In 2020, a pulp digester explosion shut down the Androscoggin paper mill in Jay, Maine for good. At its peak, the mill had employed about 1,500 people in a town 67 miles northwest of Portland. Three years later, a joint venture bought the 1.4 million-square-foot facility with plans to convert it into a data center - one of dozens of similar conversions happening across rural America as the AI industry scrambles for compute capacity.
The pattern has a shape now. A manufacturing plant closes. The community spends years in economic decline. A data center developer arrives with capital and promises of jobs. Whether that trade actually works for the community depends on math that press releases usually leave out.
The Infrastructure Fit Is Real
Former industrial sites offer data center developers something that's genuinely hard to find: existing high-capacity power infrastructure. Heavy manufacturing ran on serious, continuous power. A paper mill or steel plant connected to the grid at scales most new commercial developments never need - and rebuilding that electrical capacity from scratch is expensive and slow. Getting a new grid interconnection approved and built can take several years right now, given how backed up utility queues are.
Large footprints, existing utility connections, industrial zoning, and low land costs all stack in favor of these sites. For companies trying to build AI infrastructure fast - model training and inference (running models to generate responses for users) require enormous amounts of compute, which requires power and physical space - former industrial sites let developers sidestep multiple bottlenecks at once.
Rural locations also face less organized community opposition than metro areas. A town that lost its anchor employer a decade ago has different politics around a large industrial tenant than a suburb already dealing with traffic and noise.
The Jobs Math Is Different
A paper mill employing 1,500 people is not the same economic engine as a data center. Data centers are capital-intensive and labor-light. A facility housing tens of thousands of servers might permanently employ 50 to 200 people - facilities technicians, security staff, network engineers. The construction phase runs higher and provides real employment, but construction ends.
This doesn't make data center development bad for rural communities. Tax revenue is real. Infrastructure investment is real. But community leaders being told "data center" as a replacement for a closed manufacturing facility should pressure developers for specific numbers before signing anything: permanent jobs, wage ranges, local hiring commitments, and what happens to the tax arrangement after initial incentive periods expire.
The AI industry needs data center capacity urgently enough that it will build wherever the conditions are right. Rural industrial sites are going to keep getting converted. Whether the communities hosting this infrastructure end up better off will depend almost entirely on the deal terms they negotiate before the ground breaks.