Artificial intelligence has created an extraordinary appetite for computing power. Technology companies are investing heavily in advanced semiconductors, expanding data centers, and developing increasingly capable models. Yet the industry's growth is becoming dependent on something far more fundamental than computing technology itself.
Electricity.
The most advanced AI processors are of little use without the power needed to operate them. As data centers grow larger and more energy-intensive, securing sufficient electricity is becoming an increasingly important consideration in where facilities are built, how quickly they can begin operating, and how much they ultimately cost.
The challenge is that AI infrastructure can expand faster than the electricity systems needed to support it. Servers can be installed in months, but constructing power plants, upgrading transmission networks, and securing grid connections can take years.
This mismatch is transforming electricity from a routine operating expense into a strategic resource for the AI industry.
AI Is Driving a Sharp Increase in Electricity Demand
The scale of the challenge is becoming clearer.
According to the International Energy Agency's (IEA) April 2026 report, Key Questions on Energy and AI, global data center electricity consumption reached an estimated 485 terawatt-hours (TWh) in 2025, an increase of 17% from the previous year.
Electricity consumption at AI-focused data centers grew considerably faster, increasing by approximately 50% in 2025.
The IEA projects that total global data center electricity consumption will nearly double to 950 TWh by 2030, representing approximately 3% of worldwide electricity demand. Within that forecast, consumption at AI-focused facilities is expected to triple between 2025 and 2030.
These figures cover the entire data center industry, including conventional cloud computing, storage, and digital services. They should not be interpreted as electricity consumption attributable exclusively to artificial intelligence.
The estimates reflect the IEA's modeling of computing infrastructure, technology trends, and energy requirements. The 2030 figures are projections rather than electricity demand already realized or guaranteed to materialize.
Even so, the projected increase is substantial. Unlike some traditional industrial activities, AI computing can require enormous quantities of electricity concentrated within relatively small geographic areas.
That concentration makes access to power particularly important for data center developers.
The United States Faces an Expanding Power Challenge
The United States is experiencing some of the strongest pressure from data center development.
In June 2026, Lawrence Berkeley National Laboratory published its United States Data Center Energy Usage Report: 2025 Update.
Its reference scenario estimates that American data centers could consume 649 TWh of electricity annually by 2030, equivalent to approximately 11.8% of total U.S. electricity consumption.
The report also presents scenarios in which the industry's share of national electricity consumption ranges from 9.5% to 15.3% by 2030.
These are modeled outcomes rather than confirmed future consumption. Researchers developed the estimates using information about planned computing equipment shipments, expected electricity usage, operational characteristics, and cooling requirements.
The IEA's separate Electricity 2026 report estimates that data centers will account for approximately half of the growth in U.S. electricity demand through 2030.
The implications extend beyond technology companies. Electricity utilities must prepare for substantial additional demand while maintaining reliable service to households, manufacturers, hospitals, and other businesses.
The difficulty becomes particularly acute where data centers are concentrated. A national electricity system may possess substantial generating capacity while individual regions face shortages of available power or transmission infrastructure.
For AI developers, choosing a location is therefore becoming as much an electricity decision as a technology or real estate decision.
Power Grids Cannot Expand as Quickly as Data Centers
One of the most significant obstacles facing AI infrastructure is the time required to develop electricity networks.
The IEA's February 2026 Electricity 2026 report estimates that planning, permitting, and completing new grid infrastructure can take five to 15 years. By comparison, new data centers can be constructed in approximately one to three years.
The difference creates a fundamental timing problem.
A technology company may have financing, land, equipment, and construction partners ready, yet remain unable to operate a new facility at its intended capacity because the surrounding electricity network cannot deliver sufficient power.
The IEA also reports that more than 2,500 gigawatts of projects involving renewable generation, energy storage, and large electricity consumers are waiting in grid connection queues worldwide. This is a broader electricity-system figure, not a measurement of data center demand alone.
The consequences are already visible in the United States.
In July 2026, PJM Interconnection, the regional electricity grid operator serving parts of 13 states and the District of Columbia, reported that capacity secured for the 2028–2029 delivery year was 6,831 megawatts below its reliability requirement.
The shortfall refers to accredited generating and demand-response capacity relative to PJM's reserve target. It does not mean that an equivalent amount of electricity is currently unavailable or that power outages are inevitable.
PJM identified rapidly increasing demand, including large data center loads, as an important factor in the imbalance. In October 2026, it submitted an amended proposal for additional capacity procurement, with costs intended to be allocated to the large electricity loads driving the need for new resources.
The developments demonstrate why access to the grid is becoming a commercial consideration in the AI industry. A facility's ability to obtain a dependable power connection can influence its development timetable and economic viability.
Technology Companies Are Becoming Major Energy Buyers
The growing importance of electricity is also changing how technology companies secure their energy supplies.
Rather than relying exclusively on conventional utility arrangements, major technology companies are signing long-term electricity purchase agreements and supporting additional generating capacity.
The IEA reported in April 2026 that the technology sector accounted for approximately 40% of corporate renewable electricity purchase agreements signed in 2025. The industry has also become an important source of demand for nuclear power and advanced geothermal projects.
One example came in October 2026, when Google announced an agreement with Constellation Energy intended to add 890 megawatts of nuclear generating capacity to the PJM electricity system.
The arrangement supports upgrades at six existing nuclear plants in Illinois, Pennsylvania, and New Jersey. According to Google, the agreement brings the additional nuclear capacity it has supported through plant upgrades and restarts to more than 1.5 gigawatts.
These announcements concern planned or supported generating capacity, not electricity already delivered to Google's data centers. A megawatt measures power capacity, while a terawatt-hour measures electricity consumed or produced over time.
Nevertheless, the agreements illustrate a change in corporate priorities.
Reliable electricity supplies are no longer simply an expense to be negotiated with utilities. Technology companies are becoming more directly involved in supporting the infrastructure that produces electricity.
This development reflects the different requirements of AI computing. Data centers need dependable power throughout their operating lives, including periods when solar and wind generation may be limited.
Renewable electricity can provide significant energy supplies, but facilities also need arrangements that maintain reliable operations when generation and demand do not coincide.
That requirement is helping sustain interest in nuclear power, energy storage, geothermal generation, and natural gas.
Building More Power Is Not a Simple Solution
Producing additional electricity is only part of the problem. It must also be delivered where and when data centers require it.
New power plants may face permitting delays, equipment shortages, financing challenges, or lengthy connection processes. Even where electricity generation is available, transmission infrastructure may be insufficient.
Some developers are consequently exploring electricity generation directly at data center sites.
The IEA's April 2026 assessment projects that approximately 15 to 27 gigawatts of onsite natural gas generation could serve data centers by 2030, primarily in the United States.
However, the agency emphasizes considerable uncertainty around these projects. Its satellite-based tracking found that only about one-fifth of identified onsite gas projects had begun land clearing or construction.
Reliable onsite gas generation also presents technical challenges. According to the IEA's modeling, meeting variable data center loads can require generation capacity 30% to 70% above the facility's electricity demand.
These findings suggest that moving electricity generation closer to computing facilities does not automatically eliminate infrastructure constraints.
The supply of turbines, electrical equipment, fuel, and backup systems remains essential.
There are environmental trade-offs as well. Additional fossil-fuel generation can increase emissions, while low-emission alternatives must be developed and integrated into electricity systems quickly enough to meet growing demand.
Greater Efficiency May Not Reduce Total Electricity Consumption
One of the defining features of AI development is the rapid improvement in computing efficiency.
Advances in hardware and software mean that individual AI tasks can require substantially less electricity than they did previously.
The IEA's 2026 research finds that energy consumption per task has fallen dramatically in recent years.
Yet the overall demand for AI computing continues to increase.
The explanation lies partly in the expanding range of applications. Simple text generation may require relatively little electricity, while complex reasoning, video generation, and AI agents performing multiple operations can be far more energy-intensive.
The IEA reports that some of these applications can consume hundreds or thousands of times more electricity per query than simple text generation.
Consequently, the relationship between efficiency and total electricity consumption is not straightforward.
Making an individual task less energy-intensive does not guarantee that aggregate demand will decline if the number, complexity, and duration of tasks increase substantially.
The industry's electricity requirements will depend on how quickly efficiency improves, how widely AI is adopted, and which applications become commercially important.
This uncertainty also complicates electricity planning. Utilities must invest in infrastructure that may operate for decades while serving an industry whose technologies and computing requirements are changing rapidly.
The Cost of AI's Electricity Demand Extends Beyond Technology Companies
The growing demand for electricity raises an important economic question.
Who should pay for the infrastructure required to support the expansion of AI?
New data centers can generate investment, employment, and additional revenue for electricity providers. Under suitable conditions, large and predictable electricity customers can also help utilities spread fixed infrastructure costs across greater consumption.
However, the outcome depends heavily on local electricity supply, network capacity, regulatory arrangements, and the allocation of investment costs.
The IEA's 2026 assessment concludes that data centers do not necessarily increase electricity prices. Where spare capacity exists, additional demand can improve infrastructure utilization.
Where electricity supplies are already constrained, however, rapidly growing demand may require expensive new generation and grid upgrades. Those investments can place upward pressure on electricity prices if costs are not allocated appropriately.
The issue has attracted regulatory attention.
In June 2026, the U.S. Federal Energy Regulatory Commission directed six regional grid operators to justify or reform rules governing the connection of data centers and other large electricity consumers. The action included measures addressing transparency, transmission costs, and the prevention of inappropriate cost shifting.
For policymakers, the challenge is to accommodate a rapidly growing industry without compromising electricity affordability or reliability for existing customers.
For technology companies, the challenge is increasingly about obtaining electricity under commercially sustainable arrangements, rather than simply finding the lowest available price.
Electricity Is Becoming a Competitive Advantage
For years, the competitive landscape of artificial intelligence has been shaped by access to advanced chips, technical expertise, computing infrastructure, and investment capital.
Those resources remain essential. Electricity is not replacing them, but it is becoming an equally important constraint on the industry's expansion.
The difference is that electricity cannot be considered independently of location, network infrastructure, and the time required to develop new generating capacity.
A company may possess the most advanced computing hardware available yet still face delays if it cannot secure sufficient power.
Conversely, reliable electricity access in a suitable location can make it easier to develop and operate the infrastructure needed for AI services.
The commercial importance of electricity therefore extends beyond its price per kilowatt-hour. It includes reliability, connection capacity, infrastructure development costs, and the ability to secure power for many years.
The result is a growing convergence between two industries that have historically operated on very different timescales.
Technology companies are accustomed to rapid innovation and frequent equipment upgrades. Electricity systems require extensive planning, major capital investment, and infrastructure designed to operate for decades.
The next phase of AI development will depend on how effectively these two systems can expand together.
Artificial intelligence may be built with algorithms and semiconductors, but its continued growth will also depend on power plants, transmission lines, transformers, and the electricity networks connecting them.
The companies and regions that can secure reliable, affordable power will be better positioned to translate advances in AI computing into functioning infrastructure.
In an industry increasingly defined by the scale of its computing ambitions, electricity is becoming one of its most consequential competitive resources.




