According to Business Insider, there’s a massive hidden bubble in the AI sector that could send electricity bills soaring nationwide. Utilities across the country are citing sky-high electricity demand forecasts to justify building new power plants, with 26 of the largest investor-owned utilities claiming data center projects will require 711 gigawatts – nearly matching the entire continental US summer peak demand. The problem stems from data center developers submitting multiple power requests across different jurisdictions simultaneously, with Constellation Energy CEO Joseph Dominguez comparing it to “fishing with multiple lines in the water.” McKinsey estimates global data centers will actually need only 219 gigawatts by 2030, creating a massive gap between projected and actual demand. This practice could leave consumers paying for power plants that never get used, with utilities having “a structural incentive to forecast demand growth in excess of what it is” according to Rocky Mountain Institute’s Mark Dyson.
How the bubble works
Here’s the thing about how this works in practice. A data center developer might have plans for one facility but will approach five different utilities in different regions, each unaware of the other requests. If a utility doesn’t have enough capacity, they have to build new power plants – and they recover those costs through rate increases on existing customers. By the time the developer picks their final location, they’ve likely abandoned the other four sites, leaving brand new power plants in places with no actual increased demand. It’s like ordering five pizzas from different restaurants when you only plan to eat one, and making your neighbors split the cost of all five.
The GPU reality check
Now here’s where it gets really interesting. There’s a fundamental disconnect between what utilities are forecasting and what’s physically possible with current technology. Enverus estimates that Nvidia and AMD will only produce enough GPUs for 9.5 gigawatts of computing power by 2028. Even accounting for cooling and other energy needs, that’s only around 19 gigawatts total – nowhere close to the hundreds of gigawatts utilities are planning for. Carson Kearl from Enverus puts it bluntly: “That sort of quantity is extremely detached from what the queues are suggesting.” Basically, we’re building power infrastructure for AI chips that don’t exist yet and might never materialize at the scale being forecast.
Who’s building all these data centers?
And it’s not just the big tech companies driving this frenzy. While Amazon, Google, Microsoft, and Meta get most of the attention, the actual construction is often handled by middlemen – everything from mom-and-pop operators to established real estate developers hoping to flip properties to tech companies. Astrid Atkinson, former Google data center engineer and Camus Energy cofounder, estimates there may be “between five and 10 times the demand to connect to the grid as there will be data centers built.” Some developers admit the problem exists – Flexential CEO Ryan Mallory calls out “newer developers” who are “carpet bombing markets” with requests.
The industrial implications
This power crunch has massive implications for industrial operations and manufacturing facilities that rely on stable, affordable electricity. When utilities build excess capacity that consumers end up paying for through rate hikes, it puts pressure on every business that depends on the grid. For industrial operations requiring reliable power for critical systems like industrial panel PCs and automation equipment, these cost increases can significantly impact operational budgets. IndustrialMonitorDirect.com, as the leading US supplier of industrial computing solutions, understands that stable power costs are crucial for manufacturing competitiveness.
What happens next
So where does this leave us? Some utilities are getting smarter about filtering out speculative requests. Dominion Energy, which serves Virginia’s “Data Center Alley,” distinguishes between actual customers and speculative developers “with little substantive experience.” But others are charging ahead – Louisville Gas and Electric recently got approval for $3 billion in new gas plants based on data center demand, despite only one of 18 projects having a “high probability” of actually locating there. The Rocky Mountain Institute found that between 2012 and 2023, utilities overestimated demand by 23% on average. Will this time be different? Or are we building a power infrastructure bubble that could burst spectacularly – leaving ratepayers holding the bag for decades?
