Wired for Power
This is the first in a series of five articles on the intersection of energy and AI, culminating in an in-depth research report on the topic at the end of April 2025. This first article touches upon AI's current and expected future impact on energy demand.
Author's note: This is the first in a series of five articles on the intersection of energy and AI, culminating in an in-depth research report on the topic at the end of April 2025. This first article touches upon AI's current and expected future impact on energy demand.
How Artificial Intelligence is Driving America’s Energy Future
Artificial intelligence is no longer confined to the realm of digital assistants or self-driving car prototypes—it is now reshaping something far more fundamental: the American power grid. In just a few years, AI has transformed from a promising computational tool to one of the most formidable drivers of electricity demand in the United States. Behind every ChatGPT response, image generation tool, fraud detection algorithm, or real-time delivery system is an invisible engine humming with electrons—vast data centers, many the size of small cities, drawing power at a scale that rivals entire states.
This article presents a comprehensive narrative of how AI is driving U.S. electricity consumption, based on the latest research from the Department of Energy (DOE), Lawrence Berkeley National Laboratory (LBNL), the International Energy Agency (IEA), utilities, and energy consultancies. It traces the evolution of AI computing load from the early days of data centers to the hyperscale era, explores regional and sector-specific implications, examines the shifting load profile of the grid, and analyzes how this revolution is intersecting with—and challenging—the clean energy transition.
From Servers to Superclusters
To understand the power implications of AI, we begin with computing’s energy evolution. At the dawn of the 21st century, U.S. data centers consumed a modest share of national electricity, but this changed rapidly with the internet boom. Between 2000 and 2005, electricity use in data centers nearly doubled, rising by 90% amid explosive growth in server installations and digital services (Data Center Knowledge, 2016). By 2007, the EPA warned Congress that without intervention, data center demand could place serious strain on the national grid (EPA, 2007).
Then came a pause. Between 2010 and 2014, thanks to breakthroughs in virtualization, improved IT hardware, and energy-efficient cooling systems, total U.S. data center electricity use grew just 4%, reaching around 70 TWh—a striking slowdown compared to prior decades. These innovations saved an estimated 620 billion kWh of electricity between 2010 and 2020 compared to a no-efficiency-improvement scenario (Data Center Knowledge, 2016).
But the calm didn’t last. With the rise of cloud computing in the late 2010s and the emergence of AI workloads in the early 2020s, the curve began climbing again. From 2014 to 2023, U.S. data center electricity consumption tripled—from 58 TWh to 176 TWh—accounting for about 4.4% of all U.S. power use by 2023 (DOE, 2024). Major firms like Google and Microsoft more than quadrupled their electricity use from 2016 to 2023, largely due to expanded AI activity (Federal Reserve Bank of Kansas City, 2024).
AI Ascendant
Looking ahead, the forecasts are staggering. The U.S. DOE projects that electricity demand from data centers could rise to between 325 and 580 TWh annually by 2028, accounting for 6.7% to 12% of national power use (DOE/LBNL, 2024). Bain & Company estimates data centers will drive nearly half (44%) of all load growth between 2023 and 2028, requiring 7% to 26% more generation capacity across the grid in just five years (Utility Dive, 2024). These projections have forced a rethink of what the next decade will look like for utilities long accustomed to flat or declining demand.
McKinsey & Company projects that hyperscale AI infrastructure will grow electricity consumption by 400 TWh by 2030, with annual data center load growth averaging 23% through the decade. By then, AI-related loads may account for 30–40% of all incremental electricity demand (McKinsey, 2024). The IEA adds a global perspective: worldwide data center and crypto loads are expected to more than double from 460 TWh in 2022 to over 1,000 TWh by 2026 (IEA, 2024). The United States—home to the highest concentration of hyperscale facilities—is expected to lead this growth.
The U.S. Energy Information Administration (EIA) has had to revise its forecasts upward. Its 2025 load outlook increased eightfold in just a year, reflecting the AI surge (Kansas City Fed, 2024). This is not simply a new chapter in energy forecasting—it is a different book altogether.
Beyond 2030
Projections to 2040 show a wide range, but all suggest robust demand growth. Under a high-AI adoption scenario, total U.S. electricity consumption could reach 6,900 TWh—up from approximately 3,900 TWh today, a 75% increase (McKinsey, 2024). Pacific Gas & Electric foresees California’s power demand doubling by 2040, citing AI, electrified transport, and building decarbonization as primary drivers (Bloomberg Law, 2024).
The National Renewable Energy Laboratory (NREL) found that digitalization could raise electricity intensity of the U.S. economy by over 20%, reversing decades of decline (Kansas City Fed, 2024). EPRI scenarios estimate that AI-related data centers could consume up to 15% of U.S. power by 2040 in high-growth cases (Visual Capitalist, 2024). Even conservative forecasts foresee dozens of gigawatts of firm demand added to the grid—an outcome with profound implications for planning, reliability, and decarbonization.
Sectoral Dynamics
While not all AI loads are created equal, hyperscale data centers are undeniably the primary driver. Amazon, Google, Microsoft, and Meta are building computing campuses that consume 100 to 150 MW apiece—enough to power hundreds of thousands of homes. Each is deploying tens of thousands of high-performance AI accelerators (GPUs and ASICs) to train and deploy next-gen models (IEA, 2024). In 2023 alone, their collective capital investment in data infrastructure surpassed the entire U.S. oil and gas industry (IEA, 2024).
Enterprise data centers—those operated by corporations and institutions—have largely stagnated in energy growth. Many have migrated workloads to the cloud, while others have improved their efficiency or downsized. Financial firms and defense agencies still maintain on-premise AI clusters for latency-sensitive and secure tasks, but these facilities are small—rarely exceeding a few MW—and collectively represent a shrinking portion of the load (Data Center Knowledge, 2016).
Industrial AI—used in manufacturing, oil and gas, and R&D—is a growing but smaller category. These facilities, typically in the 1–5 MW range, use AI for seismic analysis, predictive maintenance, or autonomous systems development. As AI is embedded deeper into industrial operations, these loads are likely to rise gradually—though they will remain dwarfed by hyperscale consumption (McKinsey, 2024).
Then there’s the edge: micro data centers deployed near telecom towers, retail hubs, and industrial parks. These sites—often under 1 MW—are proliferating with 5G and the internet of things. Though small individually, they add distributed complexity to the grid. Analysts expect thousands of such sites to emerge by 2030, reshaping distribution networks and creating new demand for localized infrastructure upgrades (Goldman Sachs, 2024).
Where AI Lands
The AI boom is not evenly distributed. In Virginia, the Northern Virginia region—dubbed “Data Center Alley”—is home to the highest density of server farms on Earth. In 2023, data centers consumed 25–26% of Virginia’s total electricity. Dominion Energy expects 12% annual load growth in its service territory through 2035, driven almost entirely by these facilities (Visual Capitalist, 2024).
Texas, especially Central Texas and the Dallas-Fort Worth area, has also seen a surge in data center and crypto mining development. From 2019 to 2023, commercial power use in Texas rose by 13 billion kWh, a 10% jump largely due to large-scale computing (Data Center Dynamics, 2024). The Electric Reliability Council of Texas (ERCOT) has updated its interconnection requirements and implemented emergency procedures to account for sudden 1,000+ MW load swings when data centers or crypto mines disconnect unexpectedly (Reuters, 2024).
Other states seeing rapid growth include Iowa, Oregon, Georgia, Arizona, and even North Dakota, where one new campus drove a 37% spike in statewide electricity use from 2019 to 2023 (Data Center Dynamics, 2024). Siting criteria often include access to cheap electricity, transmission capacity, tax incentives, and cool climates. Increasingly, companies are repurposing coal plant sites—with existing grid ties and water infrastructure—for AI campuses (DOE, 2024).
The Load Profile Revolution
AI data centers are not just growing—they are changing the shape of the load curve. These facilities operate nearly 24/7 at full capacity, unlike traditional loads which peak during business hours or weather events. As a result, regional load curves are flattening. In Northern Virginia, off-peak demand is now higher than ever due to data center baseload consumption (Kansas City Fed, 2024).
This has implications for both operations and planning. A flatter curve can be beneficial: base-load generation assets like nuclear and combined-cycle gas plants can run more efficiently. But the inflexibility of AI loads presents risks during peak events. Without participation in demand response programs, data centers compound peak stress rather than relieve it.
Some operators are experimenting with “carbon-aware computing”—shifting non-urgent AI tasks to periods of high renewable availability. Google, for instance, defers batch jobs to align with solar or wind output (DOE, 2024). These efforts are early but promising, pointing toward a future where data centers become flexible grid assets rather than passive loads.
Energy Innovation on the Grid Edge
The rise of AI is prompting a wave of innovation across the power sector. Utilities are introducing special tariffs for high-load customers, incorporating AI-specific growth into integrated resource plans, and experimenting with virtual power plants that include data centers. The U.S. DOE is promoting the concept of “grid-interactive” data centers that use onsite storage, backup generation, and smart controls to support grid stability (DOE, 2024).
Some centers already feature battery-based UPS systems that could be upsized to provide peak-shaving or frequency support. There is even discussion of temporarily using backup diesel or gas turbines at data centers during emergencies—though such solutions come with emissions trade-offs.
Meanwhile, AI is helping to solve the very problems it creates. Utilities are deploying AI tools for renewable forecasting, smart EV charging, outage detection, and real-time grid balancing. As the IEA noted, digitalization and AI have “enormous potential” to accelerate the clean energy transition—if deployed wisely (IEA, 2024).
Decarbonization at a Crossroads
The elephant in the server room is carbon. If AI-driven demand is met with fossil fuel generation, it could jeopardize state and federal climate targets. Already, some utilities are delaying coal retirements or adding natural gas peaking capacity in anticipation of surging data center loads (Utility Dive, 2024).
But there is hope. Hyperscalers are leading the world in renewable energy procurement. Amazon, Microsoft, Google, and Meta are the largest buyers of wind and solar PPAs globally. Some are going further: Amazon is investing in small modular reactors (SMRs), while Microsoft has inked deals to buy power from Pennsylvania’s nuclear fleet (Puget Sound Business Journal, 2024).
In tandem, some are exploring geothermal, green hydrogen, or advanced thermal storage as solutions for 24/7 clean power. The DOE is actively supporting these technologies, recognizing the AI revolution as both a risk and an opportunity for decarbonization (DOE, 2024).
Electrifying Intelligence
Artificial intelligence is poised to become the most transformative force in U.S. electricity demand since the electrification of industry in the 20th century. The load it introduces—unrelenting, widespread, and fast-growing—requires a fundamental rethinking of how we plan, build, and operate the electric grid. If managed wisely, the AI era can support not just smarter machines, but a cleaner and more resilient power system.
But the stakes are high. If ignored or mishandled, AI could set back decarbonization by a decade. Energy professionals must treat digital load growth not as a footnote, but as a central pillar of the future power system. For the United States, the choice is clear: either build a grid ready for intelligence, or be overwhelmed by the very tools we’ve unleashed.
Bibliography
Data Center Knowledge. “U.S. Data Centers Energy Consumption.” 2016. https://www.datacenterknowledge.com.
DOE (U.S. Department of Energy) and LBNL (Lawrence Berkeley National Laboratory). “2024 Report on U.S. Data Center Energy Use.” 2024. https://www.energy.gov.
EPA. “Report to Congress on Server and Data Center Energy Efficiency.” 2007. https://www.epa.gov.
Federal Reserve Bank of Kansas City. “Powering Up: The Surging Demand for Electricity.” 2024. https://www.kansascityfed.org.
IEA. “What the Data Centre and AI Boom Could Mean for Energy.” 2024. https://www.iea.org.
McKinsey & Company. “How Data Centers and the Energy Sector Can Sate AI’s Hunger for Power.” 2024. https://www.mckinsey.com.
Utility Dive. “Data Center Load Growth and Utility Bills.” October 2024. https://www.utilitydive.com.
Visual Capitalist. “Data Center Electricity Consumption by State.” November 2024. https://www.visualcapitalist.com.
Environment America. “AI Surge and Energy Demand.” October 2024. https://www.environmentamerica.org.
Puget Sound Business Journal. “Amazon Web Services – Nuclear Energy Project News.” October 2024. https://www.bizjournals.com.
Reuters. “Big Tech’s Data Center Boom Poses New Risk to US Grid.” March 2025. https://www.reuters.com.
Goldman Sachs. “U.S. Data Center Capacity Forecast.” 2023. https://www.goldmansachs.com.