North Carolina Targets Hyperscale Costs with Proposed AI Infrastructure Bill

extract

North Carolina Targets Hyperscale Costs with Proposed AI Infrastructure Bill Proposed legislation shifts the burden of power, water, and grid expansion costs onto large data centers, challenging long-standing incentive models and reshaping site selection strategies. North Carolina lawmakers are moving to require large data centers to pay the full cost of the power, water, and infrastructure they consume, advancing legislation that would also limit their access to public incentives. The legislation reflects a growing trend among regulators to address the perceived imbalance between hyperscale demand and infrastructure capacity, as AI-driven workloads push utilities to their limits. The proposed Ratepayer and Resource Protection Act targets what lawmakers describe as cost-shifting from large data centers onto households and small businesses, as AI-driven demand accelerates power and water consumption across the state. The measure shifts from using incentives to attract data center projects to a full cost recovery model, placing the burden of grid expansion and resource use directly on data center operators. It follows similar action in Wisconsin, where regulators recently overhauled a proposed data center tariff to ensure large-load customers cover the full cost of generation and grid infrastructure. These developments signal a broader shift toward stricter frameworks to safeguard existing ratepayers while managing the rapid growth of hyperscale facilities. Full Cost Pricing The legislation requires utilities to charge large data centers rates that fully cover the marginal cost of service, including new generation, transmission, substations, and other infrastructure built to support their load. It also includes a “hold harmless” provision that prevents other customers from facing rate increases or financial risks tied to data center demand, including costs from underutilized or abandoned projects. That combination directly challenges a long-standing industry dynamic in which large-load customers benefit from shared infrastructure costs embedded in broader rate structures. Some hyperscale operators are already moving in this direction. Microsoft, Amazon Web Services, and Google have adopted dedicated power procurement, funded generation, and absorbed higher energy costs to secure capacity for large-scale builds. These deals reflect a growing willingness to pay for speed and certainty as AI demand accelerates. The proposal would turn that emerging practice into a baseline requirement. For utilities, the change reduces exposure to underutilized assets but limits flexibility in structuring large-load deals, raising questions about their ability to accommodate rapid demand growth under stricter cost allocation rules. For developers, meanwhile, it increases entry costs, especially for projects requiring new generation capacity or grid upgrades. The shift comes as power availability, not compute supply, becomes the primary constraint o AI infrastructure growth. Steven Dickens, CEO and principal analyst at HyperFrame Research, said recent earnings across the sector point to capacity constraints, with electricity generation and grid expansion lagging behind chip deployment. “This ties directly to the earnings cycle,” Dickens said. “Capacity is the binding constraint, and power generation is harder to spin up than chips.” On-Site Power Mandate The measure would require large data centers to install on-site clean generation equal to at least 25% of projected peak demand, ensuring the capacity is operational at launch. It explicitly excludes renewable energy credits and virtual power purchase agreements, requiring compliance through on-site generation.. The North Carolina Utilities Commission would retain authority to increase the required percentage if needed to maintain grid reliability. According to Dickens, the provision reflects a broader shift toward behind-the-meter power strategies, as utilities struggle to keep pace with large, concentrated AI workloads. “At the single-facility scale, it is doable,” he told Data Center Knowledge. “At campus scale across hundreds of megawatts, it becomes a serious engineering and permitting challenge.” He added that mandates without aligned utility frameworks could create stranded investment risk. The legislation also introduces new water use standards, requiring data centers to rely on closed-loop or reclaimed water systems and discouraging the use of evaporative cooling technologies. That restriction could reshape cooling design choices in the Southeastern US, where evaporative systems have been widely used to manage high thermal loads. Operators would also need to disclose both estimated and actual water use, along with cooling system efficiency metrics, in annual public reports. Incentives Cut In a significant shift from prior economic development strategy, the measure would make data centers in North Carolina ineligible for state and local incentives, including tax abatements and infrastructure subsidies funded directly or indirectly by ratepayers. In the Southeast, where tax incentives and utility partnerships have been central to attracting hyperscale development, removing those tools represents a structural break in how projects are planned and funded – particularly in a region that has relied on aggressive incentives to win large-scale deployments. At the local level, pressure is already building. Officials in Durham have proposed a moratorium on new data center development as they assess the impact of large-scale facilities on power availability, water resources, and land use. Rep. Lindsey Prather, a Democrat from Buncombe County and the bill’s sponsor, highlighted the increasing number of local governments delaying new projects as they review land-use rules. “North Carolinians want us to get ahead of this,” Prather said during a press conference. “If a company wants to build a data center in North Carolina, they can. They must simply pay their own way and stop shifting the cost onto you, me and families like ours.” The policy shift could influence the location of new capacity rather than its overall development. “It changes the site selection math, not the demand,” Dickens said. “These companies are going to build. The question is where.” If North Carolina becomes materially more expensive than other regions, he added, “the capacity goes elsewhere.” Hyperscale Threshold The bill defines a 'large data center' as one with 40 MW or greater peak demand or annual water use exceeding one billion liters (approximately 264 million gallons). That threshold aligns with the scale of modern AI infrastructure deployments, where single campuses can exceed traditional industrial loads and place concentrated pressure on regional grids. At that level, individual projects can drive the need for new generation capacity, transmission upgrades, and water system expansion. The threshold effectively targets the facilities most likely to trigger those system-wide investments and the associated cost-recovery debates. AI Drives Policy Shift The proposal lands as utilities and regulators across the US confront a widening gap between AI-driven demand growth and the pace of power and infrastructure buildout. By requiring data centers to internalize those costs, North Carolina is positioning itself at the forefront of a growing policy debate: who should pay for the AI buildout? Dickens said policies like the North Carolina proposal fail to address the root imbalance. “You cannot regulate your way out of a physics problem,” he said. “Demand is growing faster than the grid can expand.” He said similar pressure is already emerging in other markets, including Northern Virginia and parts of Europe, as infrastructure limits begin to constrain growth. “The pattern is the same,” he said. “Infrastructure hits a wall, regulators add friction, and capital moves to the path of least resistance.”