The Data Center Land Grab
Two years ago, the choke point in AI infrastructure was silicon: a company's ambitions were capped by how many GPUs it could get its hands on. That constraint hasn't disappeared, but it has been overtaken by a bigger one. Across Virginia, Texas, Ohio and beyond, the fight is no longer about who can buy the most chips it is about who can secure the electrons to run them. Utilities, grid operators and government agencies are now describing power availability, not compute, as the ceiling on how fast the AI build-out can go.
From Chip Shortage to Grid Shortage
The scale of the shift shows up in the numbers utilities are now reporting. According to the International Energy Agency's 2026 analysis, global electricity demand from data centers grew 17% in 2025, in line with earlier IEA projections, while electricity consumption from AI-focused facilities specifically surged 50% over the same period. That is not a one-year spike; the IEA's companion Energy and AI report projects global data center electricity consumption roughly doubling from 415 terawatt-hours (TWh) in 2024 to 945 TWh by 2030 — a volume that would rival the current electricity use of a mid-sized industrial economy.
Independent estimates from technology research firm Gartner point in the same direction: worldwide data center electricity demand is projected to climb from 448 TWh in 2025 to 980 TWh by 2030, with electricity drawn specifically by AI-optimized servers rising almost fivefold, from 93 TWh to 432 TWh, over that period. In the United States, the Energy Information Administration's Annual Energy Outlook 2026 projects that electricity consumed by data center servers alone could reach between 446 and 818 billion kilowatt-hours by 2050, depending on how fast power draw per server climbs.
Grid Queues Have Become the New Title Search
If chips were rationed by allocation lists, power is now rationed by queue position. Grid interconnection backlogs tracked across U.S. regional grid operators reached roughly 2,600 gigawatts (GW) of proposed generation and storage capacity awaiting approval as of early 2026 — more than double the country's entire existing operating capacity — pushing project wait times out to five years on average and as long as a decade in the most congested corridors. In Northern Virginia, still the world's densest data center market, developers report average waits of up to seven years simply to get a substation connection confirmed.
Texas illustrates how much of that backlog is now data-center-driven rather than generic industrial load. ERCOT, the state's grid operator, disclosed that its large-load interconnection process was handling more than 225 GW of requests, with a separate 410 GW large-load queue in which data centers account for an estimated 87% of demand — a concentration stark enough that ERCOT brought in outside consulting support specifically to redesign how that queue is managed.
Utilities Are Underwriting the Buildout
Regulated utilities are treating this as durable, rate-base-worthy demand rather than a passing cycle. Dominion Energy, whose Virginia service territory hosts the largest concentration of data centers on earth, lifted its five-year capital investment plan to $64.7 billion for 2026–2030, up from $50.1 billion previously, with the bulk of that increase tied directly to data center load growth and the transmission and generation needed to serve it. The company has also created a dedicated large-load rate class so that data center developers directly fund the infrastructure their facilities require, and it disclosed plans with Amazon to explore small modular reactors as a dedicated generation source.
Peer utilities are moving on a similar scale. NextEra Energy, which posted 2025 revenue of $27.4 billion against capital expenditure of $24.6 billion, has been expanding its generation development pipeline specifically to serve large-load and data center customers, reflecting a broader pattern of utilities treating power procurement, not permitting or construction, as the pacing item for new capacity.
Why Consultants Call This Structural, Not Cyclical
Management consultants tracking the sector describe the shift in physical, not financial, terms. McKinsey's analysis of the AI infrastructure buildout forecasts a 3.5-times increase in data center capacity demand between 2025 and 2030, driven less by facility count than by density: rack power draw for modern AI workloads now runs 50 to 100 kilowatts, with specialized training configurations exceeding 120 kilowatts, versus 5 to 10 kilowatts for a traditional enterprise rack. That density shift means a single new AI campus can draw as much power as a small city, which is why site selection now starts with a utility's ability to deliver gigawatt-scale power rather than with tax incentives or fiber access.
The New Rules of the Land Grab
The practical result is a land grab defined by electrons rather than acreage. Developers are chasing sites next to existing substations, decommissioned power plants and gas pipelines; some are building on-site generation — gas turbines, fuel cells, even early-stage nuclear commitments — to get operational while their grid interconnection application is still in the queue. Land, permits and even chip supply can now move faster than a utility can deliver a confirmed power date, and that single fact is reshaping where, and how, the next generation of AI infrastructure gets built.