Why Data Centers Are Shifting Next to Power Plants

Amazon is building a data center directly beside a nuclear power plant in Berwick, Pennsylvania. Not nearby. Not in the same county. *Next to it* — close enough that the facility shares a fence line with the Susquehanna nuclear generating station.

That's not a coincidence, and it's not a one-off. It's a signal.

Across North America, hyperscalers and colocation operators are quietly repositioning their site selection criteria. The old playbook — find cheap land, build near fiber, stay close to major metros — is being rewritten. Power is the new prime location factor. And increasingly, that means data centers are moving to where power is generated rather than where it's consumed.

Here's why that shift is accelerating, what it costs to ignore it, and what it means for everyone involved in infrastructure development.

The Math Behind the Move

Data centers are not modest power consumers. A single hyperscale facility can draw 100 to 500 megawatts continuously — roughly equivalent to powering a mid-sized American city. When you're operating at that scale, the distance between your servers and your power source isn't an abstract engineering concern. It's a balance sheet issue.

Transmission losses, grid interconnection queues, and utility rate structures all penalize facilities that sit far from generation assets. Moving even 50 miles closer to a power plant can meaningfully reduce transmission costs, cut exposure to grid congestion pricing, and — critically — shorten the timeline to actually get power delivered.

That last point is underappreciated. The U.S. grid interconnection queue is genuinely broken right now. Projects are waiting five to ten years for grid connection approvals in many regions. A data center co-located at or near an existing generation site can potentially negotiate a direct power purchase agreement or behind-the-meter connection that sidesteps the queue entirely. That's not a minor advantage. In a market where AI compute demand is doubling roughly every two years, shaving three to five years off your power delivery timeline is worth an enormous amount of money.

The Susquehanna Blueprint

The AWS project in Berwick is worth examining in detail because it illustrates how sophisticated the new site selection logic has become.

Susquehanna is a two-unit boiling water reactor operated by Talen Energy, with a combined capacity of approximately 2,500 megawatts. It generates carbon-free baseload power around the clock — no intermittency, no weather dependence, no fuel price volatility. For a company like Amazon that has made aggressive public commitments around clean energy, nuclear power provides something wind and solar fundamentally cannot: guaranteed uptime.

AWS reportedly entered into a direct agreement with Talen Energy for up to 960 megawatts of nuclear power capacity. That deal — structured as a campus power agreement — would allow the data center to draw power directly from the plant's output, bypassing the broader grid for much of its load. If that model proves workable at scale, it could reshape how the entire industry thinks about energy procurement.

The arrangement drew regulatory scrutiny from PJM, the regional grid operator, over concerns that pulling large loads off the grid could affect reliability for surrounding customers. That's a legitimate tension — one that other co-location projects will face. But the fact that AWS pushed forward anyway tells you how valuable the model is to them.

Reliability Is the Real Currency

Energy cost gets most of the attention in these conversations. Reliability deserves more.

A data center that goes dark costs real money — not just in lost compute time, but in SLA penalties, customer churn, and reputational damage. The closer a facility sits to its generation source, the fewer points of failure exist between the electrons leaving a turbine and the servers processing a query.

Grid-connected facilities in congested regions are exposed to demand spikes, weather events, and transmission failures that a behind-the-meter or near-generation facility can largely avoid. Proximity to power isn't just about cost efficiency — it's about building infrastructure that doesn't fail when the grid gets stressed.

This reliability premium is particularly acute for AI workloads. Training large language models requires sustained, uninterrupted compute over days or weeks. A six-hour outage doesn't just pause the job — depending on checkpointing practices, it can wipe out significant progress. The operational value of stable, guaranteed power is measurably higher for these workloads than for traditional enterprise IT.

The Regulatory Obstacle Course

None of this is simple to execute. Co-locating with generation assets introduces a regulatory complexity that standard data center development doesn't face.

Nuclear sites carry federal licensing implications managed by the Nuclear Regulatory Commission. Any construction in proximity to a licensed nuclear facility triggers additional review processes that don't apply to a greenfield industrial development. Zoning in rural Pennsylvania wasn't designed with hyperscale data centers in mind — local ordinances, water rights for cooling systems, and environmental impact reviews all require navigation.

Then there's the PJM dimension. Regional transmission organizations have legitimate authority over how large loads connect to and interact with the grid. The AWS-Talen deal ran into exactly this friction. PJM's concern — that diverting nuclear output to a private campus load could reduce supply available to the broader grid — reflects a real policy tension between enabling innovative energy structures and maintaining grid reliability for everyone else.

Expect this regulatory question to become one of the defining infrastructure debates of the next decade. As more operators attempt direct-power arrangements with generation assets, grid operators and federal regulators will be forced to develop clearer frameworks. Early movers like AWS are essentially paying to establish the rules of the road.

Permitting timelines for co-located projects should be budgeted at 18 to 36 months minimum, and project teams need legal and regulatory expertise specifically in energy law — not just standard data center development counsel.

The Carbon Calculus

The environmental picture here is more nuanced than most coverage acknowledges.

Nuclear power is genuinely low-carbon on a lifecycle basis — comparable to wind and solar when you account for construction and fuel cycle emissions. For tech companies under pressure to decarbonize their operations, securing a direct offtake from a nuclear plant is arguably a cleaner path than buying renewable energy credits from wind farms hundreds of miles away.

That said, there's a legitimate question about additionality. If a data center is pulling power from an existing nuclear plant that was already supplying the grid, it isn't necessarily creating new clean energy — it's redirecting existing clean energy away from other consumers who may then need to source from less clean alternatives. This is the same debate that surrounds tech companies buying renewable PPAs from projects that would have been built anyway.

The cleaner version of this story is when the data center's power demand directly funds new generation capacity. Some projects are heading in that direction — pairing data center development with new nuclear construction or next-generation SMR (small modular reactor) deployments. That's still years away from commercial reality at scale, but the investment signals are pointing there.

What This Means for Infrastructure Investment

For anyone involved in land acquisition, infrastructure development, or energy project finance, the implications are significant and largely underpriced by the market.

Land within transmission reach of major generation assets — particularly nuclear, large hydro, and utility-scale solar — is going to attract serious developer attention over the next five years. Parcels that were valued purely as industrial or agricultural land may carry new optionality as potential data center sites if they meet the proximity and grid connection criteria.

The competitive advantage in this market will go to developers and investors who understand both the energy and data center sides of the equation — not just one or the other.

Rural communities near power plants, which have often struggled with economic development, stand to gain substantially if they can successfully attract data center investment. The tax base and employment implications of a 200-megawatt facility are transformative for a small municipality. Local governments that streamline permitting and proactively engage with developers will capture more of this investment than those that treat it as a standard zoning matter.

The AWS project in Berwick is the visible tip of a much larger reorientation in how data center infrastructure gets built and where it gets built. Power availability has quietly displaced fiber connectivity and real estate cost as the primary site selection constraint. The developers, investors, and communities that internalize that shift first will be the ones positioning assets — and deals — ahead of where the market is heading.

The power plant isn't just a neighbor anymore. It's the address.

[CONSIDER CUTTING]

Call to Action: Ready to explore the future of data centers and energy? Discover opportunities on the InfraSale Marketplace: https://infrasale.com/marketplace.

Internal Link Suggestions:

• [INTERNAL LINK: data center site selection]
• [INTERNAL LINK: energy procurement strategies]
• [INTERNAL LINK: infrastructure investment trends]