6 Ways AI is Improving the Energy Landscape

Molly Podolefsky, Ph.D., Clarum Advisors

May 1, 2025

While AI poses challenges for the global energy economy in terms of capacity shortages, transmission and distribution constraints and obstacles to decarbonization, it is also contributing solutions to those problems by transforming energy systems and modernizing the grid. In this report we focus on 6 ways the energy landscape is benefiting from AI:

Addressing demand shortages through energy efficiency, demand management and load shifting
Relieving transmission and distribution constraints through grid modernization and optimization
Facilitating decarbonization by improving renewables integration, optimizing energy use for lower emissions and streamlining permitting and review processes
Improving grid reliability and resilience
Reducing power outages, helping protect against wildfires and aiding utilities in storm response
Helping harden the US grid against cyberattacks

AI has people on edge for many reasons, from its potential use as a disinformation tool to its competition with human workers. At the same time, AI is placing a tremendous strain on the electric grid through the rapid expansion of hyperscale data centers. Startup DeepSeek disrupted markets earlier this year with the claim that it had slashed AI energy use, though recent reports suggest its lower energy use in training is offset by higher consumption while generating answers.[1] While improvements in efficiency will happen over time, Jevon’s paradox suggests that as machines become more efficient, they become cheaper, leading to increased demand. In other words, increasing efficiency is unlikely to solve the problem of AI load growth. By focusing only on the negatives, however, we risk missing an equally important story—the emerging narrative around AI’s potential to revolutionize grid operation, allowing us to safely and reliably get more out of systems and infrastructure already in place.

The meteoric rise of computationally intensive AI applications such as chatbots, virtual assistants and text-to-image engines, in tandem with the burgeoning blockchain economy, have reversed decades of flat or low electric load growth in North America.[2] As an example, training OpenAI’s GPT-4 consumed 50 GWh of energy—enough to power 5,000 US homes for a year. Launching the next generation could be massively more energy intensive, as GPT-4 consumed 50 times more energy in training than its predecessor, GPT-3. [3] Even with ongoing improvements in energy efficiency, Jevon’s paradox suggests load growth associated with AI will only increase. Duke Energy and American Electric Power, two of the largest utility companies serving US hyperscalers recently stated they see no slowing of data center growth as a response to DeepSeek, citing the potential for energy efficiency gains to be wiped out by the resulting increase in AI use.[4] Whereas data centers currently comprise 2.5% of US electricity demand, by 2030 they are expected to consume 7.5%.[5] As a result, the US is facing load growth on a scale not seen since the 1980’s, with five-year growth rates increasing fivefold over the past two years to over 125 GW.[6]

The most immediate problem caused by exponential load growth is lack of capacity. Bringing new generation online is a years-long process involving planning, obtaining regulatory approval and navigating interconnection queues. Utilities and independent power producers will have to run a marathon sprint to bring over 100 GW new capacity online over the next five years. A subsequent

and equally thorny issue is how to meet transmission and distribution needs as all this new power comes online. While power grids around the country are generally not equipped to deal with the mammoth demand requirements of data centers, transmission and distribution networks in hyperscaler hotspots like Northern Virginia, Dallas-Ft. Worth and Atlanta feel the strain most acutely. Capacity shortfalls and transmission woes are further compounded in regions like Virginia by clean energy legislation requiring a substantial role for renewables in meeting increased demand while hampered by multi-year interconnection queue backlogs and insufficient transmission capacity.[7][8]

If these problems sound intractable, let us introduce some reasons for optimism in the form of AI. At its core, AI provides dynamic systems not just with the ability to optimize, but with the potential to learn and improve at lightning speed. The US energy grid comprises a complex system of local distribution networks and regional transmission lines carrying electricity supplied by utilities, independent power producers and distributed generation assets, orchestrated by regional transmission operators, independent system operators and aggregators, and regulated by FERC, NERC and dozens of state utility commissions. Supply and demand are balanced across the US power system each day through an intricate system of instantaneous, intra-day and longer-term bilateral and wholesale energy market transactions. AI has the potential to revolutionize the functioning of the US power system by improving the speed, scale and efficiency of buying, selling and moving electricity around the grid, thereby helping address the capacity, transmission and decarbonization issues it has created. AI also has the potential to contribute solutions to energy system problems it did not cause, from anticipating power outages and preventing blackouts to identifying system faults and protecting against wildfires and cyberattacks.

Addressing demand shortages through energy efficiency, demand management and load shifting
Over the past decade, AI has shown tremendous potential for reducing demand by improving the way companies use energy and leveraging flexibility to create capacity. As early as 2016, Google pioneered the use of AI to intelligently cool its data centers, thereby reducing energy demand. Through its DeepMind AI platform, the company was able to reduce the energy used for cooling by 40%.[9] Google has continually harnessed the power of AI to reduce carbon emissions. In 2021 Google announced plans to deploy an AI-powered “carbon-intelligent computing platform” to shift workloads between data centers to decarbonize.[10] Based on the platform’s success, the company expanded its use to create flexible load that can be treated as a virtual power plant (VPP) and shared with grid operators in response to energy crises around the world.[11] AI-powered VPPs hold the potential to unlock vast reservoirs of flexible capacity outside the realm of hyperscaler data centers. Leveraging predictive analytics and real-time data processing capabilities, AI-driven VPPs can intelligently orchestrate and dispatch solar power, battery storage, EV charging systems, vehicle-to-grid assets, smart thermostats, smart buildings, commercial HVAC and flexible industrial loads, thereby avoiding the build-out of new generation capacity. With over 30 GW of installed capacity and a forecasted growth rate of 30% through 2030, the US VPP market can help bridge the gap between available capacity and future demand.[12][13] Meanwhile, AI-enabled energy trading is creating new markets for flexibility leveraging microgrids, peer-to-peer energy trading and other disruptions to the traditional utility paradigm.
Relieving transmission and distribution constraints through grid modernization and optimization
At the same time AI is burdening transmission and distributions systems with increased load, system operators are leveraging AI to build the power grid of the future. Grid modernization through AI enables power system operators to manage and synthesize vast amounts of data from disparate sources, use that data to rapidly generate accurate forecasts of supply and demand and dynamically optimize system responses. AI enables grid operators to optimize transmission flows in real time, minimizing cost. The presence of AI features in physical grid patents has surged, increasing globally by 50% between 2010 and 2022, led by supply-demand forecasting and EV charging applications.[14] The National Renewable Energy Laboratory (NREL) has begun testing a power grid control room AI assistant, eGridGPT, to enhance the effectiveness of human grid operators by analyzing procedures, simulating scenarios and optimizing decision-making processes.[15] Other AI-driven innovations aim to expand the capacity of transmission and distributions lines themselves, allowing greater volumes of energy to be transferred using existing infrastructure. LineVision, a leading grid modernization provider, deploys AI-powered dynamic line rating (DLR) hardware and software systems enabling utilities and grid operators to increase capacity on existing transmission lines by up to 40%.[16]
Facilitating decarbonization by improving renewables integration, optimizing energy use for lower emissions and streamlining permitting and review processes
Hyperscaler growth has slowed the pace of decarbonization, encouraging affected utilities to extend the lives of coal-fired power plants and increase investments in natural gas plants to keep up with demand. Utilities across the US Southeast have sought regulatory approval to expand and fast-track development of natural gas power plants, in addition to more renewables, to meet data center demand.[17] On the other hand, AI enables real-time dynamic pricing models which allow for more efficient integration of intermittent renewables on the grid, and data center companies have begun leveraging AI to decarbonize their own operations. Google has launched an AI-powered “carbon intelligent” computing platform which it uses to shift processing loads between hyperscale data centers based on the availability of renewable energy as a step towards achieving the company’s goal of 24x7 carbon-free energy.[18] Smart buildings and home energy management systems increasingly allow energy consumers to minimize carbon emissions. Xcel Energy credits AI with helping it automate processes, predict renewable energy generation and adjust grid operations dynamically, helping the company reach its goal of net-zero emissions by 2050.[19] AI tools also have the potential to speed labor-intensive review and permitting processes, bringing renewables online faster. The US Department of Energy (DOE) is investing $20M to develop AI-powered software aiding the federal review process for National Environmental Policy Act (NEPA) compliance.[20] The DOE, in partnership with the Grid Deployment Office (GDO), is also investing up to $30MM to tackle the backlog of renewables projects in interconnection queues through an Artificial Intelligence for Interconnection (AI4IX) program which will speed the interconnection review, approval and commissioning process.[21]
Improving grid reliability and resilience
AI confers benefits on the energy system extending beyond mitigation of negative impacts on generation capacity, T&D systems and the pace of decarbonization. Over the next decade, AI holds the potential to significantly increase grid reliability and resilience. US grid infrastructure is aging—approximately 70% of large transformers have been in service for 25 years or more[22]—and the key to reliability and resilience will be proactive detection of faults and failures. Argonne National Lab is developing AI-enabled grid asset health monitoring software designed to predict failure before it occurs, reducing the potential for power failures and minimizing downtime.[23] While digital twin technology has been in use for over a decade, we are witnessing massive improvements in digital twin performance as AI enables integration and learning based on ever larger datasets, facilitating detection of patterns and anomalies to predict failure earlier and with greater accuracy.
Reducing power outages, helping protect against wildfires and aiding utilities in storm response
AI-enabled software systems will play an increasingly important role in detecting faults in the energy system that could lead to power outages and wildfires. Whisker Labs recently released information based on the analysis of data from thousands of sensors detailing voltage drops due to faults on high voltage lines in Altadena just prior to the deadly Eaton fire that erupted there in 2025.[24] Harnessing the power of AI to identify telltale signs of equipment faults and failure earlier, utilities and grid operators will be better equipped to ensure the reliability and safety of power systems in the future. Ubicquia provides AI-powered utility pole and transformer sensors to aid with fault and damage detection, wildfire prevention and storm response. Portland General Electric uses AI-powered camera systems which have improved wildfire ignition response times by over two hours compared with satellite and emergency system notifications.[25] Restoring power after a disaster involves massive mobilization of resources and can leave residents without power for long durations—it took utilities upwards of two weeks to restore power to 95% of customers after recent hurricanes along the Eastern seaboard and Gulf Coast.[26] Utilities are increasingly relying on AI to restore power faster and more efficiently. AI helps companies assess damage to assets, prioritize repairs, optimize repair routes and dispatch crews efficiently to restore power, significantly reducing the amount of time customers spend without power.
Helping harden the US grid against cyberattacks
A final frontier for AI to improve energy system resilience in the next decade will be hardening US grid infrastructure against cyberattacks. Attacks targeting US grid infrastructure increased by 70% in 2024 year-over year, with over 1000 attacks reported through Q3.[27] Using generative AI, cybersecurity software providers can synthesize vast and varied datasets, create more sophisticated defenses and leverage complex simulations to detect vulnerabilities. Fortress, a cybersecurity company specializing in solutions for energy utilities recently released a report detailing “highly exploitable” vulnerabilities in software products and code used by US utilities.[28] Fortress and other cybersecurity companies serving utilities and grid operators leverage AI to help detect and mitigate vulnerabilities and cyberthreats, increasing the resilience of the grid.

Despite shifts in federal funding and disruptive innovations like DeepSeek, the dual trends of AI-induced load growth and AI-driven energy system modernization seem likely to continue. From our vantage point today, the larger and more impactful story will be AI’s role in revolutionizing the grid. Between now and the end of this decade AI innovation will continue, and many of today’s nascent grid technologies will mature, transforming the grid in ways we cannot yet imagine. While a techno-optimist reliance on AI to fix all the problems it has created may be naïve, we should not underestimate its power to transform the energy economy for the better.

[1] MIT Technology Review, Jan. 31, 2025 https://www.technologyreview.com/2025/01/31/1110776/deepseek-might-not-be-such-good-news-for-energy-after-all/

[2] New York Times, Mar. 14, 2024 https://www.nytimes.com/interactive/2024/03/13/climate/electric-power-climate-change.html

[3] Forbes, May 24, 2024 https://www.forbes.com/sites/arielcohen/2024/05/23/ai-is-pushing-the-world-towards-an-energy-crisis/

[4] Energy Connects, Feb. 13, 2025 https://www.energyconnects.com/news/renewables/2025/february/utilities-are-full-speed-on-data-centers/

[5] S&P Global, Mar. 27, 2024 https://www.spglobal.com/market-intelligence/en/news-insights/articles/2024/3/rising-datacenter-demand-forces-reckoning-with-us-utility-decarbonization-goals-80889360

[6] Canary Media, Dec. 9, 2024 https://www.canarymedia.com/articles/utilities/data-centers-are-driving-us-power-demand-to-hard-to-reach-heights

[7] UtilityDive, Dec. 10, 2024 https://www.utilitydive.com/news/pjm-load-forecast-data-center-dominion-virginia/735056/

[8] National Public Utilities Council (NPUC), accessed Feb. 7, 2025 https://www.motive-power.com/mapped-the-age-of-energy-projects-in-interconnection-queues-by-state/

[9] Google, Jul. 20, 2016 https://blog.google/outreach-initiatives/environment/deepmind-ai-reduces-energy-used-for/

[10] Data Center Knowledge, May 18, 2021 https://www.datacenterknowledge.com/hyperscalers/google-to-shift-workloads-between-data-centers-to-follow-clean-energy

[11] UtilityDive, Nov. 7, 2023 https://www.utilitydive.com/news/google-carbon-intelligent-computing-platform-system-reliability-demand-response-grid-emergency/698958/

[12] US Department of Energy, Jan. 2025 https://liftoff.energy.gov/wp-content/uploads/2025/01/LIFTOFF_DOE_VirtualPowerPlants2025Update.pdf