The rapid expansion of artificial intelligence data centers is creating a major new source of electricity demand, with potentially significant implications for utility finances. In an OilPrice column, Leonard Hyman and William Tilles examine whether utilities can tap that opportunity without shifting the burden onto ordinary ratepayers.
The authors focus on whether the GENCO model could offer a solution. GENCO, short for generating company, refers to independent unregulated power producers that spread after energy-market deregulation. Hyman and Tilles note that many of those businesses later went bankrupt because of excess capacity, weaker-than-expected demand and too much debt. But because local utilities typically did not directly own those GENCOs, utility customers were not forced to rescue them.
That history matters for the current AI buildout. If regulated utilities construct expensive new power plants specifically for AI facilities and place those investments into the rate base, electricity bills could rise for all customers. And if the projected AI demand never fully materializes or later drops sharply, the cost of that added capacity could still remain with other utility users.
One proposed workaround would be to create a separate GENCO affiliated with the utility and its holding company, while financing it with its own debt outside the utility balance sheet. In theory, that ring-fenced structure would shield the regulated utility and its customers if the project ran into trouble. The authors argue, however, that the outcome would depend heavily on the fine print.
For example, if a utility signs a long-term contract to buy electricity from an affiliated GENCO and then resells it to an AI center on a full cost-pass-through basis, a key risk remains: what happens if the utility’s purchase commitment outlasts the AI project’s actual life. In that case, the financial burden could still flow back to the utility and its customers. The same applies to cost definitions—if they are structured improperly, either consumers or the utility itself may end up covering the shortfall.
The article also argues that the issue goes beyond legal structuring. In the authors’ view, AI companies have ample resources to exert political influence and pursue litigation, while regulators may not always have the expertise needed to challenge complicated arrangements. In that setting, consumers could still be left exposed.
From a financial standpoint, the GENCO option may also be less clean than it appears. If a utility signs a power-purchase agreement with GENCO, the generator can use that contract as collateral to raise lower-cost debt, effectively benefiting from the utility’s credit standing. According to the authors, that credit quality is supported in part by utility customers, yet those customers receive no compensation for it. In addition, if GENCO runs into distress, it could weaken the holding company that owns both the generator and the regulated utility, potentially increasing the utility’s cost of capital.
The column also considers another route: building a GENCO outside the regulated utility and selling power directly to the AI center. Even then, the authors say, regulators would need to ensure the affiliated utility does not grant the generator special treatment that disadvantages other customers or competitors. They cite storm recovery as an example: if the utility’s affiliated GENCO and its AI client get service restored ahead of others, public backlash would be likely.
Hyman and Tilles highlight two additional risks that they say are often overlooked. The first is scale. AI companies are not asking for marginal amounts of electricity but for very large volumes, with major financial and grid implications. The second is environmental exposure: new gas-fired generation built to support AI demand could become politically vulnerable, especially if stricter clean air and water standards are enforced again at the national level.
The authors’ conclusion is that AI companies should ultimately build and operate their own off-grid power resources rather than transferring the financial, reliability and environmental risks of AI expansion onto captive customers of regulated utilities.
