data centers has elevated power above most of the other factors.
“Massive natural gas projects are being talked about all over the country,” said Cox. “Those building new data centers should allow for extra space and capacity where they can serve the expanding AI market.”
