New York Imposes One-Year Moratorium on Large AI Data Centers Amid Power and Water Concerns

New York State has become the first in the U.S. to enact a statewide ban on new large-scale AI data centers, signing an executive order that halts permits for facilities requiring 50 megawatts (MW) or more of power for one year. The decision, announced by Governor Kathy Hochul on July 14, marks a significant escalation in the regulatory response to the environmental and infrastructure pressures driven by the rapid expansion of artificial intelligence. It comes on the same day that reports emerged of SpaceX's xAI operating dozens of unpermitted gas turbines at its Colossus 2 facility in Tennessee, underscoring the mounting tension between AI development and local resources.

Unlike previous restrictions, which were limited to individual cities or counties in states like Virginia, New York's moratorium represents the first legislative and administrative action at the state level. The order declares that the pace of AI-driven electricity and water consumption has become unsustainable and that new permits should be suspended until appropriate regulatory safeguards are established. The New York State Legislature has already passed the Responsible Data Center Development Act, though Governor Hochul has not yet signed it into law. The move is expected to intensify the national debate over how to balance the economic promise of AI with its growing environmental footprint.

Governor Hochul emphasized that technological progress should not come at the expense of higher utility bills, depleted water supplies, or increased noise pollution. The Data Center Coalition, an industry group, criticized the decision, arguing that it could permanently shift technology investment and jobs from New York to states with less restrictive regulations, such as Virginia and Texas. Local opposition has largely centered on the rising electricity costs associated with connecting hyperscale facilities to the grid, which often require new high-voltage transmission lines and substations. Under many U.S. electricity pricing systems, these infrastructure costs are ultimately passed on to residential and commercial customers, raising bills for the broader community. Wholesale markets operated by regional transmission organizations like PJM and ERCOT can also drive up prices when sustained demand from data centers forces the dispatch of more expensive gas- or coal-fired generators.

In response to growing public concern, some hyperscale operators have introduced community support measures. Meta, for example, recently agreed with Entergy Louisiana to bear the full cost of power generation and grid expansion for its planned 5-gigawatt Louisiana data center. The project includes the construction of seven natural gas power plants, three battery energy storage systems, and additional nuclear capacity. Meta stated that the arrangement could reduce electricity costs for local residents by a total of $2 billion over the next 20 years, and it has committed $215 million to a program supporting low-income households' electricity costs, alongside over $1 billion in direct local infrastructure investment.

On the same day as the New York announcement, Reuters reported that xAI had been operating 59 natural gas turbines without air emissions permits to power its Colossus 2 supercomputer in Memphis, Tennessee. According to the report, the turbines were installed without completing environmental reviews or public consultation procedures. xAI has argued that the equipment qualifies as portable and is therefore exempt from permitting requirements, but local residents have filed lawsuits claiming the setup functions as a permanent power plant. Reuters' analysis indicated that continuous operation could release approximately 2,500 tons of nitrogen oxides, 4,000 tons of carbon monoxide, and 22 tons of formaldehyde annually, far exceeding federal permitting thresholds. The emissions are concentrated near residential neighborhoods and elementary schools in areas with higher proportions of minority residents and elevated rates of respiratory illness. The U.S. Department of Justice and Department of Defense have intervened in support of xAI, arguing that the facility is critical infrastructure for training the Grok AI model and that delays could weaken U.S. competitiveness against China.

The challenges of powering AI data centers—requiring low-cost, low-carbon, and uninterrupted electricity—are driving the industry toward nuclear and hydroelectric power. Countries like Norway and Sweden have become attractive due to abundant renewable energy and naturally cold climates, which help achieve power usage effectiveness (PUE) values as low as 1.07. Nuclear power, despite its waste management challenges, remains one of the few reliable sources of baseload carbon-free electricity. Based on South Korea's APR1400 reactor design, each 1.4-gigawatt reactor could roughly support a 1-gigawatt AI data center. Meta's planned 5-gigawatt Hyperion facility would require the equivalent of four to five such reactors. Similarly, SK Group's plan to build 15 gigawatts of AI data center capacity would require electricity equivalent to roughly 11 APR1400 reactors, with the company evaluating locations based primarily on available power infrastructure.

Looking further ahead, two technologies are being discussed as long-term solutions: satellite-based data centers and commercial fusion energy. Although fusion power has yet to be commercialized, it is attracting significant interest for its potential to provide carbon-free, stable baseload electricity for future data centers requiring 10 to 20 gigawatts. Most fusion developers target demonstration projects around 2030 and commercial deployment by approximately 2035. Microsoft has signed a future power purchase agreement with Helion Energy, Google has invested in Commonwealth Fusion Systems, and OpenAI CEO Sam Altman has personally invested in Helion. General Fusion, which listed on Nasdaq on July 13, became the world's first publicly traded pure-play fusion company. Despite being pre-revenue and yet to demonstrate net energy gain, its shares rose 38% in the first two trading days. Speaking at SoftBank World, Masayoshi Son projected that annual AI development spending could reach $5 trillion by 2040 and that AI data centers would ultimately require 3 terawatts of generating capacity, with fusion becoming the dominant long-term energy source.