Data center developers and operators are finding power harder to come by than ever. Amid these challenges, some data centers are rethinking their approach by adopting behind-the-meter configurations, where power generation and consumption occur on the same site.
According to Boston Consulting Group, the data center power shortage could amount to more than 45 GW. More than 12,000 active projects are currently seeking grid interconnection, representing 1,570 GW of generator capacity and 1,030 GW of storage. Due to AI, data center power consumption will grow from 2.5% of U.S. electricity to 7.5% over the next five years.
During his presentation at this year’s Data Center World event, AFCOM program chair Bill Kleyman laid out the problems facing anyone in the industry looking for power.
“The number one constraint for AI data centers is power,” said Kleyman. Citing the AFCOM 2025 State of the Data Center report, he said 62% of data centers are exploring on-site power generation to boost energy efficiency or resilience. Nearly one-fifth (19%) of those surveyed said they were already implementing some form of behind-the-meter power by the end of 2024.
A behind-the-meter data center approach involves building renewable energy assets directly alongside new data centers. The on-site power generation can help bypass grid congestion, avoid transmission losses, mitigate environmental impacts, accelerate speed-to-market, and improve facility reliability by reducing vulnerability to outages.
Renewable energy interest and adoption remain high among data centers, Kleyman added. But its low availability – about one quarter of actual capacity – means AI factories cannot rely on it. Instead, they are turning to natural gas generation and small modular reactors (SMRs) as future solutions that provide the quantity and reliability levels data centers require.
The number one constraint for AI data centers is power.
AI Power Demands Force New Ideas
Industry analysts expect AI workloads to surge over the coming years. The accompanying rise in power demand presents a present-day challenge that requires immediate solutions.
“AI will drive over 50% of global data center capacity and more than 70% of revenue opportunity as AI fuels massive productivity gains across all industries,” said Vlad Galabov, research director for Omdia’s cloud and data center practice. “More than 35 GW of data center power will be self-generated by 2030.”
Those rushing headlong into the building of so-called AI factories don’t have the patience to wait for traditional utilities to provide the power they need. It can take half a decade or more for utilities to provision more capacity. AI factory developers want power now – and that means behind-the-meter and Bring Your Own Power (BYOP).
Galabov anticipates more investment and partnerships between hyperscalers, colocation providers, and developers of AI-ready facilities with those representing all parts of the power network.
According to Omdia, annual data center capital expenditure investment will reach $1 trillion globally by 2030 with physical infrastructure for power being the greatest beneficiary.
“As compute and rack densities climb, the volume of investment in physical infrastructure is rapidly accelerating,” said Galabov.
Natural Gas has Favorable Economics
Many next-generation data center developers are willing to venture far afield to find the power they need. Some are heading to Canada, while others are finding plenty of energy in North Dakota, West Virginia, or West Texas.
Traditional metro hot spots are being ignored due to the lack of power. Some developers are finding sites beside large wind and solar farms, but the bulk of newly announced mega-data centers prefer a site beside a natural gas basin.
There are huge deposits of natural gas throughout the U.S., the technology to extract it is mature, the price is low, and the nation has a vast pipeline network to take it to market. Over the next few years, these factors will see many more large data centers gravitate to areas with an abundance of natural gas.
“Between 2024 and 2030, natural gas consumption by data centers will rise by almost three times,” said Dave Bell, vice president of data center and microgrid development at VoltaGrid. “By then, data centers will be using up 4.5% of U.S. gas consumption for electricity generation.”
Bell advised data centers to find locations close to existing natural gas basics and intrastate and interstate pipelines. This includes West Virginia, Pennsylvania, Ohio, Virginia, Texas, and Oklahoma.
Time to Power
The race to deliver the latest AI services continues at a rapid pace. There is competition for the latest high-performance chips, and long lead times for electrical equipment like transformers. Some gas turbines have waiting lists of three years or more. Nuclear energy from SMRs might take five years to develop, permit, and open – perhaps longer.
“Time to power wins,” said Laura Laltrello, COO at Applied Digital. “The AI race favors builders who can energize fast – compute demand isn’t going to slow down.”
Her company favors North Dakota. That’s where Applied Digital has found stranded power that it plans to use to run its AI factory. It will come from a large wind farm and should be able to deliver more than 400 MW.
“The biggest bottleneck is power,” said Laltrello. “Major metros are tapped out, grid-constrained, and slow to scale.”
