Artificial intelligence may feel like a digital service, but the infrastructure behind it requires enormous amounts of physical energy.
Across the United States, technology companies are building massive data centers filled with high-performance computer chips used to train and operate AI systems. These facilities can consume as much electricity as a town or small city, and their rapid expansion is forcing utilities to build new power plants, substations and transmission lines.
In some regions, part of that cost may ultimately appear on the monthly bills paid by households and small businesses.
The relationship is not identical in every state, and data centers are not the only reason electricity prices are rising. Fuel costs, severe weather, aging infrastructure, wildfire protection, inflation and general grid investment also affect rates. However, regulators and consumer advocates are increasingly concerned that ordinary customers could be asked to help finance infrastructure built mainly to serve some of the world’s largest technology companies.
America’s Electricity Demand Is Growing Again
US electricity consumption remained relatively flat for much of the period between the mid-2000s and early 2020s.
That pattern has now changed. The US Energy Information Administration says national electricity use has been rising steadily since 2020, with data centers playing a major role in the increase. Demand is also growing because of new factories, electric vehicles and the wider electrification of buildings and industrial processes.
The shift matters because electric grids are built around forecasts of how much power customers will need. When demand rises faster than expected, utilities may have to accelerate investment in generation, transmission and distribution equipment.
The US Energy Information Administration reports that electricity demand grew by an average of approximately 1.7 percent annually between 2020 and 2025, compared with only about 0.1 percent per year from 2005 through 2019.
Data Centers Already Consume a Significant Share of US Power
Data centers used approximately 176 terawatt-hours of electricity in 2023, accounting for around 4.4 percent of total US electricity consumption, according to research led by Lawrence Berkeley National Laboratory.
The same analysis estimated that data-center electricity use could rise to between 325 and 580 terawatt-hours by 2028. Depending on broader national electricity growth, that would represent roughly 6.7 to 12 percent of all US power consumption.
The increase is being driven partly by conventional cloud computing, video streaming and online services. AI is accelerating the trend because graphics-processing units and other specialized chips require large amounts of electricity and generate substantial heat.
Cooling those chips requires additional energy for pumps, fans, chillers and water systems. The facility must also maintain backup power, networking equipment and other infrastructure even when computing demand changes.
The full Berkeley Lab data-center energy report provides detailed estimates of current and projected consumption.
Why New Data Centers Can Affect Household Rates
A large data center may require hundreds of megawatts of continuous electrical capacity.
Before connecting it, a utility may need to construct substations, replace transformers, expand transmission lines or secure additional power-generation capacity. These projects can cost hundreds of millions or billions of dollars.
Utilities generally recover approved infrastructure spending through rates charged to customers. The central policy question is how much of the cost should be assigned directly to the data-center operator and how much should be distributed across households and other businesses.
When a technology company pays the complete cost of its connection and commits to purchasing electricity for many years, the additional demand could potentially benefit other customers by spreading fixed grid costs across a larger amount of electricity.
The opposite can happen when a planned data center requires expensive infrastructure but delays construction, reduces its electricity use or cancels the project. Existing customers may be left paying for equipment that was built around demand that never fully appeared.
California regulators and consumer advocates have therefore recommended special rate structures requiring very large customers to make advance payments, cover more infrastructure costs and commit to paying for the power capacity they request.
Capacity Charges Are Already Rising in Some Regions
The impact has become particularly visible in the PJM electricity market, which serves parts of 13 states and the District of Columbia.
PJM includes major data-center markets such as northern Virginia, Ohio, Pennsylvania and New Jersey. Rapid demand growth has increased concern about whether the region has enough available generating capacity during periods of peak use.
Capacity markets pay electricity generators to remain available when demand is highest. When projected demand increases faster than available supply, capacity prices can rise sharply.
Those costs are passed through utilities and energy suppliers and may eventually reach residential, commercial and industrial customers.
A July 2026 Reuters investigation described how an Ohio brick manufacturer experienced a roughly 90 percent increase in electricity costs, driven largely by higher capacity charges amid surging demand associated with data-center development.
Federal regulators have recognized that data centers create major questions for both reliability and consumer costs. The Federal Energy Regulatory Commission has required grid operators to address how extremely large computing facilities connect to power plants and share grid infrastructure.
Virginia Shows How Quickly Demand Can Change
Northern Virginia contains one of the largest concentrations of data centers in the world.
Between 2019 and 2025, commercial electricity sales in Virginia increased by nearly 30 million megawatt-hours, according to the EIA. The agency said the growth was largely driven by the state’s concentration of data centers, along with electric vehicles and building electrification.
Data centers can benefit local governments by generating tax revenue and supporting construction activity. They may also create fewer permanent jobs than their enormous physical and electrical footprint would suggest.
The challenge is that concentrated demand can require utilities to build infrastructure much faster than they would for ordinary population or business growth.
A facility can be approved and developed within several years, while a major transmission line or power plant may take much longer to permit and construct. That timing mismatch can tighten power supplies and raise short-term costs.
Americans Are Not All Experiencing the Same Impact
It would be misleading to say that every increase in every American electricity bill has been caused by AI.
Electric rates are regulated differently across states, and utilities vary in how they assign infrastructure costs to large customers. Some data centers negotiate their electricity directly, construct their own power generation or pay special tariffs designed to protect households.
A 2026 working paper examining US data from 2015 through 2024 found that data centers may actually have produced modest average reductions in retail electricity rates during that historical period by spreading fixed system costs over a larger customer base. The researchers cautioned that future supply constraints could reverse that result as demand accelerates.
Another analysis found little statistically significant evidence that simply living close to a data center had increased household electricity bills historically. Its more closely matched estimate suggested an increase of about 90 cents per month, but the finding was not statistically conclusive.
These findings do not eliminate the concern. They show that the impact depends on timing, market design, available generation and whether data centers pay their full share of new infrastructure.
The strongest evidence is therefore not that AI has already raised every household bill by a specific amount. It is that data-center growth is creating a substantial risk of future rate increases in regions where supply and grid construction cannot keep pace.
The Cost Can Be Hidden Inside Broader Rate Increases
Electric bills do not normally contain a separate line labeled “AI data-center charge.”
Instead, costs may appear through higher generation prices, transmission charges, capacity payments or approved utility investments. A consumer may see only that the price per kilowatt-hour or a monthly delivery charge has increased.
That makes the effect difficult to isolate.
A utility may request a rate increase covering several projects at once, including storm recovery, grid modernization, renewable-energy connections, wildfire prevention and data-center infrastructure.
The data-center portion can therefore be buried inside a much larger regulatory filing that few customers read.
Consumer advocates are calling for greater transparency so regulators can identify exactly how much infrastructure is being constructed for large computing facilities and who will pay if projected demand fails to materialize.
Data Centers Could Also Increase Fossil-Fuel Use
The electricity problem is not limited to price.
Utilities and technology companies are exploring new natural-gas plants, delayed coal-plant retirements, nuclear generation, renewable energy and onsite power systems to meet data-center demand.
The EIA has warned that faster-than-expected electricity growth could increase fossil-fuel generation, particularly when new demand arrives before enough renewable generation, storage and transmission capacity can be completed.
A 2026 environmental analysis identified dozens of planned gas-fired power projects intended partly or entirely to support data centers. Critics warned that these facilities could produce substantial greenhouse-gas and local air-pollution emissions.
Some technology companies argue that building onsite generation can reduce pressure on the public grid. However, onsite natural-gas plants may still create environmental and public-health concerns for nearby communities.
AI Facilities Need Reliable Power Around the Clock
Many ordinary electricity customers change their usage throughout the day. Residential consumption usually rises during morning and evening periods, while commercial demand follows working hours.
Large data centers often operate continuously.
AI models may be trained over days or weeks, while online AI services must remain available whenever users submit requests. The facilities also require highly reliable electricity because interruptions can damage equipment, interrupt customer services and waste expensive computing time.
This demand can be helpful to utilities when it remains stable and predictable. A large customer using power throughout the day may improve the economics of certain generation and transmission assets.
The difficulty appears during system peaks, when households, businesses and data centers all need electricity at the same time. Utilities must maintain enough capacity for the highest-demand hours, even when that capacity is used only occasionally.
Technology Companies Say They Are Paying Their Share
Major technology companies frequently argue that their investments strengthen local economies, expand the tax base and support the construction of new clean-energy projects.
Some companies sign long-term agreements to purchase renewable electricity or fund new generation. Others are pursuing nuclear power, geothermal energy, battery storage and onsite gas generation.
These arrangements can add new electricity supply rather than simply drawing more energy from the existing system.
Supporters also argue that large customers can lower rates if they pay substantial fixed charges and help spread the cost of maintaining the grid.
Whether those benefits reach households depends on the details of utility tariffs and regulatory decisions. A poorly structured agreement can protect the data center from financial risk while shifting costs to other customers. A carefully designed agreement can require the facility to pay for the infrastructure and generation it needs.
Regulators Are Considering Special Data-Center Rates
Several states and utilities are developing dedicated tariffs for extremely large electricity users.
These rate structures may require data centers to pay upfront connection costs, guarantee a minimum level of electricity payments and provide financial protection if a project is cancelled.
Regulators may also require companies to reduce consumption during grid emergencies or supply some of their own power.
FERC has ordered the PJM grid operator to develop clearer rules for very large facilities connecting directly or indirectly to power plants. The commission said the issue has major implications for reliability and fair consumer rates.
New York imposed a one-year moratorium on certain large new data centers in July 2026, citing concerns about utility bills and pressure on natural resources. The policy applies to projects using at least 50 megawatts of power and reflects growing political resistance to unchecked data-center development.
Data Centers Could Become More Flexible
Not every AI workload must run at the same time or in the same place.
Researchers and technology companies are exploring systems that allow data centers to reduce power use temporarily during periods of grid stress. Nonurgent computing jobs could be delayed or shifted to facilities in regions with more available electricity.
A 2026 research paper demonstrated that a GPU-based computing cluster could reduce and shift electrical demand while maintaining priority services. The authors argued that flexible AI facilities could become grid-supporting resources rather than operating as constant, inflexible loads.
This approach could reduce the need for expensive power plants used only during peak periods.
However, flexibility requires cooperation from technology companies and electricity pricing that rewards them for reducing demand when the grid is under strain.
What Consumers Can Do
Individual households have limited control over where data centers are built or how utilities assign their costs.
Consumers can still participate in public utility commission proceedings, where proposed rate increases and special data-center tariffs are reviewed. These proceedings are often open for written public comments.
Customers can also examine their utility bills to determine whether generation, transmission or capacity charges are increasing. Comparing current rates with previous years can help separate higher consumption from higher electricity prices.
Energy-efficiency improvements may reduce household use, but customers should not be expected to solve a system-level affordability problem through personal conservation alone.
The larger responsibility belongs to utilities, regulators and data-center operators to ensure that the companies creating extraordinary new demand pay an appropriate share of the resulting costs.
The Main Takeaway
AI data centers are contributing to the fastest period of US electricity-demand growth in decades.
They already consume a meaningful share of national power, and Berkeley Lab projects that their share could rise to between 6.7 and 12 percent by 2028.
That demand does not automatically raise household rates. Data centers can sometimes spread fixed grid costs across more electricity sales and help finance new generation.
The risk arises when utilities build costly infrastructure for data centers without ensuring that those facilities will pay for it over the long term. In constrained markets, rapidly rising demand can also increase capacity and generation prices before new supply becomes available.
Americans may never see “AI” printed on their electricity bills. The cost can instead appear through higher delivery charges, capacity costs and utility-rate increases.
The question is no longer whether AI requires enormous amounts of electricity. It is whether regulators will require the technology companies benefiting from that electricity to cover the full cost or allow part of it to be quietly passed to everyone else.
