AI’s Energy Appetite and the Challenge of Crafting an Energy Transition
The explosive growth of artificial intelligence and the urgent shift toward renewable energy are colliding in ways both troubling and transformative.
As AI data centers proliferate across the globe, consuming electricity at unprecedented rates, they simultaneously threaten the renewable energy transition.
The numbers surrounding AI’s energy consumption are significant indeed.. According to the International Energy Agency, data centers consumed approximately 415 terawatt-hours of electricity in 2024, representing about 1.5% of global electricity consumption. But this figure masks a more concerning trajectory: data center electricity consumption has grown at 12% annually over the past five years and is projected to more than double by 2030, reaching around 945 TWh in the IEA’s base case scenario—just under 3% of total global electricity consumption.
Within the United States, data center power demand is expected to surge from 35 gigawatts to 78 gigawatts by 2035, accounting for 8.6% of the nation’s electricity consumption. Goldman Sachs Research forecasts an even more dramatic trajectory, projecting that global power demand from data centers will rise 165% by 2030 compared with 2023 levels.
These projections aren’t abstract for they’re already reshaping regional power grids. In Northern Virginia’s “Data Center Alley,” utilities forecast that peak demand will rise by more than 75% by 2039 if data center growth continues at current rates, compared to just 10% growth without them. In Texas, data centers have become unequivocally the largest source of new power consumption, with demand projected to double by 2035.
The human cost of this expansion is becoming visible in electricity bills. A recent Bloomberg analysis revealed that wholesale electricity prices have increased as much as 267% in areas near data centers compared to five years ago.6 In Baltimore, residents have experienced an average monthly electric bill increase of over $17 in the last year, with expectations of another $4 jump in 2026.6 More than 70% of locations recording significant price increases are within 50 miles of major data center activity.6
Despite lofty corporate sustainability pledges, the reality on the ground paints a sobering picture. Currently, coal remains the biggest single electricity source for data centers globally, largely due to numerous facilities in China, while fossil fuels overall provide nearly 60% of power to data centers worldwide. This dependency on carbon-intensive energy sources threatens to undermine decades of climate progress.
The situation may worsen before it improves. Goldman Sachs Research forecasts that approximately 60% of the increasing electricity demands from data centers will be met by burning fossil fuels through 2030, potentially increasing global carbon emissions by about 220 million tons. To put this in perspective, driving a gas-powered car for 5,000 miles produces about 1 ton of carbon dioxide—meaning data centers could add the equivalent of 220 million such journeys annually.
In practice, data center expansion is already being used to justify prolonged reliance on fossil fuels. In Virginia, utility company Dominion Energy has delayed the retirement of fossil fuel power plants, including the Clover Power Station, citing projected energy demand from data centers. Similar patterns are playing out across the Southeast, where utilities in Virginia, Georgia, North Carolina, and South Carolina have proposed building 20,000 megawatts of new gas power plants by 2040, with data centers responsible for at least 65% of projected load growth in Virginia, Georgia, and South Carolina.
Some regions have even seen a resurgence in coal consumption. The U.S. coal industry experienced increased demand in 2025, with utilities running their plants harder, a reversal of the decades-long trend of declining coal use. Data center energy needs, combined with concerns about grid reliability, have contributed to this unexpected coal revival.
Natural gas has emerged as the supposed “bridge fuel” for data centers, but its environmental credentials are questionable at best. While natural gas produces less carbon dioxide per kilowatt-hour than coal, approximately 430 grams compared to coal’s 1,000 grams, it still emits more than ten times the emissions of solar power and forty times the emissions of wind or nuclear energy. Moreover, the extraction and distribution of natural gas leak methane, a greenhouse gas 86 times more potent than carbon dioxide over a 20-year period.
Gas-powered generation dedicated to data centers is expected to more than double from 120 TWh in 2024 to 293 TWh by 2035, with much of this growth concentrated in the United States.7 About 38 gigawatts of captive gas plants currently in development—roughly a quarter of all such projects—are specifically planned to power data centers.
Despite these challenges, the AI boom is simultaneously driving unprecedented investment in renewable energy infrastructure. The technology sector has emerged as a dominant force in clean energy procurement, consistently accounting for more than 68% of U.S. corporate renewable energy deals tracked over recent twelve-month periods. In 2024 alone, Google signed contracts to purchase approximately 8 gigawatts of clean energy generation capacity—more than in any prior year in the company’s history.
This corporate demand is creating powerful economic incentives for renewable energy deployment. Without the purchase commitments from tech companies and hyperscalers, many renewable energy projects would lack the financial viability to proceed. Microsoft announced a monumental $10 billion renewable energy deal with Brookfield Asset Management, with Brookfield set to deploy more than 10.5 gigawatts of renewable energy capacity beginning in 2026—equivalent to the output of ten nuclear power plants.
The scale of these commitments is transforming energy markets. Google’s partnership with Intersect Power and TPG Rise Climate aims to build industrial parks with gigawatts of data center capacity co-located with new clean energy plants, with the first phase expected to be operational by 2026. This approach addresses grid constraints by developing renewable energy and data centers as integrated infrastructure projects rather than separately.
These investments are accelerating the broader renewable energy transition. The momentum from data center demand is helping overcome supply chain constraints and driving technological innovation. AI itself is being leveraged to optimize renewable energy supply chains and accelerate operational efficiencies. Companies are using AI-powered robots to install large solar deployments in deserts specifically to power data centers, dramatically increasing deployment speed.
The fundamental tension between AI’s needs and renewable energy lies in their temporal mismatch. Data centers require reliable, 24/7 power to maintain uninterrupted operations and protect expensive hardware—AI-specific equipment like GPUs and TPUs has become increasingly costly and cannot tolerate power fluctuations. In contrast, solar and wind energy are inherently intermittent, producing power only when the sun shines or wind blows.
This variability creates what energy planners call “the duck curve” or sharp imbalances between renewable energy production and consumption that make these sources difficult to control and integrate into grid systems. For data centers housing sensitive computing equipment worth millions of dollars, any power disruption can result in significant downtime, hardware damage, and catastrophic data loss.
Battery Energy Storage Systems have emerged as the critical technology bridging this gap. These systems store excess renewable energy during high-production periods and discharge it during low-production periods, effectively time-shifting renewable generation to match constant data center demand. Lithium-ion batteries currently dominate the storage landscape, but alternative technologies including flow batteries, nickel-zinc systems, and emerging solid-state batteries are gaining traction for their improved safety profiles and longer duration capabilities.
The scale of storage deployment is accelerating rapidly. The global liquid cooling market for data centers is expected to reach $17.8 billion by 2027, with battery storage playing an integral role.2 Leading data center operators are already implementing battery systems that provide one to two hours of energy storage to stabilize the grid, reduce diesel generator needs, lower energy costs, and enable better utilization of renewables while maintaining availability.
Advanced implementations are even more sophisticated. Singapore’s SP Group employs AI to manage one of Southeast Asia’s most advanced urban energy grids, with AI-powered battery storage systems dynamically adjusting supply in real time and reducing peak demand stress by 15%. China’s Dalian Flow Battery Energy Storage Peak-shaving Power Station, a 100-megawatt facility, leverages AI-driven analytics to optimize charge and discharge cycles, ensuring efficiency and longevity.
These storage systems serve multiple functions beyond simply addressing intermittency. They provide backup power during grid outages, offer seamless continuity unlike diesel generators that require several minutes to start, and enable data centers to participate in demand response events and provide grid services like frequency regulation and voltage stabilization. Some facilities are exploring battery storage systems with wide-duration capabilities ranging from 2 to 110 hours of discharge, allowing them to smooth out renewable power source intermittency while ensuring stable electricity supply even during extended lulls in production.
Perhaps the most intriguing dimension of this relationship is AI’s potential to solve the very problems it creates. Machine learning algorithms are increasingly being deployed to optimize grid operations, predict renewable energy output, and manage the complex integration of distributed energy resources.
Research from Argonne National Laboratory demonstrates how AI can revolutionize grid maintenance by analyzing vast amounts of sensor data throughout the power network, creating predictive models that forecast wear and tear over time and recommend repairs before problems occur. This capability is crucial as the U.S. power grid with more than 240,000 high-voltage transmission lines and 50 million transformers, 70% of which have been in service for 25 years or more—faces increasing strain from growing load and volatile renewable energy integration.
AI-powered forecasting has achieved remarkable accuracy in predicting renewable generation. Shanghai’s energy forecasting models predict power generation fluctuations with 95% accuracy, enabling more precise grid balancing and reducing transmission losses through optimized distributed energy storage. Research from Argonne National Laboratory and MISO showed that machine learning models can perform daily grid planning calculations 12 times faster than traditional methods, reducing calculation time from nearly 10 minutes to just 60 seconds.
The National Renewable Energy Laboratory is examining ways to use generative AI to revolutionize the power grid by providing proactive decision support and predictive online control to improve efficiency, reliability, and resilience. Machine learning models are being developed to address the optimal power flow problem—determining the most cost-effective way to generate and distribute electricity while managing the uncertainty of renewable energy generation and electric vehicle demand.
Studies demonstrate measurable impacts from these AI applications. Research published in peer-reviewed journals found that machine learning optimization led to a 15% improvement in grid efficiency and a 10-20% increase in battery storage efficiency, with Random Forest algorithms reducing prediction error by approximately 8.5%. The research concluded that machine learning can significantly enhance renewable energy system performance, potentially helping close the “ambition gap” by 20% in supporting global efforts to meet the 1.5°C Paris Agreement targets.
Grid operators are already integrating these technologies into critical operations. The Department of Energy recognizes AI’s transformative potential, recently awarding $3 billion in grants to various smart grid projects that include AI-related initiatives.26 The agency is also developing AI tools to improve how energy projects are sited and permitted at federal, state, and local levels through initiatives like PolicyAI at Pacific Northwest National Laboratory.
Beyond technological innovations, the AI industry is exploring geographic and operational flexibility to better align with renewable energy availability. The reality is that some AI workloads, particularly training large language models, can be scheduled flexibly or relocated to areas where renewable power is abundant and cheap.
Time-shifting represents another strategic approach. AI training workloads that don’t require real-time processing can be scheduled to run when renewable energy generation peaks—during sunny afternoon hours for solar-powered facilities or windy periods for wind-powered infrastructure. Alphabet is piloting demand response methods that allow reduced data center power demand during periods of grid stress by shifting computing tasks to alternative times and locations.15
This operational flexibility could transform data centers from grid burdens into grid assets. The Department of Energy’s strategy explicitly calls for “enabling data center flexibility through onsite power generation and storage solutions” to help stabilize grids rather than strain them. With proper controls and storage infrastructure, data centers can absorb excess renewable energy when it’s plentiful, store it, and even provide grid services during periods of scarcity.
As data center operators grapple with the need for reliable, carbon-free baseload power, nuclear energy has reentered the conversation in dramatic fashion. Microsoft signed a power purchase agreement to restart the nuclear reactor at Pennsylvania’s Three Mile Island that shut down in 2019, marking an unprecedented move by a tech company.9 Amazon, Google, and Oracle have all explored small modular reactors as potential solutions for their data center energy needs.
Small Modular Reactors are emerging as a particularly promising option—they’re scalable nuclear reactors that could provide up to 300 megawatts of carbon-free power to data centers.32 While commercial deployment isn’t expected until 2030 at the earliest, SMR agreements are anticipated to double in 2025, signaling growing enthusiasm for the technology.32 Google signed a first-of-its-kind partnership with Kairos Power to unlock clean power from a series of SMRs.15
Microsoft has gone even further, entering into the world’s first fusion energy purchase agreement with Helion Energy for provision of electricity from its first fusion power plant expected to be online by 2028.15 While highly speculative, such investments demonstrate the industry’s willingness to explore all potential pathways to reliable, carbon-free power.
However, nuclear solutions face significant hurdles. Regulatory approval processes are lengthy, safety concerns persist regarding waste handling and accident risk, and construction costs remain high. Most critically, new nuclear capacity cannot be built quickly enough to meet near-term data center demand, which can come online in just one to two years compared to the many years required for nuclear plant construction.
The collision between AI growth and climate goals has created complex policy challenges that governments and regulators are only beginning to address. Different jurisdictions are taking markedly different approaches, creating a patchwork of incentives and requirements that shape where and how data centers develop.
Seventeen U.S. jurisdictions have statutory 100% clean energy requirements covering utilities, with attainment years starting in 2032. The European Union’s 2020 digital strategy called for data centers to become climate neutral by 2030, with the 2023 EU Energy Efficiency Directive requiring data centers to report on energy consumption, water usage, and use of renewable energy.
However, these policies face implementation challenges. North Carolina regulators approved a controversial green tariff proposal by Duke Energy that provides a pathway for data center customers to accelerate solar energy developments—but critics questioned whether the tariff simply allows data centers to subsidize renewable energy development that would have occurred anyway, given that Duke is legally required to cut its carbon emissions by 95% by 2050.9
Some states have taken more aggressive stances. Virginia’s regulatory environment has become contentious, with the state’s clean energy mandates clashing against Dominion Energy’s grid planning that uses data center demand to justify fossil fuel plant extensions.9 The issue has become central to gubernatorial elections in Virginia and New Jersey, where power costs and data center policy have emerged as key voter concerns.
Rate structures are also evolving. More jurisdictions are considering specific large-load tariffs for data centers with stringent requirements to ensure ratepayers are buffered from costs associated with data center expansion during a time of unprecedented utility rate increases. Some utilities have moved to require tech firms to put up more collateral or pay for specific amounts of electricity even if they use less, attempting to ensure fair cost allocation.
In short, the collision between AI’s exponential growth and the urgent need for energy system decarbonization represents one of the defining challenges of our time.