Data centers circumvent multi-year electrical grid queues by deploying on-site generation—nuclear, SMR, and natural gas.
Grid interconnection delays are pushing data centers toward behind-the-meter (BTM) power generation at an accelerating pace. A new forecast from Enverus Intelligence Research projects that BTM will serve 25.5 GW of demand from new U.S. industrial facilities through 2030, with data centers accounting for 22.5 GW—or 88%—of that total. This represents about 36% of all U.S. data center capacity additions expected during the five-year period.
"We built the forecast by heavily risking both whether new load comes online and how much goes BTM," said Alex Nevokshonoff, senior analyst at Enverus Intelligence Research. "We are confident the numbers are realistic, and if anything, likely to rise as more data centers are pushed BTM. About 60% of the expected BTM load is tied to projects we view as high-confidence to come online within the forecast timeline."
Most operators prefer grid connections for simplicity and reliability, but delays in energization are reshaping project economics. "All 22.5 GW comes from new data centers targeting operation by 2030, and the shift to BTM is driven predominantly by grid interconnection delays," Nevokshonoff said. "Developers are turning to BTM power to come online faster." The trend is particularly pronounced in grid-constrained regions. PJM is expected to account for 7.5 GW of BTM demand, while ERCOT is projected at 5 GW—together representing nearly half the forecast.
Meeting this BTM demand will require 31.6 GW of new generation capacity, including 29.6 GW of natural gas-fired generation. "About 62% (19.5 GW) of the 31.6 GW is tied to high-confidence loads expected online by 2030," Nevokshonoff said. "The remainder is allocated using top-down analysis, such as data-center capex spend and industrial trends. Some high-confidence projects have ordered more generation than we assume comes online by 2030, so we risk generation to the load we expect to materialize, based on factors like contracted compute, rather than taking nameplate load or generation contracts at face value."
The race for power is reshaping what gets built. Developers are favoring faster-deploying technologies over large-frame turbines. Reciprocating engines, small- and medium-frame gas turbines, and fuel cells are expected to account for 61% of projected gas-fired BTM capacity—a choice driven by speed. These technologies can reach commercial operation in roughly 18 to 24 months, compared with as long as 80 months for a large-frame combined-cycle plant.
"Large-frame gas turbines have traditionally delivered reasonably priced baseload power with low heat rates, which cuts operating costs," Nevokshonoff said. "But demand has sold out many major manufacturers into 2030 and pushed prices up, so developers who need power quickly are looking elsewhere. Data centers that ordered large-frame units a year or two ago will receive them and energize on time. Anyone who did not order in 2024 or early 2025 has been pushed toward smaller, modular options such as aeroderivatives and reciprocating engines, which typically deliver in about 24 months."
Modular technologies also require less excess capacity to maintain reliability. Enverus estimates reciprocating engines and fuel cells can operate with generation overbuild factors as low as 1.1 times demand, compared with about 1.5 times for large-frame plants. Redundancy favors smaller units because many BTM data centers plan to overbuild primary generation for reliability. Modular fleets typically need only 10-20% overbuild, while 100+ MW large-frame units can require as much as 50%.
The economics are shifting decisively. "Initially, reciprocating engines and small turbines won on availability, since they could be delivered sooner despite costing more per kW than a CCGT plant," Nevokshonoff said. "Now economics are shifting in their favor too." Demand has pushed large-frame prices above reciprocating engines for a new order placed today. As grid interconnection timelines grow longer, technologies deployable within the next two years have a significant advantage.