Electricity has become the defining constraint of the data center era. As AI drives an unprecedented construction boom, the questions have moved from the server room to the utility bill: how much power do these facilities actually use, is the grid able to keep up, and why is the electricity bill going up in states where data centers cluster? The answers involve some genuinely large numbers and a fierce, still-unresolved fight over who pays for the grid these facilities require. This is a plain-numbers reference on data center electricity use in 2026, what the real figures are, where they come from, and what is driving the strain.
TL;DR
Data center electricity use has become one of the fastest-growing and most contested forces on the US grid. The essentials:
- US data centers used about 176 terawatt-hours in 2023, roughly 4.4% of national electricity. That is the widely cited federal benchmark, and it is projected to climb sharply.
- That share could reach 6.7 to 12% of US electricity by 2028. Federal analysis projects data center demand roughly doubling or tripling to 325 to 580 TWh within a few years, driven by AI.
- A single hyperscale facility can draw 100+ megawatts, enough to power around 80,000 homes. AI has pushed rack densities from 5 to 10 kilowatts to 50 to 100 kilowatts, concentrating enormous demand in single buildings.
- The grid strain is now showing up on household bills. In the PJM region, data centers accounted for an estimated $9.3 billion capacity-market cost increase in 2025-26, pushing typical residential bills up by roughly $16 to $18 a month in some states.
- A political fight over who pays is underway. Seven major AI companies signed a Ratepayer Protection Pledge in 2026 to self-fund grid infrastructure, and states are passing laws requiring large loads to cover their own costs, mirroring the ratepayer-protection push in Texas, Ohio, Virginia, and beyond.
The single most important thing to understand: data centers are industrial-scale electricity consumers running continuously, and their demand is arriving faster than the grid can be built. The result is a genuine capacity and cost strain, concentrated in the regions where facilities cluster, and an unresolved question of whether operators or ordinary ratepayers should pay to expand the grid to serve them.
The Headline Numbers
The most-cited figures come from federal analysis, and they describe a sector whose electricity use is both large and rising fast.
| Metric | Figure | Source basis |
|---|---|---|
| US data center electricity use, 2023 | ~176 TWh (~4.4% of US total) | Lawrence Berkeley National Laboratory |
| US data center electricity use, 2014 | ~58 TWh | Lawrence Berkeley National Laboratory |
| Projected use by 2028 | ~325 to 580 TWh (~6.7 to 12% of US total) | Lawrence Berkeley National Laboratory |
| Single hyperscale facility draw | 100+ MW (≈ 80,000 homes) | Industry reporting |
| AI rack power density | 50 to 100 kW (vs 5 to 10 kW traditional) | Industry reporting |
| US data center demand growth | ~23 GW (2023) to ~42 GW (2026) | Industry reporting |
A few things anchor these figures. The federal benchmark, from Lawrence Berkeley National Laboratory’s 2024 report for the Department of Energy, is that US data centers consumed roughly 176 TWh in 2023, about 4.4% of national electricity, up from about 58 TWh in 2014. That is the number most credible sources cite, and it is the baseline for the projections. Looking forward, the same analysis projects data center demand roughly doubling or tripling to between 325 and 580 TWh by 2028, which would put data centers at somewhere between 6.7 and 12% of all US electricity consumption within a few years, a trajectory summarized in Congressional Research Service analysis.
The reason those numbers are climbing so fast is AI. Traditional server racks drew 5 to 10 kilowatts; AI racks packed with GPUs draw 50 to 100 kilowatts, an order of magnitude more, concentrated in the same physical footprint. A single hyperscale campus can now draw more than 100 megawatts, roughly the electricity use of 80,000 homes, and the largest proposed AI campuses are being planned at gigawatt scale, the draw of a small city. That density is what turns a data center from a large commercial customer into a system-level force on the grid.
Why Data Centers Strain the Grid
The strain is not simply about total consumption. It is about how, where, and how fast that demand arrives, and about a grid that was not designed for it.
The Speed and Concentration Problem
Data center demand is arriving faster than grid infrastructure can be built. New generation, transmission lines, and substations take years to permit and construct, while a data center campus can be built in a fraction of that time. Interconnection queues (the waiting lists to connect new load and generation to the grid) now stretch five to seven years in many regions, and critical equipment like large transformers has fallen into severe backlog. The result is a mismatch: enormous new demand wanting to connect on a timeline the grid cannot match. Compounding this is concentration. Data centers cluster in specific regions (Northern Virginia, Ohio, Texas, Georgia, and a handful of others), so the strain is intensely local rather than spread evenly across the national grid.
The Continuous, Inflexible Load
Data centers run 24 hours a day, every day, at high utilization. Unlike many industrial loads that vary with shifts or seasons, a data center’s demand is close to constant, which means the grid must have firm capacity available for it at all times, including peak periods. This continuous, largely inflexible profile is harder and more expensive to serve than variable demand, and it is a poor match for the intermittent renewable generation that makes up a growing share of new supply. Regional grid operators built around a balance of competitive generation and predictable demand are finding that a single fast-growing class of customer with this profile stresses the system in ways it was not designed for.
The Reliability Question
Beyond cost, the sheer scale of new demand raises reliability concerns. When demand growth outpaces new supply, the margin between available generation and peak demand narrows, and some regions have flagged the risk of tighter reserves. This has pushed operators toward self-supply: contracting power directly from private producers, and increasingly building their own on-site generation (gas turbines, and in some high-profile cases nuclear, including small modular reactors) to secure capacity and timelines rather than wait for the grid. The move to on-site power is itself a signal of how strained grid interconnection has become.
Data Centers and Electricity Bills
The most contested part of the data center power story is its effect on ordinary electricity bills, and here the numbers have become concrete enough to quantify.
The PJM Case
The clearest evidence comes from PJM, the regional grid operator covering 13 states from Illinois to North Carolina and about 65 million people. In PJM’s capacity market (the mechanism that pays generators to guarantee future supply), data centers drove an estimated $9.3 billion cost increase in the 2025-26 delivery year. PJM’s own market monitor has attributed roughly 7.9 GW of additional data center demand in 2025-26 and about 12 GW in 2026-27 to the sector, and estimated that removing data centers from the forecast would cut capacity payments by roughly $9.3 billion, a large majority of the total. Capacity prices in the region rose sharply, by several hundred percent, between recent delivery years.
Those wholesale costs flow through to households. In PJM, the data center-driven capacity increase is expected to raise the typical residential bill by roughly $16 a month in Ohio and $18 a month in parts of Maryland. These are not projections of some distant future; they are increases already flowing into bills.
The Broader Picture
The effect extends beyond PJM. National retail power prices rose about 2.3% year-over-year recently, with data center demand growth cited as a primary driver, and one major-bank analysis estimated that data center-driven electricity demand would add measurably to core inflation in 2026 and 2027. The Federal Reserve Bank of Dallas has estimated that with data center demand expected to double within five years, wholesale power prices could rise by as much as 50%.
The important nuance is that the impact varies enormously by region. In markets with abundant cheap generation (parts of Texas, where data center demand has coincided with heavy renewable investment), the bill impact has been more muted; in constrained markets where new supply lags demand, it is acute. The reason ordinary ratepayers feel it at all is structural: when a utility builds new generation, transmission, and substations to serve large new loads, those grid-upgrade costs have traditionally been shared across the entire customer base, so households and small businesses help pay for infrastructure built primarily to serve data centers, unless specific ratepayer protections are put in place.
The Fight Over Who Pays
That last point, cost allocation, has become the defining regulatory battle of the data center era, and 2026 has seen it move quickly.
The Ratepayer Protection Pledge
In a notable 2026 development, seven of the largest AI and cloud companies (Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI) signed a White House-facilitated Ratepayer Protection Pledge, committing to fund the grid infrastructure needed to serve their facilities rather than relying on shared utility investment paid by all ratepayers. The pledge reflects growing political pressure, bipartisan in character, to ensure the companies driving demand also bear its infrastructure cost. Whether and how the pledge is enforced in practice remains to be seen, but its existence signals that the “socialize the cost across ratepayers” model is under serious challenge.
The State Legislation Wave
States have moved in parallel, and quickly. The common thread is defining a “very large customer” or large-load category and ensuring those customers pay for the infrastructure they require. Texas Senate Bill 6 requires large new loads (above 75 MW) to participate in demand-response programs and allows emergency disconnection during grid stress. Ohio, Georgia, Oklahoma, Minnesota, and other states have enacted or proposed measures pushing large-load infrastructure costs onto data center operators rather than the general ratepayer base, and utilities in several states have proposed dedicated data center rate classes or surcharges. This wave of state action, mirrored across the fast-growing data center markets, is the practical mechanism by which the cost-allocation question is being answered, one jurisdiction at a time.
The Unresolved Tension
Underneath the policy activity is a genuine and unresolved tension. Data centers bring substantial capital investment, significant construction employment, and a large property-tax base, and, operators argue, are willing to pay for the power they use. They also create relatively few permanent operating jobs per facility once built, and the grid upgrades they require are expensive and long-lived, raising the question of who should carry the risk if a facility’s demand does not materialize as projected. Consumer advocates argue ratepayers should be insulated entirely; operators and some economists argue that large loads already pay substantial costs and that blanket restrictions could push investment elsewhere. The balance struck between these positions, still being negotiated in 2026, will shape both electricity bills and where the next wave of data centers gets built.
Putting the Numbers in Context
Because the figures are abstract, a few grounded comparisons help, with the same caution the water debate requires: match the comparison to the right scale.
A single hyperscale data center drawing 100+ megawatts uses roughly the electricity of 80,000 homes, running continuously. That is an accurate single-facility comparison.
At the national level, data centers at ~4.4% of US electricity are a meaningful but not dominant share, comparable in scale to a few percent of total consumption, still smaller than sectors like residential heating and cooling or industrial manufacturing in aggregate. The reason data centers draw outsized attention despite that modest national share is, again, concentration and growth rate: the demand is rising far faster than any other load, and it lands in specific regions where it can represent a very large share of local consumption. In Virginia, the country’s densest data center market, data centers consume more than one in four kilowatt-hours of the entire state’s electricity. That local intensity, not the national average, is what strains grids and moves bills.
The honest framing is that data center electricity use is a rounding error at the national level in some respects and a genuine crisis at the local level in others, and both statements are true at once. The national number reassures; the local numbers alarm. Which one matters depends entirely on whether you live near the cluster.
What This Means for Infrastructure Planning
The electricity story connects to hardware in a way that is easy to miss: the same forces straining the grid are compressing hardware lifecycles and reshaping how facilities think about efficiency and refresh.
Efficiency Pressure Drives Refresh
When power is the binding constraint and its cost is rising, the efficiency of the hardware inside a facility becomes a direct economic lever. Newer chips deliver far more computation per watt, so operators facing power limits and rising energy costs have a strong incentive to refresh to more efficient hardware sooner. Power scarcity, in other words, is itself a driver of the compressed refresh cycles now common in AI infrastructure. That has a direct downstream effect: more frequent refresh means more retired hardware, sooner, and a larger, faster-moving stream of decommissioned equipment.
Reclamation as a Capacity Strategy
Where grid interconnection is the bottleneck and new power takes years to deliver, the energized, powered capacity a facility already has becomes its most valuable asset. Reclaiming that capacity, by efficiently decommissioning retired equipment and clearing powered space for redeployment, becomes a genuine capacity strategy in constrained markets, not merely a cleanup task. The facilities best positioned in a power-constrained era are the ones that use their existing energized footprint most efficiently, which makes disciplined decommissioning and refresh planning part of the power strategy itself.
The Retirement Volume
Finally, the sheer scale of the buildout means the volume of hardware moving through refresh and retirement is large and growing. As the AI capacity being energized now reaches the end of its compressed lifecycle, it generates a substantial stream of high-value, high-density equipment that has to be decommissioned, sanitized, and recovered, at a scale the power buildout implies.
Frequently Asked Questions
How much electricity do data centers use?
US data centers consumed roughly 176 terawatt-hours in 2023, about 4.4% of total national electricity consumption, according to Lawrence Berkeley National Laboratory’s 2024 report for the Department of Energy, up from about 58 TWh in 2014. That figure is projected to rise sharply, to between 325 and 580 TWh by 2028, which would represent roughly 6.7 to 12% of all US electricity. The growth is driven overwhelmingly by AI, which has pushed rack power densities from 5 to 10 kilowatts to 50 to 100 kilowatts. The national share is meaningful but not dominant; the strain comes from how fast the demand is growing and how concentrated it is in specific regions.
How much power does a single data center use?
A large hyperscale data center can draw more than 100 megawatts of electricity, roughly the amount used by 80,000 homes, running continuously around the clock. The largest AI campuses now being planned are designed at gigawatt scale, comparable to the electricity draw of a small city. The dramatic increase is due to AI hardware: traditional server racks drew 5 to 10 kilowatts, while AI racks packed with GPUs draw 50 to 100 kilowatts, concentrating an order of magnitude more demand in the same physical space. This density is what makes modern data centers a system-level force on the electricity grid rather than just large commercial customers.
Why do data centers strain the electric grid?
Data centers strain the grid for several reasons beyond total consumption. Their demand arrives faster than new generation, transmission, and substations can be permitted and built, with interconnection queues stretching five to seven years and equipment like transformers in severe backlog. They run continuously at high utilization, an inflexible load that requires firm capacity at all times and matches poorly with intermittent renewables. And they cluster in specific regions, so the strain is intensely local. Together these factors mean data center demand stresses regional grids in ways the system was not designed to handle, pushing some operators to build their own on-site power.
How do data centers affect electricity bills?
When utilities build new generation, transmission, and substations to serve large new data center loads, those grid-upgrade costs have traditionally been shared across all customers, so households help pay for infrastructure built mainly to serve data centers unless specific protections exist. In the PJM region covering 13 states, data centers drove an estimated $9.3 billion capacity-market cost increase in 2025-26, expected to raise typical residential bills by roughly $16 a month in Ohio and $18 a month in parts of Maryland. National retail prices have risen with data center demand cited as a primary driver. The impact varies by region, and is most acute where new supply lags demand.
How much will data centers increase electricity prices?
The impact varies widely by region. In the PJM market, data centers drove a roughly $9.3 billion capacity cost increase already flowing into bills as $16 to $18 per month in some states. The Federal Reserve Bank of Dallas has estimated that with data center demand expected to double within five years, wholesale power prices could rise by as much as 50%. National retail power prices rose about 2.3% year-over-year recently with data centers cited as a primary driver, and analysis has suggested a measurable effect on core inflation. However, in markets with abundant cheap generation, the impact has been more muted, so regional conditions determine how much prices actually rise.
Who pays for the grid upgrades data centers require?
This is the central policy fight of 2026. Traditionally, grid-upgrade costs are socialized across all ratepayers, meaning households subsidize infrastructure serving data centers. That model is under serious challenge. Seven major AI companies (Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI) signed a Ratepayer Protection Pledge in 2026 to self-fund their grid infrastructure. States including Texas, Ohio, Georgia, Oklahoma, and Minnesota have passed or proposed laws requiring large loads to cover their own infrastructure costs, often by defining a “very large customer” class, and utilities have proposed dedicated data center rate classes. The balance between insulating ratepayers and attracting investment is still being negotiated.
Does AI increase data center electricity use?
Yes, dramatically. AI is the primary driver of data center power growth. AI training and inference require high-density hardware, pushing rack power from the traditional 5 to 10 kilowatts to 50 to 100 kilowatts, an order of magnitude increase. This is why federal projections show data center electricity use potentially doubling or tripling by 2028. A single AI-focused hyperscaler can consume as much electricity as a small city. AI’s power intensity is also driving operators toward on-site generation and reshaping grid planning nationally. The one counterweight is efficiency: newer AI chips deliver more computation per watt, giving operators a strong incentive to refresh to more efficient hardware as power costs rise.
How does data center electricity use compare to other countries or industries?
At roughly 4.4% of US electricity in 2023, data centers are a meaningful but not dominant national consumer, smaller in aggregate than sectors like residential heating and cooling. What sets them apart is growth rate and concentration: no other major load is growing as fast, and the demand lands in specific regions where it can dominate local consumption. In Virginia, the densest US market, data centers consume more than one in four kilowatt-hours of the entire state’s electricity. So data center power use can be simultaneously a modest national share and a genuine local crisis, which is why national averages and local experiences diverge so sharply in the public debate.
How does the data center power buildout affect equipment retirement?
The same forces straining the grid compress hardware lifecycles. When power is the binding constraint and its cost is rising, hardware efficiency becomes a direct economic lever, so operators refresh to more efficient chips sooner, generating more retired equipment faster. Where grid interconnection is the bottleneck, the energized capacity a facility already has becomes its most valuable asset, making efficient decommissioning that reclaims powered space a genuine capacity strategy rather than a cleanup task. And the sheer scale of the buildout means the volume of high-density hardware moving through refresh and retirement is large and growing, all of which has to be decommissioned, sanitized, and recovered responsibly.
The Bottom Line
Data center electricity use has become one of the most consequential forces on the US power system. The federal benchmark is clear: US data centers used about 176 terawatt-hours in 2023, roughly 4.4% of national electricity, and that could reach 6.7 to 12% by 2028 as AI drives demand from 325 to 580 TWh. A single hyperscale facility now draws more than 100 megawatts, the equivalent of 80,000 homes, because AI has pushed rack densities an order of magnitude higher. But the defining issue is not the national total; it is that this demand arrives faster than the grid can be built and concentrates in specific regions, straining local systems and pushing costs onto ordinary ratepayers. In PJM alone, data centers drove a roughly $9.3 billion capacity cost increase now showing up as $16 to $18 a month on household bills.
That strain has ignited the central policy fight of the era: whether data center operators or ordinary ratepayers should pay to expand the grid. The 2026 Ratepayer Protection Pledge from seven major AI companies and a wave of state legislation defining large-load cost responsibility are the early answers, but the balance is still being negotiated. For the industry, the power constraint has a direct hardware consequence: rising energy costs and grid limits push operators to refresh to more efficient hardware sooner, compressing lifecycles, while the value of already-energized capacity makes efficient decommissioning and reclamation a genuine capacity strategy. The electricity story, in other words, is not only about generation and bills; it reaches all the way to how quickly the hardware inside these facilities is retired and replaced.
How ROC Telecom Fits In
ROC Telecom is an R2v3, RIOS, NIST 800-88, and ITAR-compliant ITAD specialist. Power and grid capacity are utility questions rather than disposition ones, but they connect directly to hardware: rising energy costs and grid constraints push operators to refresh to more efficient equipment sooner, and in power-constrained markets the value of already-energized space makes reclaiming it a real priority. Both dynamics turn decommissioning and asset recovery into part of the infrastructure strategy.
That is where ROC helps:
- Data center decommissioning with 48-hour rapid-response mobilization, reclaiming energized capacity efficiently in markets where new power takes years to deliver
- GPU and AI infrastructure asset recovery for the high-density, high-value hardware displaced by efficiency-driven refresh cycles, with speed-to-remarketing that protects value against generational decay
- Specialist asset recovery across routing, switching, optical transport, and compute, with direct buyer relationships
- NIST 800-88 data destruction with per-asset serialized Certificates of Destruction and full chain-of-custody documentation
- R2v3 Appendix E materials recovery with in-house dismantling and direct-to-refiner processing
- Mass-balance recovery reporting for the ESG and sustainability disclosures that facility resource use increasingly demands
15+ years of ITAD experience, $25M+ in client capital recovered, 45M+ pounds diverted from landfill.
Request a Free ITAD Program Assessment
Tell us about your infrastructure and what you are building, refreshing, or retiring. A specialist will reach out to discuss decommissioning timelines, recovery value, and certified disposition. No commitment, no spam.
Prefer to talk directly? Call 585-406-1249 or email info@roctelecom.com.
"*" indicates required fields
Related reading:
- How Much Water Does a Data Center Use? The 2026 Numbers
- Nuclear Power and Data Centers: The 2026 Energy Overview
- Why AI Is Shortening Your Hardware Lifecycles (and What to Do About It)
- The Northern Virginia Data Center Market: A 2026 Overview
- Data Center Asset Recovery: The 2026 Strategic Guide
- The Texas Data Center Market: A 2026 Overview
- The Oklahoma Data Center Market: A 2026 Overview
