The meeting was scheduled for seven in the evening in a school gymnasium in Chester County, and by six-thirty the parking lot was full. More than three hundred people had shown up — farmers, retirees, school board members, a retired electrician who drove forty minutes each way — to talk about something that had not existed as a concept in their vocabulary five years ago: the data centre. Two proposals were on the table, one from a hyperscale cloud provider and one from a smaller colocation operator, together representing approximately 800 megawatts of proposed electrical load in a county whose entire residential and commercial demand currently peaks at around 600 megawatts. The room, predictably, did not stay civil for long.
The Pennsylvania scene is playing out with minor variations across every region of the United States where the arithmetic of AI infrastructure intersects with the reality of existing electrical grids, water supplies, and local zoning codes. The artificial intelligence boom has created a demand signal for computing capacity so large and so sudden that it has overwhelmed the planning assumptions of utilities, municipalities, and transmission operators simultaneously. A data centre that would have been considered large at 50 megawatts five years ago is now a routine project specification; facilities of 500 megawatts to one gigawatt are in active development in multiple US states. The cumulative electrical demand of announced projects exceeds the generating capacity of several mid-sized national grids.
The complaints at the Chester County town hall were concrete and specific. A dairy farmer named Glen Hostetler said his family had watched electrical reliability deteriorate over the previous eighteen months as a nearby facility came online and drew unpredictably on distribution infrastructure that was designed for a different load profile. A representative from the local watershed authority raised the issue of cooling water consumption — hyperscale data centres can consume millions of gallons of water daily in evaporative cooling towers — in a region already managing drought stress. A high school teacher asked, with some heat, what proportion of local property tax revenue would actually remain in Chester County once federal depreciation schedules and negotiated assessment agreements were factored in.
“People are not opposed to technology,” said Councilwoman Renata Okonkwo, who chairs the county’s land use committee and organised the meeting. “They are opposed to bearing costs that get socialised while benefits get privatised. That is not a technology question. That is a governance question.” Her framing resonated with attendees in a way that the operators’ prepared presentations — which emphasised job creation figures and capital investment totals — conspicuously did not.
The governance gap is real and structural. Data centres, when they emerged as an asset class in the early 2000s, were classified under industrial or commercial zoning designations that were designed for warehouses and light manufacturing. The regulatory frameworks that govern their development have not kept pace with their scale, their energy intensity, or their water consumption. Most US states lack specific data centre impact assessment requirements. Utility interconnection queues — the formal process by which new large loads connect to the grid — were designed for industrial facilities that take years to plan and build, not for hyperscale operators who can assemble a facility in eighteen months.
The economic case for data centres is not fictional. A facility of 200 megawatts typically employs 50 to 200 permanent staff — a low job density relative to its capital base — but generates significant property tax revenue, supports a broader local construction economy during the build phase, and, in some configurations, co-invests in grid infrastructure upgrades that benefit other ratepayers. The question is not whether data centres create value, but whether existing frameworks for negotiating the distribution of that value between operators, utilities, and host communities are adequate. The Chester County residents would argue, with considerable evidence, that they are not.
For markets outside the United States that are actively courting data centre investment — and the UAE is explicitly and aggressively doing so, with both Abu Dhabi and Dubai offering incentive packages to hyperscale operators — the Pennsylvania experience offers a warning worth heeding in advance rather than retrospect. The infrastructure demands of AI-scale computing are not compatible with planning frameworks designed for a previous generation of industrial development. Jurisdictions that establish clear, community-informed impact assessment frameworks before the proposals arrive will be better positioned than those that negotiate each facility under ad hoc conditions that favour whichever party holds more information at the table.
Back in Chester County, the meeting ended without a vote. That is how town halls work: they are not decision-making forums but sentiment-registering ones. The operators’ representatives left with a clearer understanding of what they were dealing with. Whether that understanding translates into better community engagement, modified facility designs, or simply more effective public relations strategy will depend on choices that will be made in boardrooms, not gymnasiums. The residents of Chester County are hoping it is not the latter. History suggests their scepticism is not unwarranted.