Middlesex Township is a quiet municipality of roughly 14,000 people in Cumberland County, Pennsylvania, unremarkable in most respects — except that it is now at the centre of a national debate about who pays the true cost of artificial intelligence.
A joint venture called Carlisle Development Partners — comprising Pennsylvania Data Center Partners and PowerHouse Data Centers — has secured a $14.1 million agreement with the township granting its 18-building AI data center facility access to municipal water and sewer infrastructure. Under the terms, the complex may consume up to 400,000 gallons of water daily. That figure represents approximately 40 percent of Middlesex Township's excess water supply — the equivalent of the daily consumption of roughly 2,367 homes.
Township Supervisor Phil Neiderer confirmed the deal, saying the revenue would "fund a lot of projects that have already been in the books that are completely unrelated to the data center." The revenue argument is not trivial for a small municipality with limited fiscal options. But residents have raised pointed questions about the long-term implications: rising utility rates, potential strain on aging water infrastructure, and the precedent of dedicating a major share of a public utility to a single private tenant whose water needs, unlike those of a household or small business, may expand indefinitely.
The Hidden Thirst of the AI Economy
The Middlesex Township deal is not an anomaly. It is a preview.
Data centers cool their server racks with water — enormous volumes of it, circulated through cooling towers that evaporate much of it into the atmosphere in a process distinct from, and in addition to, the electricity they consume. The AI revolution has dramatically accelerated data center construction and scale, and with it, water demand. A peer-reviewed study in Nature Sustainability warns that data centers could soon consume water equivalent to the annual usage of 10 million Americans and emit carbon at the scale of 10 million vehicles.
A single large-scale AI training run — of the kind used to develop models underpinning widely used AI assistants — can consume hundreds of thousands of litres of water for cooling. Multiply that by the number of queries processed daily across global AI infrastructure and the aggregate demand becomes significant. Unlike electricity consumption, which is directly metered and billed, water use by data centers has received comparatively little regulatory or public scrutiny.
The grid implications compound the concern. The PJM Interconnection, the electricity market operator serving over 65 million people across 13 states including Pennsylvania, projects a 6-gigawatt power shortage by 2027, driven substantially by surging data center demand. An independent market monitor has described the regional grid situation as "at a crisis stage." Data centers are not only competing with residents for water — they are competing for electricity on an increasingly strained system.
Climate Justice in a Small Town
For communities like Middlesex Township, the calculus is complex. The $14.1 million payment is real money for a small municipality. But the transaction raises questions that extend beyond local budgeting.
Public water systems are, by design, held in trust for residents. They are built with public investment, maintained through rate payments, and governed as a collective resource. The effective wholesaling of a major share of a public utility's surplus to a private AI infrastructure company represents a qualitative shift in how that resource is managed, even if the transaction is technically legal.
Residents interviewed by local advocates have expressed concern about rate increases that could follow as infrastructure ages and maintenance costs rise. Water pricing in U.S. municipalities is notoriously regressive: lower-income households spend a disproportionately higher share of their income on water bills than wealthier ones. If a data center deal eventually drives rate increases to cover infrastructure upgrades, the burden falls unevenly.
The climate dimension compounds the justice question. Data centers concentrate their physical footprints in communities that lack the political leverage to resist them while offering tax revenues and employment that local governments genuinely need. The net result is a pattern in which the infrastructure costs of the AI economy are socialised into working-class municipalities while the profits accrue to large technology firms and their shareholders in distant cities.
The Regulatory Vacuum
No federal framework currently requires data center operators to disclose water consumption, account for it in environmental impact assessments, or offset demand increases on municipal systems. Several states are beginning to examine data center water reporting requirements, but legislation has advanced slowly relative to the pace of AI infrastructure buildout.
Environmental advocates argue the gap is urgent. "We're locking in infrastructure commitments for the next 20 to 30 years right now," said one water policy researcher familiar with similar cases across the United States. "The permitting decisions being made in townships like Middlesex will define the water footprint of the AI economy for decades."
Township Supervisor Neiderer and officials have defended the deal's financial merits and indicated that engineering reviews assessed the water system's capacity before the agreement was finalised. The data center developer has not responded publicly to questions about long-term water projections or contingency plans in drought conditions.
In climate policy, as across environmental challenges, urgency must meet solutions — science demands action, but despair achieves nothing. Middlesex Township's deal may fund parks, roads, and municipal services its residents genuinely need. But the broader pattern it represents — a technology industry consuming shared public resources faster than governance can respond — demands the kind of national policy framework that has not yet arrived.
