In our new artificial intelligence era, power isn’t just about computing abilities—it’s about who controls the energy grid.
As artificial intelligence (AI) propels the economy into uncharted territory, a less visible but equally seismic shift is jolting the foundation beneath it: the US electric grid and the rules of who gets access to it. Once engineered for steel mills and suburbs, the grid is now being reshaped by data centers—vast, humming fortresses of compute that guzzle electricity at levels reminiscent of heavy industry.
From northern Virginia to central Texas and the Nevada desert, the expansion of AI infrastructure is forcing utilities, regulators, and grid operators to rewrite the rulebook on electricity planning, sparking a quiet reckoning that could alter the nation’s energy landscape for decades to come.
On July 15 in Pittsburgh, Pennsylvania, President Trump announced a massive set of data center and energy investments totaling roughly $90 billion. The major tech companies, including Meta, Microsoft, and Alphabet, were there along with energy giant Exxon Mobil and investment house BlackRock.
Sorting through the hyperbole, breathless announcements, and reports is not a simple matter. The frenzied pace of activity may well lead to some bad investments.
Artificial Intelligence Is Reshaping Demand
McKinsey & Co., in a 2025 report, estimated that data centers could require $6.7 trillion in global capital expenditures by 2030—with the bulk of that amount for AI demand. The firm also estimates data center demand to reach 80 gigawatts (GW) by 2030 (up from 25GW in 2024) in the US alone.
Already, the United States has installed more than 53 gigawatts of data center capacity—accounting for nearly nine percent of regional average power demand, according to a new estimate by the International Energy Agency. Much of this activity will be powered by natural gas—at least in the short term.
But national figures don’t capture the drama playing out on the ground. In Virginia, which hosts more than a fifth of the world’s hyperscale data centers, Dominion Energy tripled its load forecast through 2035 and pivoted from a clean energy roadmap to plans for nine gigawatts of new gas-fired plants. The company’s 2023 Integrated Resource Plan points to the scale of data center demand as a key driver in its strategy.
Energy Developers Are Racing to Catch Up
Recent business developments show that US energy developers may not be ready to fully take on the energy demand challenge. This is underscored by NRG Energy’s $12 billion acquisition of gas-fired power plants, as the Texas-based provider hopes to chip into the US’ data center heartland. Meanwhile, Microsoft and Meta are working to get old nuclear power plants back online and build new gas-fired plants as quickly as possible.
“The real constraints of the physical world will assert themselves over the dreams of the virtual one,” a Bloomberg columnist recently wrote to this point.
Now, local authorities are scrambling to best dictate who will get the energy of the future—and where that energy should come from. An excellent new report from E9 Insight, a regulatory research firm, details how this scramble is unfolding across the United States, revealing a patchwork of custom rate structures, long-term contracts, and controversial infrastructure proposals—often with significant implications for other electricity customers.
Data Centers Are Getting Their Own Rules
According to the E9 report, among the most striking developments is a proliferation of dedicated rate classes for hyperscale customers—typically those demanding five megawatts or more. These arrangements include 10- to 20-year contracts, minimum billing thresholds that guarantee utility revenues even if facilities are idle, and incentives for around-the-clock energy usage. The result is a regulatory environment increasingly tailored to the business models of artificial intelligence giants.
In Nevada, NV Energy’s Clean Transition Tariff allows large non-residential customers to source electricity from new clean resources while paying a hybrid of fixed and variable rates—a structure hailed as a potential national model. In Indiana, regulators approved a 12-year tariff for customers exceeding 70 megawatts, complete with exit fees and 80 percent minimum billing—but deferred the contentious question of who foots the bill for grid upgrades.
Other states are tightening scrutiny. In Minnesota, Amazon’s proposal to install 600 megawatts of diesel backup generators at a new data center was blocked by regulators, who argued the project required a formal certificate of public need. Meanwhile, Kentucky utilities are proposing $3.7 billion in new gas and battery projects to meet an estimated six gigawatts of upcoming demand from data centers. And Crusoe has secured nearly five GW of natural gas generators in a deal with Chevron, Engine 1, and GE Vernova.
Duke Energy has modeled an alternative: a concept called curtailment-enabled headroom, where modest, short-duration cutbacks in power usage could integrate large new loads without massive new generation. Yet, few utilities have adopted this approach in their formal Integrated Resource Plans.
This is a competitive race for building and managing infrastructure.
Who Benefits and Who Pays
The implications go beyond grid physics. Are ordinary ratepayers subsidizing upgrades that benefit a handful of hyperscale customers? Do opaque, bilateral deals between corporations and utilities undermine public oversight? And are long-term contracts locking the grid into different generation pathways than previously foreseen?
These questions are playing out globally as well. Tensions are high in Ireland around the addition of huge data centers on a small power system. And many countries across Asia are running into similar crossroads.
The stakes are reminiscent of past inflection points in the power sector—from the rise of electrified manufacturing in the early 1900s to the suburban explosion of air conditioning in the postwar era.
How Much Power Will Artificial Intelligence Really Use?
That raises the possibility that everyone is overestimating just how much energy artificial intelligence will end up using. Long-range energy forecasting is notoriously difficult, and in the past, this led to poor investments and power planning decisions. While current AI energy demands are unprecedented, there are already impressive advances and trends in efficient cooling systems, chips, and systems.
But this time, the change is more centralized, more capital-intensive, and more politically powerful. We saw that play out with President Donald Trump’s recent visit to the Middle East, signing some $200 billion of new bilateral deals with Gulf states that crucially focused on AI, supercomputing, and energy.
As artificial intelligence continues its breakneck march forward, the question isn’t whether the machines will run. It’s who—and what—gets left in the dark.
About the Author: Morgan Bazilian
Morgan D. Bazilian is the Director of the Payne Institute and Professor at the Colorado School of Mines, with over 30 years of experience in global energy policy and investment. A former World Bank lead energy specialist and senior diplomat at the UN, he has held roles at NREL, the Irish government, and advisory positions with the World Economic Forum and Oxford and Cambridge Universities.
Image: Meta.N/Shutterstock
