Within days of each other in April 2026, a state legislature passed a first-of-its-kind ban on data center construction, and a trillion-dollar AI infrastructure project was quietly paused. Taken together, they signal something more consequential than a political backlash against big tech. They are some of the first systemic indications that the regulatory framework underlying the AI build-out is not keeping pace with the intentions of tech companies and major AI labs.
Maine became the first state in the country to pass a temporary statewide ban on new data center construction when its legislature approved LD 307. The bill, which now heads to Governor Janet Mills for signature, prohibits the development of any data center drawing 20 megawatts or more of power until November 2027. That threshold is roughly equivalent to the electricity demand of 15,000 to 20,000 homes. In many cities, that might be small. In Maine, a small load pocket in ISO-NE, that 20 megawatts can be material. The legislation also creates the Maine Data Center Coordination Council, tasked with studying the impact of data center development on electricity prices, water supplies, and grid reliability before the moratorium expires.
The same week, OpenAI confirmed it was pausing its Stargate UK initiative, a high-profile partnership with the British government aimed at establishing sovereign AI computing infrastructure in the United Kingdom. In a statement, OpenAI cited “the cost of energy and regulatory issues” as the reasons for the pause, saying it would move forward “when the right conditions enable long-term infrastructure investments.”
Neel Somani, a technologist and former quantitative researcher at Citadel, has been writing about the economics of AI infrastructure and the misalignment between where capital flows and where value is actually created. In his view, the energy problem and its regulatory implications are not temporary obstacles. They reflect deeper structural issues with how AI infrastructure has been planned and whether the grid is truly ready for the build-out.
The Math That Underpins The Decision
Maine’s decision did not come out of nowhere. The state already has some of the highest residential electricity rates in the country, and between 2021 and 2026, its total electricity bills surged roughly 60 percent. Distribution costs alone rose by approximately 140 percent over that period. When proposals for large data centers began in the state, including a $415 million underwater facility proposed off the coast near Eastport and projects in the former mill towns of Jay and Sanford, residents and lawmakers reacted with alarm.
Data centers are among the most energy-intensive industrial facilities. A single hyperscale facility can draw as much power as a small city. American data centers consumed approximately 183 terawatt-hours of electricity in 2024, accounting for more than 4 percent of total US power consumption. That figure is expected to more than double by 2030 as AI workloads expand.
The standard industry response to energy cost concerns has been to argue that data centers create jobs, generate tax revenue, and attract investment, and that companies building them often pledge to develop new power generation alongside their facilities. Maine lawmakers were skeptical. Seth Berry, executive director of Our Power, a nonprofit focused on energy policy, argued during the legislative debate that even company-supplied power generation would not be sufficient to prevent electricity costs from rising further for ordinary payers. The House rejected an amendment to create exemptions for specific projects by a vote of 115 to 29.
What Maine did, in effect, was force an analysis. Before any further large-scale data center construction proceeds, the state wants to know who will pay for the required infrastructure and whether the benefits justify the costs. Though the delay is painful, that is a reasonable question. Due to the geographically bespoke nature of power modeling, the question is also one that cannot be easily answered systematically at the state or federal level, which is part of why it has now arrived in the form of a legislative ban.
Stargate UK and the Global Energy Problem
The OpenAI pause on Stargate UK points to the same underlying dynamic playing out at a different scale. Stargate UK was announced with considerable fanfare in September 2025 as a partnership between OpenAI and the UK government designed to build sovereign AI computing infrastructure inside Britain. The goal was to allow the government to run advanced AI models locally, particularly for applications where jurisdiction and data residency matter. OpenAI simultaneously announced it would offer similar arrangements to other countries through an initiative called OpenAI for Countries, with Australia, Greece, the UAE, Slovakia, and Kazakhstan among the participants.
The pause, confirmed this week and attributed to energy costs and regulatory friction, possibly because building a data center in the United Kingdom means navigating UK energy markets, UK planning regulations, and UK grid constraints, which are significantly more restrictive and expensive than those in the US markets, where OpenAI and its infrastructure partners have built most of their capacity to date.
The UK is not unique in this respect. Energy costs for industrial users are substantially higher across Europe than in the United States. Regulatory timelines for large infrastructure projects are longer. Grid capacity constraints are more acute. The Stargate UK pause is a data point, but the dynamic it reveals may apply across the countries OpenAI for Countries is targeting.
The Structural Problem With How AI Infrastructure Has Been Built
Somani’s broader argument about the economics of AI infrastructure is that the current build-out has been predicated on assumptions about energy availability and cost that are now being tested in real time.
The business model of large-scale AI infrastructure depends on electricity being cheap, abundant, and reliably available at the locations where data centers are built. In the United States, this has been achievable in specific markets: Texas, Virginia, parts of the Southeast, and other regions with surplus power generation and favorable permitting. Virginia has the world’s largest concentration of data centers, and Texas is expected to surpass it.
But the economics that made those markets appealing are not infinitely scalable. As data center demand has grown, it has started to outpace local power generation in some markets, pushing up electricity prices and threatening grid stability for surrounding residents and businesses.
That dynamic is what triggered Maine’s backlash. It is what underlies the similar bills now progressing in at least a dozen other states, including New York, South Carolina, and Oklahoma. It is what Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez are attempting to address with a proposed federal moratorium. And it is what OpenAI ran into when it tried to extend its infrastructure model to the United Kingdom.
The core issue, as Somani frames it, is that AI infrastructure is both a technology and a finance problem, as well as an energy infrastructure problem. Capital flows into data center construction have vastly outpaced investment in the power generation and transmission infrastructure needed to support them. Since power generation often has a direct beneficiary, that beneficiary can be charged correspondingly via contractual ties. Socialized costs, such as distribution, on the other hand, cannot be allocated so easily. The gap between the two investment curves is now manifesting as political resistance, regulatory obstacles, and halted projects.
What Comes Next
Maine’s moratorium is temporary. The bill mandates a council report by February 2027, and expires in November of that year. The intent, as articulated by the bill’s sponsor, Representative Melanie Sachs, is to create space for the state to develop a regulatory framework that allows data center development to proceed on terms that protect ratepayers and grid reliability.
That framing is worth taking seriously as a model. Based on the language of the legislation, a blanket ban is not a permanent outcome that most legislators in Maine are aiming for. What they want is a framework that answers the energy-cost question before construction begins, rather than after.
For the AI industry, the more urgent problem may be the global version of the same dynamic. If Stargate UK is paused because energy costs and regulatory conditions in the UK do not support the economics of large-scale data center investment, that might be true for other parts of the world. The OpenAI for Countries initiative envisions AI infrastructure across dozens of sovereign jurisdictions. If the economics do not work in Britain, the pipeline of similar agreements could face the same friction.
Somani’s interpretation of the situation is that the AI infrastructure build-out is entering a phase where energy infrastructure can no longer be treated as a secondary consideration. The companies and investors best positioned for the next phase of this build-out will be the ones that treat energy as a primary constraint, not an afterthought. That means investing in power generation alongside data center capacity, engaging with grid operators and regulators earlier in the development process, and being more selective about which markets can actually support the infrastructure economics required by large-scale AI.
