Electric utilities across the U.S. are facing a critical challenge: determining how much of the soaring demand for electricity from artificial intelligence data centers is real, and how much is speculative. As tech companies announce massive plans for server farms that could consume power on par with entire cities, utilities are struggling to assess the true scale of future energy needs. Willie Phillips, former chairman of the Federal Energy Regulatory Commission, noted that some regions have dramatically revised down their initial projections after realizing they may have overestimated demand. This reflects growing skepticism about whether all the promised data center expansions will materialize as expected. AI firms are pitching similar large-scale projects to multiple utilities across different states, seeking the fastest path to power. Brian Fitzsimons, CEO of GridUnity, a company that tracks power connection requests across the fragmented U.S. grid, said the same-sized projects are being proposed in widely separated areas—making it hard for utilities to gauge actual demand. This “data center shopping” complicates grid planning and risks overinvestment. Electricity prices are already rising due to supply constraints, and inaccurate forecasts could lead to billions in unnecessary infrastructure spending. FERC Chairman David Rosner warned that even small differences in load projections can have massive financial and consumer impacts. Joe Dominguez, CEO of Constellation Energy, expressed concern that demand is being overstated. He urged caution, saying, “I think the load is being overstated. We need to pump the brakes here.” While data centers are undeniably consuming more power—gigawatt-scale facilities are now common, far surpassing the 50-megawatt standard of just a few years ago—the long-term sustainability of these projections remains uncertain. Grid Strategies estimates that by 2030, AI-driven data centers could add up to 60 gigawatts of demand—equivalent to Italy’s peak hourly power use in 2024. Fitzsimons insists this is not a bubble but a transformation that demands a long-term energy strategy, calling for a 50-year national energy policy. Still, utilities need firm commitments from data center developers to make reliable forecasts. Without them, planning remains speculative. The Edison Electric Institute reports that utilities spent $178 billion on grid upgrades last year and are projecting $1.1 trillion in capital investments through 2029. Despite the uncertainty, renewable energy is emerging as the fastest way to add new capacity. More than 90% of projects currently waiting for grid connections are solar, wind, or battery storage, according to Enverus. Supply chain issues and high costs have slowed the deployment of natural gas turbines, making renewables the preferred option. However, political uncertainty looms. President Donald Trump has favored coal, natural gas, and nuclear power while opposing solar and wind, raising questions about whether sufficient new generation can be built in time. Utilities will reject data center requests if they lack the power to serve them, as reliability remains their top priority. Some AI companies are now exploring off-grid solutions—building their own power plants on-site, known as “behind the meter” generation. Nvidia CEO Jensen Huang told CNBC that self-generated power at data centers could move faster than traditional grid connections. “We should invest in just about every possible way of generating energy,” he said, emphasizing the need for rapid, flexible energy solutions to keep up with AI’s explosive growth.