US data center used an estimated 312.6 TWh of electricity in 2025, a 25.5% year-over-year increase – nearly eight times the growth rate of overall US electricity generation. Total US electricity generation rose 3.2% to 4,772 TWh, according to the Energy Institute’s 2026 Statistical Review of World Energy..
The U.S. accounted for 39.7% of the 787.8 TWh consumed by data centers worldwide. Global data center demand rose 19.7% in 2025 and had nearly doubled from 410.8 TWh in 2020.. The widening gap is making power the first filter in site selection.
Proximity to electrical infrastructure no longer guarantees access to hundreds of megawatts. Under these conditions, operators need to know what can be delivered, when it will arrive, and whether the supply can scale with the campus.. “The key shift is from asking, ‘Where do we want to build?’ to ‘Where can we actually secure and deliver the power?’” said Siddharth Muzumdar, senior vice president of research at DC Byte, in an interview with Data Center Knowledge..
Related:For High-Density AI, Available Data Center Space May Not Be Usable. Power Now Sets the Search Area. Historically, site searches started with a preferred market and then evaluated land, fiber, water, and utility service.
In constrained hubs such as Northern Virginia, Dallas, and Phoenix, that sequence has flipped. Developers are prioritizing where larger power blocks may be available.. Being near generation is not enough.
Transmission capacity and utility construction schedules must support the load. Emerging markets often lack the dense network infrastructure found in established hubs, adding another layer of complexity.. “It could take years to construct a new AI data center, and just as long to build the fiber networks needed to connect it,” said Bill Major, CEO of FiberLight.“Planning for both needs to happen in parallel.”.
Operators should engage utilities and network providers before committing to land. They must verify whether quoted capacity is firm and funded, or merely indicative of what could be available after upgrades that have yet to be scoped or financed.. Operating Costs Extend Beyond the Utility Rate.
Securing electricity for an initial building is insufficient when a 100 MW deployment can quickly evolve into a multi-phase campus requiring several hundred megawatts. Forecasts should account for future phases, workload growth, rack density, and changes in AI infrastructure.. “A project that initially requires 100 MW could ultimately require several hundred megawatts as additional phases come online,” Muzumdar noted..
Plans should map each phase to utility delivery dates and stress-test 12-, 24-, and 36-month delay scenarios. A low nominal power price is meaningless if transmission work or equipment shortages delay the delivery of revenue-generating capacity.. Related:Solid-State Transformers Power Next-Gen AI Data Centers.
“There’s more of a focus on time to revenue than cost,” said Sean Farney, vice president of data center strategy at JLL. “What’s not as easy to fix is time-to-revenue pressure or constraints.”. “There’s more of a focus on time to revenue than cost.” – Sean Farney, vice president of data center strategy at JLL.
Small differences in electricity prices compound over time at continuously operating facilities. Operators should calculate the all-in delivered cost, including interconnection work, transmission charges, demand fees, site infrastructure, backup generation, and rate escalation, and then weigh it against reliability and expansion headroom..
“The most attractive market is not necessarily the one with the cheapest electricity,” Muzumdar said. “It is the one that can provide competitively priced power reliably and at the scale the operator needs.”. Behind-the-Meter Power Goes Mainstream.
Data centers require continuous power, and many developers cannot wait years for grid upgrades. As a result, some are adopting bring-your-own-power strategies on the customer side of the meter.. Related:AI Rack Density’s Real Limits: Power, Cooling, Failure Risk.
Natural gas turbines, reciprocating engines, and generators currently offer firm, dispatchable output. Renewables and battery storage can diversify that supply, but each project must account for intermittency, fuel and pipeline access, air permits, emissions, capital costs, and maintenance..
“For large-scale development, [behind-the-meter power] is here,” Farney said. “It has gone from resistance three or four years ago to being openly accepted right now.”. Behind-the-meter generation can bridge the gap until utility capacity arrives or can become a permanent component of a hybrid power architecture.
Operators must decide who will own and operate the assets, how much redundancy is required, and whether the system meets uptime requirements and sustainability commitments.. Projects relying entirely on conventional grid interconnections face mounting schedule risk.. “If their only plan is to interconnect with the grid, their data center may never break ground,” Major said..
Operators Need Proof, Not Power Promises. Power constraints are reshaping the expectations of lenders, investors, and tenants. General assurances from a utility are increasingly insufficient.
Stakeholders want firm delivery dates backed by demonstrated operating expertise and clear allocation of upgrade costs.. “These deals are not getting financed without firm, real power-delivery dates backed by evidence of infrastructure investment and know-how,” Farney said.. Before selecting a site, operators should determine how much power is available today and what can be delivered over the next three to five years.
They need visibility into required substations and transmission projects, their completion dates, and who will pay for them.. They should also know what share of their supply can come from renewables and how the plan performs during outages or construction delays. The goal is to prove that the entire power strategy can support the construction schedule, eventual scale, and operating economics..
“Power can no longer be an assumption in the site-selection process,” Major said. “It has to be a fundamental part of the plan from Day One.”
