Artificial Intelligence and U.S. Electricity Demand: Trends and Outlook to 2040
AI is driving a surge in U.S. electricity demand, led by hyperscale data centers and cloud computing. By 2040, AI-related loads could double national power use. Utilities must adapt grid planning, clean energy deployment, and policy to align AI growth with decarbonization goals.
At the dawn of the internet age, few imagined that the flicker of server lights in early data centers would one day rival the electricity draw of entire states. Today, a new digital revolution is underway—driven not by web hosting or video streaming, but by artificial intelligence. From the clustered GPUs training large language models to edge devices parsing real-time data, AI is rapidly transforming how we compute—and how we consume power.
This article charts the trajectory of U.S. electricity demand through the lens of artificial intelligence, tracing the historical rise of computing power, the present surge in energy use by AI infrastructure, and the outlook through 2040. The story is not simply about servers and silicon, but about the electric grid itself—and whether it can adapt to meet the demands of a machine-learning future.
A Decade of Data: From Efficiency to Explosion
In the early 2000s, U.S. data centers consumed a modest share of national electricity—about 30 terawatt-hours (TWh) annually. But as internet traffic grew, so too did the demand for power. By 2007, the Environmental Protection Agency warned Congress that unchecked growth in server farms could overwhelm parts of the grid. And then, unexpectedly, the curve bent.
Between 2010 and 2014, efficiency gains in server hardware, cooling systems, and virtualization decoupled computing output from energy consumption. Data center electricity use grew just 4 percent during this period, despite a dramatic increase in processing demand. Analysts estimate that continued improvements through 2020 avoided over 600 billion kWh of electricity use.
But that reprieve has ended. From 2014 to 2023, total U.S. data center power consumption tripled, reaching roughly 176 TWh—about 4.4 percent of national electricity use. The culprit: exponential growth in digital services and, more recently, AI workloads.
AI Enters the Grid
Training a large AI model today can require as much electricity as hundreds of homes use in a year. When these models are deployed across millions of devices and applications, the cumulative load becomes staggering. The U.S. Department of Energy estimates that data center energy use could double or triple between 2023 and 2028—reaching 325 to 580 TWh. That would mean AI-related infrastructure alone could consume 10 percent or more of national power by decade’s end.
Utilities, long accustomed to flat or declining load growth, are scrambling to adapt. In some states, AI-related electricity demand now outpaces growth from electric vehicles or building electrification. Consulting firm Bain & Company found that data centers could account for nearly half of all new U.S. electricity demand between 2023 and 2028.
Long-term projections vary, but the arc is clear. By 2040, U.S. electricity consumption could rise by 50 to 75 percent compared to today, driven by the combined forces of AI, electrification, and industrial reinvention. In a high-AI scenario modeled by McKinsey, data centers alone could account for up to 15 percent of U.S. electricity use by 2040.
Sector Spotlight: Where the Power Goes
The AI electricity boom is not spread evenly. The bulk of growth is concentrated in hyperscale cloud providers—Amazon, Microsoft, Google—who are building 100-megawatt-class facilities across the country. These giants are deploying tens of thousands of GPUs and AI accelerators, housed in massive campuses with sophisticated power and cooling systems. For them, electricity is a controllable cost—often less than 20 percent of operations—and access to reliable power is a strategic imperative.
Meanwhile, smaller enterprise and industrial data centers are also evolving. Financial institutions and manufacturers are deploying AI locally for fraud detection, automation, and predictive analytics. While these facilities are smaller—1 to 5 megawatts—their proliferation adds new distributed loads to the grid.
The edge is also emerging. Telecom companies, autonomous vehicle platforms, and IoT applications are driving deployment of small AI inference centers in local networks. While individually modest, their collective impact introduces new complexity for distribution grids.
Regional Flashpoints
The data center boom is reshaping electricity demand maps across the country. Nowhere is this more apparent than Northern Virginia, where data centers consumed more than 25 percent of statewide electricity in 2023. Dominion Energy, the region’s largest utility, expects this load to continue growing at double-digit rates, requiring massive upgrades to substations, feeders, and even long-distance transmission.
Texas is another hotspot. Driven by competitive electricity pricing and abundant land, data centers and crypto mines have added gigawatts of load to the ERCOT grid. The result has been both boom and risk—grid operators have recorded multiple “near misses” where large, uncoordinated load tripping threatened stability.
Other states—from Iowa to Oregon, Georgia to Arizona—are courting data center investment with renewable energy incentives and infrastructure deals. The IEA reports that in some U.S. states, data centers already account for more than 10 percent of electricity use.
As siting pressures mount, companies are exploring new regions—often repurposing retired coal plant sites with pre-built transmission access. The Department of Energy has encouraged this strategy as a way to modernize energy communities while meeting digital infrastructure needs.
Load Shape and Utility Planning
Unlike residential or commercial loads that peak at certain hours, data centers typically draw power at a steady rate, 24/7. This high load factor flattens the grid’s demand curve, raising off-peak loads and potentially helping base-load generators. But it also reduces flexibility. During summer peaks or extreme weather events, data centers rarely curtail demand—requiring utilities to plan for their full capacity at all times.
Some AI workloads offer a path to flexibility. Training tasks, unlike real-time applications, can be scheduled for periods of renewable abundance. Google has experimented with “carbon-aware computing” that shifts tasks to hours when solar or wind is plentiful. If broadly adopted, this could transform data centers into grid assets rather than inflexible burdens.
Utilities are beginning to factor AI-driven loads explicitly into their integrated resource plans. Special tariffs, demand response programs, and co-investment in on-site storage are all under consideration. In some cases, data centers may use their backup generators or batteries to support the grid in emergencies—blurring the line between consumer and participant.
Uncertainty and Scenarios
Forecasting electricity demand from AI is fraught with uncertainty. Adoption rates, hardware efficiency, and workload distribution (cloud vs. edge) all affect outcomes. In optimistic efficiency scenarios, advanced chips and liquid cooling could dramatically reduce energy per computation. In high-growth scenarios, these savings are overwhelmed by the sheer scale of new AI applications.
Flexible demand models, where data centers throttle during grid stress, show promise but remain rare. And while some loads may shift to edge devices, the current trajectory strongly favors centralized cloud processing—especially for large model inference and training.
Policymakers and planners are now treating data center growth as a primary load scenario, not a marginal variable. Utilities are stress-testing their systems against aggressive growth cases, modeling what happens if 500 or 1000 megawatts of unexpected demand materialize in a region. The planning stakes are high—and the lead time for grid upgrades is long.
Climate Collision—or Catalyst?
Perhaps the biggest question is how AI’s electricity appetite intersects with climate goals. The U.S. power sector is under pressure to decarbonize, yet the rise of new loads—especially constant, non-deferrable loads—could complicate that trajectory.
Some worry that AI growth will force utilities to delay fossil retirements or build new gas plants. Others see opportunity. The largest hyperscalers have committed to 100 percent renewable energy, often via power purchase agreements that underwrite new solar and wind farms. In 2023 alone, tech companies were among the top five buyers of clean energy globally.
There’s also renewed interest in carbon-free firm power. Microsoft and Amazon are exploring nuclear options—including small modular reactors—to supply data centers. DOE is encouraging this pairing, viewing AI infrastructure as a potential customer base for next-generation zero-carbon power.
On-site energy storage, smarter cooling, and even waste heat reuse are part of the emerging toolkit. And AI itself may help decarbonize the grid—through better forecasting, demand response algorithms, and optimization of renewables and batteries.
Conclusion: Bits and Electrons
AI is not just changing how we think, work, or communicate—it is transforming the physical infrastructure beneath our economy. The electric grid, once shaped by homes and factories, must now accommodate a rising tide of servers, accelerators, and distributed intelligence.
This transformation carries risks: unplanned load growth, infrastructure constraints, carbon emissions. But it also offers a chance to rethink how power is produced, managed, and consumed. AI-driven demand can be a problem—or a lever for innovation—depending on how we respond.
The coming decades will test our ability to plan wisely, invest strategically, and align the goals of the digital economy with those of a decarbonized grid. One thing is certain: the story of electricity in the 21st century will be written not just in kilowatts, but in code.
If you found this analysis valuable…
The convergence of AI and energy is accelerating, and the headlines only scratch the surface. AI × Energy delivers independent, deeply researched insights at the frontiers of technology, infrastructure, and policy—without hype, paywalls, or partisanship.
To stay ahead of the signal, not the noise, subscribe to receive the latest articles, opinion pieces, and in-depth research reports. The future grid is being built now. Know what’s coming.