U.S. AI Data Center Demand Is Boosting Gas and Solar, While Power Grids Fall Behind

U.S. AI data center demand is forcing utilities to plan for loads that could reach 12 percent of national electricity use by 2028. Natural gas and solar projects are advancing together, despite their sharply different climate profiles. The conflict is no longer simply clean energy versus fossil fuels. It is fast power versus a grid that takes years to expand.

A report published by The Athens Messenger on October 7 highlighted how AI infrastructure is stimulating both natural gas and solar development. It also described the pressure landing on regional power grids. That combination matters because data centers need continuous electricity, while new transmission lines and substations face long development timelines.

The result is an uncomfortable race. Technology companies want computing capacity within a few years, utilities must preserve reliability, and communities want protection from higher costs and pollution. Solar can add substantial energy quickly, but it cannot deliver every hour without storage or other firm resources. Gas plants can run when needed, but their construction risks locking emissions into the grid for decades.

AI Data Center Energy Demand Has Entered Utility Planning

The data center boom has moved from technology forecasts into the physical plans of electric utilities.

The Department of Energy estimated that U.S. data centers consumed 176 terawatt-hours of electricity in 2023. That represented approximately 4.4 percent of national electricity consumption. In 2014, the same sector used 58 terawatt-hours.

The department’s data center forecast placed 2028 consumption between 325 and 580 terawatt-hours. That equals between 6.7 and 12 percent of total U.S. electricity use.

The width of that range is important. Utilities cannot wait for perfect certainty before ordering equipment or planning transmission. However, building for the highest estimate can leave customers paying for infrastructure that later proves unnecessary.

AI intensifies this planning problem because its electricity demand arrives in unusually large blocks. A traditional commercial development might connect customers gradually. An AI campus can request hundreds of megawatts, sometimes with expectations for later expansion into gigawatt-scale operations.

Graphics processing units also concentrate more computing inside each building. Their power becomes heat, which creates additional demand for cooling, pumps, fans, and supporting equipment. A data center therefore needs electricity for both computation and the systems that keep its servers operating.

The International Energy Agency expects global data center electricity consumption to rise from about 460 terawatt-hours in 2024 to more than 1,000 terawatt-hours in 2030. AI represents the most important driver, alongside continued growth in conventional digital services.

The United States occupies a particularly exposed position. The IEA projects that data centers will account for nearly half of U.S. electricity demand growth through 2030. By then, processing data could consume more electricity than manufacturing energy-intensive goods combined.

These projections do not mean every announced campus will be built. Developers sometimes submit overlapping requests while comparing multiple locations. Some projects will shrink, move, or disappear if financing and AI demand disappoint.

However, utilities must still evaluate those requests. They must determine whether generation, transmission, and local substations can support each project without weakening service for existing customers.

That process changes the central question. The issue is no longer whether AI uses substantial energy. The question is which projects are real, when they will connect, and who must fund the required infrastructure.

This uncertainty explains why AI data center energy demand pressures utilities before a server hall opens. Forecasts influence generation purchases, transmission plans, reserve requirements, and customer rates. A speculative project can therefore create real planning costs long before it consumes electricity.

The demand is also geographically concentrated. Data centers cluster near fiber networks, available land, favorable permitting, and existing cloud regions. Local grids can face severe pressure even when data centers remain a modest share of national electricity use.

Northern Virginia remains the clearest example, but the pattern is spreading. Ohio, Pennsylvania, Texas, Arizona, Georgia, and other markets are receiving large connection requests. Each region has a different combination of fuel supplies, transmission constraints, water availability, and regulatory rules.

That geography determines what kind of power gets built. A solar-rich region might favor photovoltaic capacity and batteries. A location near gas pipelines may turn toward combustion turbines or onsite generators. Areas with existing nuclear plants can seek long-term contracts for firm, low-carbon electricity.

The common requirement is speed. Data center operators compete to install computing capacity before rivals. Electric infrastructure, by contrast, still follows development cycles designed for slower and more predictable demand growth.

Natural Gas Wins the Race for Firm Power

Natural gas is gaining because it offers dispatchable electricity on a schedule that better matches urgent data center development.

Dispatchable generation is a power source that operators can call upon when demand rises. Gas turbines provide that control, unlike solar output that changes with daylight and weather.

The IEA estimates that natural gas already supplies more than 40 percent of the electricity physically consumed by U.S. data centers. Renewables provide about 24 percent, nuclear supplies around 20 percent, and coal contributes roughly 15 percent.

Those figures describe the electricity reaching facilities through local grids and onsite systems. They differ from corporate procurement claims, which can include renewable certificates generated in another place or at another time.

Through 2030, the IEA expects natural gas to provide over 130 terawatt-hours of additional annual U.S. data center electricity. Renewables are projected to add approximately 110 terawatt-hours during the same period.

The IEA’s energy supply outlook therefore supports the central reversal. AI companies remain major renewable energy buyers, yet their expansion also strengthens the market for fossil generation.

Gas offers several operational advantages. Plants can produce power through the night, respond to changing demand, and support the grid when solar or wind output falls. The United States also has extensive gas production and pipeline infrastructure.

Some developers are considering onsite generation because grid connections take too long. Behind-the-meter generation supplies a facility without relying entirely on the public grid. It can include gas turbines, fuel cells, batteries, or hybrid systems.

This approach can shorten the time between constructing a data center and energizing its servers. It also shifts part of the generation investment from utilities toward technology companies and infrastructure partners.

However, onsite gas does not eliminate local constraints. Projects still need fuel delivery, air permits, noise controls, and backup arrangements. Pipeline capacity can become another bottleneck, especially when heating demand peaks during cold weather.

The emissions consequences also extend beyond the plant. Gas production and transportation can release methane, a greenhouse gas with strong near-term warming effects. New turbines can operate for decades unless owners retire them early or convert them to lower-carbon fuels.

Utilities face a difficult investment decision. A gas plant can meet a large customer’s near-term schedule, but the plant may outlive the computing project that justified it. Data centers can change ownership, become more efficient, or close before power infrastructure reaches the end of its useful life.

This creates stranded-asset risk. If a large customer leaves, regulators must decide whether other customers should continue paying for the plant, pipeline, or transmission upgrades.

Gas generation can also complicate technology companies’ climate commitments. Annual renewable purchases do not guarantee that a data center receives carbon-free electricity during every operating hour. A facility can buy enough solar energy across a year while consuming gas-fired electricity overnight.

Some companies have responded by pursuing hourly carbon-free energy matching. Under that approach, clean generation must align more closely with when and where electricity is consumed. That standard demands more storage, transmission, firm clean generation, or flexible computing.

Gas therefore wins the near-term race partly because the clean alternatives remain incomplete. Nuclear plants take years to build or restart. Advanced geothermal is not available everywhere. Long-duration storage has not reached wide commercial deployment.

The risk is that a temporary bridge becomes permanent infrastructure. Once utilities invest in gas plants and pipelines, they gain financial and operational reasons to keep those assets running.

The case for gas is strongest when data center demand is inflexible. If facilities can shift selected computing jobs away from stressed hours, the system needs fewer plants reserved for its highest peaks.

That possibility puts software design, workload scheduling, and electricity contracts at the center of the energy debate. A data center that changes when it computes can reduce the amount of firm capacity required.

Solar Growth Does Not Remove the Reliability Gap

Solar can provide large volumes of relatively fast electricity, but energy volume is not the same as round-the-clock capacity.

Solar projects are attractive to data center operators for several reasons. They can be developed faster than many conventional power plants, their operating costs are predictable, and they support corporate emissions goals.

Technology companies have also become major buyers under power purchase agreements. These contracts give renewable developers a committed customer, which can help projects secure financing.

The IEA reported that technology and data center companies procured 86 gigawatts of renewable capacity in the United States between 2015 and 2024. That purchasing power has helped accelerate wind and solar construction.

Solar fits daytime data center demand well. It can also charge batteries, which release electricity during evening peaks or short grid disturbances. Pairing solar with storage makes the resource more useful than solar alone.

Still, even a large battery normally covers a limited number of hours. Several cloudy days, seasonal changes, and overnight demand require other resources or a much larger storage system.

This is the central mechanism behind simultaneous solar and natural gas growth. Solar supplies low-cost energy when available. Gas supplies controllable capacity when renewable output does not match demand.

The pairing can reduce fuel consumption compared with running a gas plant continuously. It does not create a fully carbon-free system, especially when data centers operate at high utilization throughout the day.

Transmission adds another complication. The best solar resources often sit far from major data center clusters. Moving that electricity requires high-voltage lines, transformers, substations, and completed interconnection studies.

Interconnection is the process that evaluates whether a new generator or customer can safely join the grid. Queues have grown as solar, storage, manufacturing, and data center projects compete for limited network capacity.

A solar farm can therefore be physically complete but unable to deliver its full output. Likewise, a data center can have buildings and servers ready while waiting for an upgraded substation.

The Electric Power Research Institute modeled how different operating rules affect this mix. Its generation scenarios show natural gas dominating incremental supply under reference policies.

When data centers follow round-the-clock carbon-free commitments, the model shifts toward more solar, wind, storage, and other low-emission resources. This finding places significant responsibility on the buyers, not only the utilities.

Flexible demand produces another shift. Some AI tasks need immediate responses, including user-facing inference and safety systems. Other jobs, such as portions of model training or batch processing, can tolerate delays.

A data center could pause flexible tasks during a grid emergency, then resume when renewable output or system capacity improves. Operators could also move selected work between facilities in different regions.

That flexibility is not free. Hardware represents expensive capital, and operators want it processing jobs as often as possible. Delaying work can affect product schedules, customer commitments, and the economics of an AI cluster.

Electricity prices and contracts must make flexibility worthwhile. A flat rate gives a facility little reason to reduce demand during stressed hours. Time-sensitive pricing or capacity payments can reward helpful behavior.

The grid also needs reliable communication with participating facilities. Utilities must know how much load can move, how quickly it can respond, and whether performance matches contractual promises.

Solar growth remains essential under almost every credible data center scenario. The IEA expects renewables to meet nearly half of additional global data center electricity demand through 2030.

Yet that statistic should not be mistaken for a complete reliability solution. Electricity systems must satisfy demand during their most difficult hours, not only balance total annual production.

The decisive contest is therefore not solar versus gas as isolated technologies. It is flexible, coordinated data center operation versus inflexible demand backed by new combustion capacity.

The Grid Is the Hardest Component to Scale

Generation receives more attention, but transmission and distribution determine whether electricity can reach a data center at the required time and place.

A power plant and a server campus do not form a functioning system by themselves. Electricity must travel through transmission lines, substations, transformers, switches, and local distribution equipment.

Many of these components have long manufacturing and construction timelines. Large transformers can require specialized materials and customized engineering. Transmission projects also face land acquisition, permitting, environmental review, and disputes over cost allocation.

A data center can be built faster than this supporting network. That mismatch encourages developers to seek locations with spare capacity or consider onsite power.

Research published in 2026 found that a U.S. data center can take 18 to 36 months to construct. Securing full grid power can require more than four or five years in constrained regions.

That gap changes site selection. Access to firm electricity can matter as much as land prices, taxes, water, or proximity to users. A location with available transmission capacity can become more valuable than one offering larger financial incentives.

Grid operators must also prepare for peak demand, not just average consumption. They maintain reserves for equipment failures, weather extremes, and unexpected changes in supply or demand.

PJM Interconnection manages the wholesale electricity system serving all or parts of 13 states and the District of Columbia. Its territory includes major data center markets in Virginia, Ohio, and Pennsylvania.

PJM’s 2025 planning materials projected approximately 31,600 megawatts of peak-demand growth by 2030. The forecast identified data centers as a primary contributor.

That growth arrives while some older power plants are retiring. New generation and transmission must therefore cover both replacement needs and additional demand.

Planning becomes harder when connection requests exceed realistic construction. A developer may seek capacity in several markets before choosing one site. If utilities count every request, they can overbuild. If they discount too many, they can face shortages.

This is why financial commitments matter. Utilities and regulators can require deposits, construction milestones, minimum bills, or long-term contracts before investing for a large customer.

Special electricity rates for large loads are also spreading. These tariffs can require data centers to cover dedicated infrastructure and protect households from stranded costs.

The details determine whether that protection works. A minimum payment that expires too soon may not cover a transmission asset expected to last several decades. Confidential contracts can also make public oversight difficult.

Grid reliability is not simply a question of whether enough electricity exists nationally. Congestion can trap generation on one side of a transmission constraint while demand rises on the other.

The United States Geological Survey noted this location problem in its infrastructure analysis. Reliable power and fiber connectivity both shape where data centers can operate, while colocating generation introduces additional technical tradeoffs.

Communities experience these decisions locally. New lines require land, gas plants produce nearby emissions, and substations occupy large sites. Water use can also rise through cooling systems or thermal power generation.

The benefits are uneven as well. Data centers create construction work and tax revenue, but their permanent staffing can be modest relative to their electricity demand. Communities may question whether public infrastructure commitments match the local economic return.

Higher wholesale prices present another concern. When large loads tighten supply, expensive generators can operate more often. Capacity markets can also become costlier when the system needs additional plants reserved for future peaks.

However, blaming all rate increases on AI would oversimplify the issue. Aging infrastructure, extreme weather, fuel prices, electrification, manufacturing, and reliability investments also affect bills.

The more precise question is whether each data center pays the costs it creates. That includes dedicated equipment, additional generation, transmission upgrades, reserve requirements, and risks attached to uncertain forecasts.

Clear cost allocation can preserve support for projects that genuinely strengthen a region. Weak rules can transfer commercial risk from large technology companies to households and small businesses.

The Largest Risk Is Building Against the Wrong Forecast

The grid must expand, but the difference between the low and high demand cases represents hundreds of terawatt-hours and billions in infrastructure exposure.

Forecasting AI electricity use requires assumptions about hardware efficiency, model design, user demand, and the business value of AI services. Each variable can change faster than traditional utility planning expects.

New chips can perform more calculations per unit of electricity. Better cooling and power systems can also reduce facility overhead. These improvements lower the energy needed for a fixed amount of computing.

Efficiency can still increase total electricity consumption. Lower computing costs often encourage companies to train larger models, serve more users, and apply AI to more tasks. This response is sometimes called the rebound effect.

The amount of computation also depends on whether companies receive adequate returns. If AI revenue disappoints, some proposed campuses will be delayed or canceled. Strong adoption could push demand toward the upper end of current forecasts.

The National Academies summarized this uncertainty in its AI electricity workshop. Participants discussed estimates ranging from restrained growth to scenarios where data center energy use doubles or triples by 2030.

This uncertainty cuts in both directions. Underbuilding threatens reliability and economic development. Overbuilding can burden customers with unnecessary plants, pipelines, and power lines.

Utilities traditionally recover approved investments through rates over many years. That model works when demand grows steadily across many customers. It becomes riskier when one enormous customer drives a project.

Data centers can reduce that risk by signing contracts that survive delays, ownership changes, or lower utilization. Regulators can also require companies to fund dedicated assets upfront.

Forecast quality needs improvement as well. Utilities should distinguish early inquiries from projects with land, financing, equipment orders, and binding service agreements.

Transparency remains limited because developers consider power requirements commercially sensitive. Utilities may also withhold project details while negotiations continue.

That secrecy makes community review harder. Residents can see a new rate case or gas proposal without knowing how much demand is committed, which customer created it, or what alternatives were evaluated.

Climate claims deserve similar scrutiny. A company can announce renewable procurement while the host utility builds gas capacity to preserve reliability. Both statements can be technically accurate, yet neither describes the complete physical system.

Hourly matching offers a better view than annual accounting. It tests whether clean electricity is available when a facility actually operates.

Even hourly matching has limits. Contracting for existing clean generation can redirect that electricity from other customers without adding new supply. Stronger standards consider location, timing, and whether procurement causes new capacity to be built.

Natural gas faces its own uncertain assumptions. Developers may describe plants as temporary bridges or future hydrogen facilities. Those conversion pathways involve costs, fuel availability, equipment changes, and unresolved emissions questions.

Claims about future conversion should not justify present infrastructure without detailed technical and financial plans. A plant permitted for gas should be evaluated as a gas plant until a credible alternative is secured.

Solar and storage projections also require caution. Battery prices have fallen over time, but a system designed for several hours differs greatly from one covering multiple days.

The most resilient strategy is a portfolio. Solar, wind, batteries, nuclear power, geothermal resources, transmission, efficiency, and flexible computing can each address different parts of the problem.

A portfolio still needs disciplined contracts. Data center operators should carry the risks associated with their schedules and demand forecasts. Existing customers should not become default insurers for speculative AI expansion.

Three Signals Will Show Whether Power Supply Is Catching Up

The next stage will be decided by binding grid agreements, hourly operating behavior, and the actual generation entering service.

The first signal is the conversion of announced data center projects into enforceable electricity contracts. Signed agreements, deposits, and minimum payments reveal which campuses are moving beyond speculation.

A growing share of committed projects would strengthen the case for sustained electricity demand. Widespread delays, cancellations, or downsizing would weaken high-growth forecasts.

Investors and communities should watch the timing as closely as the headline capacity. A project scheduled for several phases creates a different grid challenge from one requesting its full load immediately.

The second signal is measurable demand flexibility. Pilot programs must show that data centers can reduce or shift meaningful loads during grid stress without breaking essential services.

The Electric Power Research Institute’s modeling suggests flexible operation can support more renewable generation and reduce long-term gas additions. Models are useful, but verified operating data matters more.

Utilities should publish aggregated results from flexibility programs. Useful figures include the megawatts available, response time, event duration, and the percentage of promised reductions actually delivered.

Successful performance would support a grid where computing follows electricity availability. Weak participation would reinforce the argument for additional firm generation.

The third signal is the physical supply mix connected before 2030. Announcements for nuclear reactors, geothermal systems, or long-duration storage do not provide electricity until projects receive permits, financing, equipment, and grid access.

Current evidence favors a mixed expansion. The IEA expects U.S. natural gas generation to grow alongside solar through 2030. Its electricity forecast projects substantial gas growth, while solar reaches 11 percent of U.S. generation by the decade’s end.

The speed of transmission construction will influence that balance. Faster interconnection and stronger regional lines make remote renewable resources more useful. Continued delays increase the appeal of local gas generation.

The underlying story is therefore larger than rising electricity consumption. U.S. AI data center demand is forcing technology companies to operate inside the constraints of energy infrastructure.

Software can scale globally in months. Power systems cannot. They require factories, fuel, land, permits, construction crews, and public approval.

That mismatch explains why gas and solar are growing at the same time. Both can enter service sooner than many alternatives, but they solve different parts of the electricity problem.

Gas provides controllable output while creating emissions and fuel risks. Solar supplies low-cost electricity while depending on storage, transmission, and complementary generation.

The grid sits between them. If planners modernize transmission, price flexibility correctly, and assign project costs carefully, AI demand can support a broader energy buildout.

If those measures lag, utilities will favor whatever source delivers firm power soonest. That path likely means more natural gas, even as technology companies purchase record volumes of renewable electricity.

The next few months should bring more utility filings, large-load tariffs, and data center power agreements. Readers should look past promised megawatts and ask three questions: Is the load contractually committed, can it respond during grid stress, and what generation will physically serve it?

Those answers will reveal whether AI infrastructure is financing a more flexible power system or locking communities into expensive capacity built around uncertain forecasts.