Utilities grapple with a multibillion question: How much AI data center power demand is real

Electricity providers across the United States are grappling with the challenge of estimating the actual power demand stemming from the rapid growth of artificial intelligence (AI). As the market anticipates massive investments in infrastructure to support an influx of data centers, industry experts express skepticism about the accuracy of these projections. Willie Phillips, former chairman of the Federal Energy Regulatory Commission, highlighted concerns regarding the reliability of forecasts, noting that certain regions have scaled back their initial estimates of demand. As AI companies unveil ambitious plans for extensive server farms, some of which may require electricity equivalent to that consumed by entire cities, the competition for power access is intensifying. Brian Fitzsimons, CEO of GridUnity, explained that the tech industry is presenting similar large-scale projects to multiple utilities, complicating the ability of electricity providers to ascertain how much power generation will be necessary to maintain grid reliability. Rising electricity prices are already affecting consumers, as the supply struggles to keep pace with escalating demand. FERC Chairman David Rosner cautioned that even minor discrepancies in electricity load predictions could result in significant financial implications, potentially impacting billions in investments and consumer bills. Constellation Energy CEO Joe Dominguez echoed these concerns, suggesting that the anticipated load might be overstated and urging caution in planning. Despite these warnings, the stock market remains bullish on data centers, contributing to a notable surge in utility company stocks—an increase of around 21% this year alone. This rally has added nearly $500 billion in value to U.S. electricity suppliers over two years, the most substantial growth for the sector since the early 2000s. OpenAI CEO Sam Altman expressed concerns about a potential AI bubble, advising investors to temper their excitement. While the precise scale of future demand remains uncertain, experts agree that the U.S. is on the brink of a historic spike in electricity consumption, following years of stagnant growth. Rob Gramlich, president of Grid Strategies, emphasized the existing electricity demands of current data centers, which are consuming significantly more power than in the past. Projections indicate that an additional 120 gigawatts of electricity demand could arise by 2030, with 60 gigawatts attributed to data centers alone—roughly equivalent to Italy's peak hourly demand. Fitzsimons believes that this transformation will fundamentally reshape the energy landscape, emphasizing the need for a long-term energy policy. However, the uncertainty surrounding financial commitments from data center operators presents a challenge for utilities in planning effectively. Utilities face the potential risk of over-investing in infrastructure that may not be required. Fitzsimons noted that the current market environment, marked by supply chain disruptions and inflation, is less conducive to overbuilding than in the past. In a recent development, Altman announced a partnership with Nvidia to create 10 gigawatts of data centers, raising concerns about the feasibility of securing the necessary power. As demand grows, competition for limited electrical infrastructure is intensifying, leading to rising costs for essential components. The pressing question remains how quickly new energy generation can be developed. Natural gas turbines are largely sold out for the next several years, while advanced nuclear technologies are not expected to be commercially viable until at least the 2030s. In contrast, renewable energy sources, particularly solar and battery storage, are poised for rapid deployment, with a significant majority of current power projects being renewables. However, policy uncertainties under the current administration pose challenges for expanding renewable capacity. Utilities may have to decline requests from data centers if they cannot guarantee sufficient power availability, which remains their top priority. In an innovative approach, some AI companies are exploring options to generate power on-site, independent of the grid. Nvidia CEO Jensen Huang advocated for investment in diverse energy generation methods, suggesting that self-generated power at data centers could expedite energy solutions more effectively than relying on grid infrastructure.