Released August 20, 2025 | SUGAR LAND
en
Written by John Egan for Industrial Info Resources (Sugar Land, Texas)--Economists say solving problems involves a combination of supply-side and demand-side approaches. For most of the last year, electricity providers have relied nearly exclusively on the supply side--building more power plants and transmission & distribution (T&D) infrastructure--to meet soaring projections of electric demand growth from artificial intelligence (AI). But there is a growing recognition that electricity providers cannot build their way out of the AI electric challenge.
On the "build" side of the equation, two PPL Corporation (Allentown, Pennsylvania) subsidiaries, Kentucky Utilities (Lexington, Kentucky) and Louisville Gas & Electric (Louisville, Kentucky) last month cleared a regulatory hurdle that will enable them to build 1,300 megawatts of gas-fired power to serve expected data center and other loads. In testimony filed at the Kentucky Public Service Commission (Frankfort, Kentucky), the two utilities said they expected about 1,875 MW of new data center load, and 580 MW of other commercial and industrial load, by 2032.
The two Kentucky utilities join their brethren in Texas, Georgia, Virginia, Alabama, Indiana, Ohio, Illinois, Maryland and Pennsylvania, among others, in readying plans to build new electric generating capacity to meet the strong demand growth created by the expected surge of data centers, many of which will have AI capabilities. Across the U.S., a number of electric utilities have increased their proposed capital budgets to reflect this buildout. For more on that, see July 14, 2025, article - Ohio Plots Path for Data Centers as Large-Scale Buildouts Loom; July 17, 2025, article - Big Talk of AI, Power Development at Pennsylvania Summit; April 25, 2024, article - Data Centers, Oil & Gas Industry to Drive Surge in Texas Electricity Demand; and April 16, 2024, article - Data Center Construction Propels Electric Load Growth and Utility Capex.
Energy agencies, including the U.S. Energy Information Administration and the International Energy Agency (Paris, France) have issued ever-rising estimates about the electric demand growth stemming from AI. For more on that, see July 1, 2025, article - EIA: U.S. Data Center Power Demand to More Than Triple by 2050.
By and large, utilities, concerned about potential electric shortfalls, have taken a traditional "build first" approach to meeting future electric demand of large loads. But utility plans to build new power plants or T&D infrastructure will have to win approval of state or federal officials, and some of those officials have started pushing back over how a given utility plans to recover planned capital outlays made to serve large-load customers.
The costs and lead times needed for the Power industry to build its way out of the AI power challenge have caused a growing number of observers to ask about less-expensive ways to meet the electric needs of the AI industry, or whether all that demand growth will materialize.
For example, in Colorado last month, the chair of the state's Public Utilities Commission (Denver, Colorado) had sharp questions for the state's largest utility, Xcel Energy (Minneapolis, Minnesota), about its plans to raise residential electric prices approximately 50% through 2031, mainly to recover the cost of building new generation and infrastructure to meet expected load from data centers, while prices for industrial and commercial customers remained flat.
The word "over-build" came up in regulatory filings and news articles by those who questioned Xcel's plans. Another term coming into focus: "project fallout." Intervenors in the Xcel docket noted that developers of large loads have withdrawn some requests for new service. One intervenor said data center service requests are, in some cases, speculative.
The Colorado PUC is not the only utility regulatory body concerned about potentially overbuilding long-lived assets like power plants or T&D infrastructure. Utility commissions in Georgia and Ohio, the Federal Energy Regulatory Commission (FERC) and lawmakers in Virginia are starting to push back against utility plans to build new assets to meet future large loads from proposed data centers, AI and advanced manufacturing facilities.
A few voices in the electricity business are starting to raise similar concerns. During first-quarter earnings calls, the chief executive officers (CEOs) of competitive merchant electric companies Constellation Energy Corporation (Baltimore, Maryland) and Vistra Corporation (Houston, Texas) reportedly warned investors about a potential power-industry AI-driven over-build, according to a recent article from the Institute for Energy Economics and Financial Analysis (Valley City, Ohio). Many data center requests for electric service likely are duplicative, said one CEO. Said the other, "It's hard not to conclude that the headlines are inflated."
"Over-build" is precisely what longtime energy guru Amory Lovins warned about earlier this summer, when he issued a research report which warned there was a risk that electric utilities could waste hundreds of billions of dollars building assets that will not be needed to serve data center and AI loads.
In a May 10 paper, "Artificial Intelligence Meets Natural Stupidity: Managing the Risks," Lovins, founder of the sustainability consultancy Rocky Mountain Institute (Boulder, Colorado) and an energy consultant to scores of Fortune 500 companies, wrote: "Future electricity needs for AI are wildly uncertain -- shaped by unproven concepts, disputed performance, limited trust, volatile markets, unpredictable adoption and technical efficiency that quadruples roughly each year. Yet a speculative surge is driving massive investment in data centers and new electricity supplies, risking a 12-figure overbuild. Avoiding an electricity bubble requires clear-eyed analysis, disciplined planning and using markets to allocate risks fairly to potential beneficiaries."
Lovins, head of Lovins Associates LLC (Snowmass, Colorado), has made a history of zigging when the rest of the energy industry zagged. Five decades ago, he posited that there was a "soft energy path" to meeting the world's energy needs: energy efficiency and renewable energy. His article on that topic, initially published in Foreign Affairs magazine, turned into a book. Fortune 500 clients sought him out as they scrambled to cut energy outlays.
He is not alone in asking hard questions about the dominant "build first" ethos among electricity providers for meeting the future power needs of large loads. Nick Wayth, CEO of the Energy Institute (London, England) this summer commented, "Massive hyperscalers don't want to spend lots of money on data center cooling. They want to drive the cost down. So there's research into semiconductors that can operate at much higher temperatures, rather than having to be kept in chilled rooms."
In fact, that's exactly what data center and AI operators are doing. A recent IIR article explored ways to lower the electric demand of AI data centers. For more on that, see March 17, 2025, article - On a Diet: As Data Centers Strain the Grid, Can They Curb Their Power Appetite?. On July 31, Constellation said it inked a deal with GridBeyond (Dublin, Ireland) to launch an AI-powered demand response program in the PJM wholesale market to improve grid flexibility, and potentially save customers millions of dollars.
The risk of a 12-figure over-build raised by Lovins, which would be up to hundreds of billions of dollars, was an eye-opening assertion. But he has made those before, and more often than not, over the long term, he has proved to be prescient.
The AI energy riddle is unlikely to be solved through exclusive reliance on either supply or demand approaches. As Wayth of the Energy Institute noted, supply- and demand-side approaches function as yin and yang, interdependent and interconnected elements of an integrated solution that influence each other.
Another economic maxim, "the best cure for high prices is high prices," appears to be at work as well, as data center and AI developers investigate ways to lower power demand while regulators work on fairly allocating the costs of those facilities. All of that suggests a relatively high level of project fallout that could cut sharply into current projected electric demand growth and power provider capital budgets.
On the "build" side of the equation, two PPL Corporation (Allentown, Pennsylvania) subsidiaries, Kentucky Utilities (Lexington, Kentucky) and Louisville Gas & Electric (Louisville, Kentucky) last month cleared a regulatory hurdle that will enable them to build 1,300 megawatts of gas-fired power to serve expected data center and other loads. In testimony filed at the Kentucky Public Service Commission (Frankfort, Kentucky), the two utilities said they expected about 1,875 MW of new data center load, and 580 MW of other commercial and industrial load, by 2032.
The two Kentucky utilities join their brethren in Texas, Georgia, Virginia, Alabama, Indiana, Ohio, Illinois, Maryland and Pennsylvania, among others, in readying plans to build new electric generating capacity to meet the strong demand growth created by the expected surge of data centers, many of which will have AI capabilities. Across the U.S., a number of electric utilities have increased their proposed capital budgets to reflect this buildout. For more on that, see July 14, 2025, article - Ohio Plots Path for Data Centers as Large-Scale Buildouts Loom; July 17, 2025, article - Big Talk of AI, Power Development at Pennsylvania Summit; April 25, 2024, article - Data Centers, Oil & Gas Industry to Drive Surge in Texas Electricity Demand; and April 16, 2024, article - Data Center Construction Propels Electric Load Growth and Utility Capex.
Energy agencies, including the U.S. Energy Information Administration and the International Energy Agency (Paris, France) have issued ever-rising estimates about the electric demand growth stemming from AI. For more on that, see July 1, 2025, article - EIA: U.S. Data Center Power Demand to More Than Triple by 2050.
By and large, utilities, concerned about potential electric shortfalls, have taken a traditional "build first" approach to meeting future electric demand of large loads. But utility plans to build new power plants or T&D infrastructure will have to win approval of state or federal officials, and some of those officials have started pushing back over how a given utility plans to recover planned capital outlays made to serve large-load customers.
The costs and lead times needed for the Power industry to build its way out of the AI power challenge have caused a growing number of observers to ask about less-expensive ways to meet the electric needs of the AI industry, or whether all that demand growth will materialize.
For example, in Colorado last month, the chair of the state's Public Utilities Commission (Denver, Colorado) had sharp questions for the state's largest utility, Xcel Energy (Minneapolis, Minnesota), about its plans to raise residential electric prices approximately 50% through 2031, mainly to recover the cost of building new generation and infrastructure to meet expected load from data centers, while prices for industrial and commercial customers remained flat.
The word "over-build" came up in regulatory filings and news articles by those who questioned Xcel's plans. Another term coming into focus: "project fallout." Intervenors in the Xcel docket noted that developers of large loads have withdrawn some requests for new service. One intervenor said data center service requests are, in some cases, speculative.
The Colorado PUC is not the only utility regulatory body concerned about potentially overbuilding long-lived assets like power plants or T&D infrastructure. Utility commissions in Georgia and Ohio, the Federal Energy Regulatory Commission (FERC) and lawmakers in Virginia are starting to push back against utility plans to build new assets to meet future large loads from proposed data centers, AI and advanced manufacturing facilities.
A few voices in the electricity business are starting to raise similar concerns. During first-quarter earnings calls, the chief executive officers (CEOs) of competitive merchant electric companies Constellation Energy Corporation (Baltimore, Maryland) and Vistra Corporation (Houston, Texas) reportedly warned investors about a potential power-industry AI-driven over-build, according to a recent article from the Institute for Energy Economics and Financial Analysis (Valley City, Ohio). Many data center requests for electric service likely are duplicative, said one CEO. Said the other, "It's hard not to conclude that the headlines are inflated."
"Over-build" is precisely what longtime energy guru Amory Lovins warned about earlier this summer, when he issued a research report which warned there was a risk that electric utilities could waste hundreds of billions of dollars building assets that will not be needed to serve data center and AI loads.
In a May 10 paper, "Artificial Intelligence Meets Natural Stupidity: Managing the Risks," Lovins, founder of the sustainability consultancy Rocky Mountain Institute (Boulder, Colorado) and an energy consultant to scores of Fortune 500 companies, wrote: "Future electricity needs for AI are wildly uncertain -- shaped by unproven concepts, disputed performance, limited trust, volatile markets, unpredictable adoption and technical efficiency that quadruples roughly each year. Yet a speculative surge is driving massive investment in data centers and new electricity supplies, risking a 12-figure overbuild. Avoiding an electricity bubble requires clear-eyed analysis, disciplined planning and using markets to allocate risks fairly to potential beneficiaries."
Lovins, head of Lovins Associates LLC (Snowmass, Colorado), has made a history of zigging when the rest of the energy industry zagged. Five decades ago, he posited that there was a "soft energy path" to meeting the world's energy needs: energy efficiency and renewable energy. His article on that topic, initially published in Foreign Affairs magazine, turned into a book. Fortune 500 clients sought him out as they scrambled to cut energy outlays.
He is not alone in asking hard questions about the dominant "build first" ethos among electricity providers for meeting the future power needs of large loads. Nick Wayth, CEO of the Energy Institute (London, England) this summer commented, "Massive hyperscalers don't want to spend lots of money on data center cooling. They want to drive the cost down. So there's research into semiconductors that can operate at much higher temperatures, rather than having to be kept in chilled rooms."
In fact, that's exactly what data center and AI operators are doing. A recent IIR article explored ways to lower the electric demand of AI data centers. For more on that, see March 17, 2025, article - On a Diet: As Data Centers Strain the Grid, Can They Curb Their Power Appetite?. On July 31, Constellation said it inked a deal with GridBeyond (Dublin, Ireland) to launch an AI-powered demand response program in the PJM wholesale market to improve grid flexibility, and potentially save customers millions of dollars.
The risk of a 12-figure over-build raised by Lovins, which would be up to hundreds of billions of dollars, was an eye-opening assertion. But he has made those before, and more often than not, over the long term, he has proved to be prescient.
The AI energy riddle is unlikely to be solved through exclusive reliance on either supply or demand approaches. As Wayth of the Energy Institute noted, supply- and demand-side approaches function as yin and yang, interdependent and interconnected elements of an integrated solution that influence each other.
Another economic maxim, "the best cure for high prices is high prices," appears to be at work as well, as data center and AI developers investigate ways to lower power demand while regulators work on fairly allocating the costs of those facilities. All of that suggests a relatively high level of project fallout that could cut sharply into current projected electric demand growth and power provider capital budgets.
