Hook
Maine Governor Janet Mills vetoed a bill that would have banned new data center construction in the state. The proposed ban emerged because these facilities—the physical backbone of cloud computing, streaming services, and AI—consume massive amounts of electricity and water, straining local grids that weren’t built for them. Mills sided with economic development: data centers bring jobs and investment. But the tension underneath doesn’t disappear with a veto. When a single building can draw as much power as 50,000 homes, who gets the electricity? What happens when a region’s grid reaches capacity and every megawatt becomes a choice between competing uses?
This isn’t an abstract policy question. It’s a material constraint surfacing everywhere digital demand meets physical infrastructure.
What Are Data Centers
A data center is a warehouse filled with servers. Rows and rows of computers, stacked in racks, running continuously. They’re the physical places where “the cloud” lives. When you stream a show, query ChatGPT, or access files from cloud storage, you’re pulling data from one of these buildings.
Most people imagine the internet as weightless—information floating in some digital ether. It’s not. Every email, every video, every AI interaction runs on hardware in a specific location consuming real electricity and producing real heat. Data centers house that hardware at scale.
These facilities are enormous. A single data center can span hundreds of thousands of square feet—think multiple football fields under one roof. Inside, the servers generate intense heat. Cooling systems run around the clock to prevent hardware failure, using either air conditioning or water-based systems. Many data centers consume millions of gallons of water annually just for cooling.
The electricity load is the bigger constraint. A mid-sized data center draws 20 to 30 megawatts continuously. That’s enough to power roughly 20,000 homes. Large hyperscale facilities built by companies like Amazon, Google, or Microsoft can exceed 100 megawatts. They run 24/7. There’s no off-peak. Uptime is the business model.
Maine’s debate centers on this: if you add a data center to a regional grid, you’re adding a load equivalent to a small city that never sleeps.
The Energy Problem
Grids have capacity limits. They’re designed to serve a certain peak demand based on the mix of residential, commercial, and industrial users in a region. When total demand approaches that limit, grid operators face choices: deny new connections, upgrade infrastructure, or ration power during peaks.
Data centers hit those limits fast. The average U.S. household uses about 30 kilowatt-hours per day—roughly 1.25 kilowatts sustained. A 30-megawatt data center uses the equivalent of 24,000 households. If a utility planned for gradual residential and commercial growth, a single data center application can consume years of projected spare capacity in one project.
Maine’s grid is relatively small. The state’s total electricity consumption averages around 1,200 megawatts. A large data center requesting 50 megawatts would claim over 4% of statewide capacity. That’s a meaningful fraction—especially when grids need headroom for peak demand days and contingency reserves.
Now add AI. Training a large language model like GPT-4 required an estimated 50 gigawatt-hours of electricity—the annual consumption of about 5,000 homes, compressed into months. Inference—running the model for user queries—also demands substantial power. As AI adoption grows, data centers built five years ago are undersized. New facilities are being planned specifically for AI workloads, with even higher power densities.
The International Energy Agency estimates global data center electricity use could double by 2026, driven primarily by AI and cryptocurrency mining. That doubling isn’t evenly distributed—it concentrates where companies can secure grid access and favorable energy prices. Regions with available capacity become magnets. Regions near capacity, like Maine, face pressure they didn’t anticipate.
The Trade Off
Here’s the dilemma facing Maine and similar regions: data centers offer economic benefits—construction jobs, ongoing tech employment, local tax revenue. But they consume resources other development needs.
If a utility has 100 megawatts of spare capacity and a data center takes 60, that’s 60 megawatts unavailable for housing developments, hospitals, manufacturing plants, or electric vehicle charging infrastructure. Every megawatt allocated is a megawatt denied elsewhere.
Upgrading the grid is theoretically possible but practically difficult. Transmission line projects take years to permit and construct. Substation upgrades require capital most rural utilities lack without rate increases. New generation capacity—whether renewable or fossil—faces regulatory hurdles and public opposition. Meanwhile, data center developers want answers in months, not decades.
Communities face a zero-sum calculation. Say yes to the data center and you get jobs now but constrain other growth for years. Say no and you lose investment but preserve capacity for diversified development. There’s no option that avoids the constraint—only choices about who bears it.
Water creates a parallel tension. Data centers using evaporative cooling withdraw millions of gallons from local aquifers or rivers. In regions with drought risk or competing agricultural use, that’s water someone else doesn’t get. Mesa, Arizona rejected a data center proposal in 2023 partly due to water concerns during a prolonged drought. The electricity question and the water question compound each other.
Mills vetoed the ban, arguing Maine needs economic development and shouldn’t close the door on an entire industry. The bill’s sponsors argued the state’s grid and water resources can’t absorb more data centers without harming existing residents and businesses. Both positions reflect real constraints. The veto doesn’t resolve the underlying scarcity—it just decides who has standing to claim the next slice.
Global Pattern
Maine isn’t alone. This collision between digital demand and physical capacity is surfacing worldwide.
Ireland hosts a high concentration of data centers due to tax policy and access to transatlantic fiber cables. By 2021, data centers consumed 14% of Ireland’s total electricity—up from 5% in 2015. The grid operator warned that peak demand could exceed supply by 2028 without major infrastructure investment. Dublin implemented a de facto moratorium on new data center connections in certain areas, forcing projects to queue for years.
Singapore, a regional hub for cloud services, paused new data center approvals in 2019 due to power and land constraints. The city-state’s total generation capacity is fixed by geographic limits, and data centers were outcompeting other industries for the available load. The moratorium lasted until 2022, when Singapore lifted it conditionally—new centers must meet strict energy efficiency standards and most can’t connect until new generation comes online.
The Netherlands restricted data centers in the Amsterdam and Haarlemmermeer areas after electricity demand growth outpaced grid capacity. Developers must now prove their projects meet heightened efficiency thresholds and contribute to renewable energy generation.
These aren’t anti-technology policies. They’re triage responses to infrastructure bottlenecks. Regions that said yes to early data center projects without accounting for cumulative load are now forced to ration access.
The global pattern is clear: data centers cluster in places with cheap electricity and permissive policy until the grid can’t absorb more. Then bottlenecks surface as political fights between growth factions. AI accelerates this timeline. A region that had five years of spare capacity in 2023 may have two years in 2026 as AI training and inference workloads multiply.
Energy markets respond, but slowly. Google, Microsoft, and Amazon are investing in new renewable generation and grid-scale batteries to offset their consumption. Microsoft signed a deal to restart a reactor at Three Mile Island to power data centers. But these projects take years to operationalize, and not every region has nuclear plants or renewable potential at the scale needed.
The mismatch between digital ambition and physical infrastructure is structural, not temporary.
Close
Mills chose jobs over the ban, but the grid’s capacity hasn’t changed. Maine will face this question again—and so will everywhere else—as AI makes data centers hungrier and grids stay finite.