For most of the history of the internet, data centers were a manageable part of the electrical grid — a growing slice, but a predictable one. That predictability ended with the AI boom, and the speed of the change has caught utilities, regulators, and entire regional power systems off guard. As covered in Essay #1, every server generates heat that must be removed, and removing that heat at scale requires enormous amounts of electricity. But the deeper story now is not about cooling — it’s about the raw, explosive growth in computing demand itself, and the grid’s inability to keep pace.
To put that growth rate in context: Gartner’s June 2026 forecast projects data center electricity consumption will reach over 1,200 terawatt-hours globally by 2030, a level at which the firm states plainly that grid supply will be insufficient to meet construction demands, affecting every category of data center user — not just new AI facilities, but the conventional cloud infrastructure already running today.
Traditional data center workloads — hosting websites, running email servers, storing files — are relatively light on power per task. AI workloads are not. Training and running large AI models requires sustained, intensive computation that draws vastly more electricity than the workloads data centers were originally built to handle.
This is why a relatively small number of AI-focused facilities can strain a regional power grid in ways that hundreds of conventional data centers never did. The concentration of demand is the problem: it isn’t spread evenly across the system, it’s slammed into specific substations, specific transmission corridors, and specific utility service territories, often faster than those systems were ever designed to absorb.
Until recently, the primary constraint on data center construction was hardware — the availability of advanced processors and the chip manufacturing supply chain. That has changed. Industry analysts now describe the bottleneck as having “migrated from the server rack to the substation.” Utility interconnection queues — the process by which a new electricity customer gets approved to connect to the grid — are now delaying projects by years in major data center corridors. Building the high-voltage transformers and transmission infrastructure required to serve a new gigawatt-scale facility can take the better part of a decade, while the data center building itself can be constructed in 18 to 24 months.
That mismatch — fast construction timelines colliding with slow grid upgrade timelines — is now considered the single greatest structural barrier to the industry’s continued growth. It is also the reason developers are increasingly pursuing on-site power generation, including natural gas turbines and small nuclear reactors, to bypass the public grid entirely. That approach solves the developer’s interconnection problem, but it raises its own set of environmental and community concerns, which later essays in this series address directly.
When a new data center requires grid upgrades — new substations, new transmission lines, expanded generation capacity — someone has to pay for that infrastructure. In most states, the cost is socialized across the entire utility’s customer base through rate increases, rather than being borne entirely by the data center operator that created the need.
Virginia is the clearest case study because it has the longest data center history and the most data center capacity of any state. Data centers already consume more than one in four kilowatt-hours generated in Virginia, and that demand drove half of all U.S. electricity demand growth in the past year. The national average residential electricity rate hit 17.45 cents per kilowatt-hour in January 2026 — a 9.5 percent year-over-year increase that significantly outpaced general inflation.
Utilities and data center operators dispute how directly these increases are tied to data center load. Dominion Energy has publicly stated that data centers “do not currently have an outside influence on customer energy bills.” Virginia’s own Joint Legislative Audit and Review Commission largely agreed in a December 2024 study, finding that data centers were, at that point, paying their fair share of system costs — while warning that residential customers could begin absorbing more of the burden if policy didn’t change. In response to mounting pressure, Dominion created a dedicated rate class for large data center customers, an acknowledgment that the old rate structure wasn’t built for loads of this size.
The unresolved question, in Virginia and everywhere else this fight is playing out, is whether new rate structures actually isolate data center costs the way they’re supposed to, or whether residential customers end up subsidizing infrastructure that primarily benefits a handful of trillion-dollar companies. That question is exactly why utility rate cases have become one of the most important — and least understood — battlegrounds in this fight, a topic covered in greater depth in Essay #11.
Perhaps the most striking figure in the current data is this: AI data centers are expected to triple their share of total U.S. electricity consumption, from 4.4 percent in 2023 to 12 percent by 2028. That is not a marginal shift in the country’s energy mix — it is a fundamental restructuring of where American electricity goes, compressed into a span of roughly five years. No other sector of the economy has grown its electricity footprint this fast, this suddenly.
“Power availability is now officially the primary constraint to new construction… a critical maturity gap has emerged between the rapid innovation cycles of data center technologies and the slow, linear deployment pace of the underlying electrical grid infrastructure.” — Industry grid analysis, February 2026
If a data center is proposed near you, the electricity question deserves the same scrutiny as the water question covered in Essay #2. A few things are worth finding out before a project gets approved.
The electricity story behind the data center boom isn’t really about data centers anymore — it’s about whether the American electrical grid, built for a different era of demand, can absorb a tripling of high-density industrial load in less than a decade without passing the cost and the instability on to everyone else who depends on that same grid. The industry’s own growth projections suggest the answer, without major intervention, is no. Communities that understand this dynamic are in a far stronger position to demand real answers — about rate structures, about who pays for upgrades, and about whether the lights stay on and affordable for the people who lived there long before the server racks arrived.