A couple called Ken and Carol Apacki discovered their electricity bill had increased by 60% in only a single year in a central Ohio neighborhood. There were now 130 data centers in the area, and Carol thought she knew why. She was correct. The bill, which was once manageable, now depicts an event occurring at a scale that has surpassed the grid that was intended to serve everyone equally.
The grid was constructed in the 1950s and 1960s for a nation that had no idea that it would eventually need to train artificial intelligence models that consumed more power than entire cities. If all planned facilities are built, the pipeline of new data centers under construction will add 140 gigawatts of new load to the present US total, which is now less than 15 GW. It is uncomfortable to think about that math.
| Category | Details |
|---|---|
| Scale of Problem | Current US data center sector draws under 15 GW of power — planned new facilities could add 140 GW more |
| US Power Demand Trend | Total US electricity consumption projected to rise from 4,110 billion kWh (2024) to 4,260 billion kWh (2026) — record levels after two decades of flat demand |
| Virginia Impact | In 2023, data centers consumed 26% of Virginia’s total electricity — “Data Center Alley” in Ashburn nearly caused regional blackout in 2024 when 60 centers dropped off grid simultaneously |
| Grid Operator Warning | PJM Interconnection (serves 65 million people) projects a 6 GW shortfall in reliability requirements by 2027 |
| Consumer Cost Impact | Average residential bills rising — Ohio families paying 60% more; Carnegie Mellon study projects 8% average US bill increase by 2030 |
| Power Intensity of AI | A single AI task can consume up to 1,000 times more electricity than a traditional web search |
| Geographic Shift | Tech companies moving to power-rich regions — Microsoft investing $15.2B in UAE; Meta building $10B campus in Louisiana; Canada’s Alberta emerging as AI hub |
| Austin Warning | City analysis found proposed data centers seeking more power than the entire city’s peak load can deliver |
| Energy Source Problem | Approximately 60% of data center electricity still comes from fossil fuels |
| Further Reading | Grid and energy policy analysis at US Energy Information Administration |
After two decades of virtually unchanged power demand, the total amount of energy consumed in the United States is expected to hit record levels in 2025 and 2026, growing from around 4,110 billion kilowatt-hours in 2024 to more than 4,260 billion kilowatt-hours. Neither an industrial revival nor a population boom is the catalyst. Server farms are involved.
Because a single AI-related job can use up to 1,000 times more electricity than a typical online search, a small number of AI facilities can put a regional power supply under stress that hundreds of conventional data centers would never experience. The businesses constructing these facilities were aware of the figures. In many instances, the utility regulators and grid operators did not respond fast enough.
Compared to most, Virginia has experienced this earlier and more keenly. About 26% of Virginia’s electrical supply was used by data centers in 2023. The concentration in Ashburn, referred to as Data Center Alley in the industry, has become a case study of what occurs when demand exceeds infrastructure. Sixty data centers simultaneously switched off the grid and switched to backup generators in 2024 due to a protective system failure.
This incident barely prevented severe power outages and discreetly alerted grid engineers around the region. The biggest grid operator in the United States, PJM Interconnection, which serves more than 65 million people in 13 states, now predicts that it will fall six gigawatts short of its reliability standards in 2027. According to independent expert Joe Bowring, who keeps close eye on PJM, he has never witnessed the grid under such anticipated stress.
The biggest tech corporations’ reactions have been instructive. They are transferring their funds to the actual location of the power rather than waiting for the grid to catch up. Microsoft pledged $15.2 billion to build data centers in the UAE, which is directly related to the region’s capacity to supply enough power and promote renewable energy.
In Louisiana, Meta is constructing a $10 billion campus. Because of its energy resources, Alberta, Canada, is becoming a destination for hyperscale AI. Electricity follows capital, not the other way around. The location of AI infrastructure, as well as the jobs and economic activity it creates, will be affected by this geographic change. It’s feasible that a disproportionate amount of AI development over the next ten years will go to the nations and areas who invest in grid capacity now. That wager is as much about infrastructure as it is about technology.
It is difficult to ignore the unequal distribution of the costs associated with all of this. To supply power to data centers, the grid must be increased, but everyone must pay for this growth, not just data centers. According to a Carnegie Mellon study, data center and cryptocurrency demand is expected to raise US electricity prices by 8% on average by 2030; in the markets with the highest demand, this rise could reach 25%.
Companies are facing an increasing conflict between AI capabilities and their own environmental obligations, since over 60% of the energy used in data centers still comes from fossil fuels. Utility commissions, state legislatures, and congressional offices are only now starting to take the issue of who will pay to upgrade the grid, and how quickly, seriously. The grid that built the American economy over the course of seventy years is now being asked to carry a load it was never designed for.