America’s Data Center Siting Bet Is Built on Fossil Ground?: the real signal is the immediate adjustment required in cash, risk, and execution
Signals That Are Accumulating
Three distinct lines of evidence point in the same direction. A 2026 academic mapping study published in Nature Cities found that US data center siting correlates systematically with legacy urban infrastructure and fossil fuel industrial footprint—not with renewable energy availability or grid decarbonization timelines. The geographic pull toward established power corridors, substations, and transmission capacity is functionally reinforcing the fossil fuel grid rather than transitioning away from it.
Power infrastructure constraints have become the dominant bottleneck for new capacity. Roughly half of planned US data center builds in 2026 face delay or cancellation, driven by shortages of power infrastructure and critical electrical components. The transformer supply chain is a specific chokepoint: delivery timelines that ran under 30 months before 2020 have reportedly stretched toward five years in some cases. A meaningful share of US transformer and switchgear imports remains dependent on Chinese manufacturing, according to industry reporting referenced in the academic source material.
The fossil fuel dependency is not incidental to these constraints—it is structural. Reporting cited in the research documents Georgia Power keeping coal plants online specifically to serve data center load growth, a Pennsylvania gas project described as the largest-ever US facility of its type planned explicitly to power AI data centers, and peaker plants being brought back into service as AI demand exceeds clean dispatch capacity. Nuclear restarts—Three Mile Island for Microsoft, a separate plant for Meta—signal willingness to pay for firm, low-carbon power, but they remain limited relative to overall growth. The aggregate picture, which the Nature Cities mapping work attempts to quantify for the first time, is that AI capacity growth is being physically anchored to the fossil fuel grid by the path-dependent logic of infrastructure availability.
Why No One Is Naming It Yet
The pattern is easy to miss for two structural reasons. At the project level, each siting decision looks like a reasonable response to power availability, interconnection timelines, or tax incentives—not a portfolio-level commitment to fossil-grid dependency. The academic mapping work is among the first systematic attempts to see the aggregate effect rather than the individual project rationale.
At the communications level, hyperscaler renewable energy announcements—PPAs, REC purchases, 24/7 CFE commitments—dominate the public narrative. Those commitments represent real procurement activity, but the underlying grid mix supporting actual uptime may diverge significantly from the purchased instruments. When a facility sits on a fossil-heavy subregional grid while purchasing RECs sourced elsewhere, carbon accounting can pass external scrutiny while physical energy supply remains carbon-intensive on a location-based basis.
State regulatory attention has not yet crystallized into a unified operating constraint. The research cites emerging 2026 state legislation on data center energy and tax issues, but the framework remains fragmented across jurisdictions. That asymmetry is the core risk: physical siting decisions made now will lock in grid relationships for 15 to 25 years, while the regulatory environment that will eventually govern those relationships is still forming. Operators acting on current regulatory signals are optimizing for a framework that will likely look different at the midpoint of asset life.
What Happens If the Pattern Continues
If AI capacity continues clustering around legacy fossil infrastructure, three forward pressures become harder to manage on a portfolio basis.
Carbon accounting exposure is the most direct. As CDP, GRI, and potential regulatory frameworks tighten around Scope 2 methodology—particularly any shift from market-based toward location-based accounting—data centers physically co-located with or dependent on coal and gas dispatch will face growing gaps between reported and actual carbon intensity. Operators who built long-term PPA portfolios without addressing location-based exposure will find that gap difficult to close at renewal without significant additional cost or contract restructuring.
Cost volatility on the fossil side is a second-order pressure. Peaker plants and aging coal facilities brought back into service carry variable cost structures that natural gas price movements amplify. If those facilities become a meaningful share of actual dispatch supporting uptime commitments, the energy cost base acquires commodity exposure that was not underwritten in original site models—and that sits outside the PPA hedging strategy.
Community and political risk is the third front. Survey and academic research cited in the study documents growing local opposition to data center environmental impacts even as state-level voter awareness remains low. Municipal zoning resistance is already shaping development in Northern Virginia. The fossil fuel siting pattern adds a visible environmental narrative that local opposition can anchor to, and infrastructure reporting from rural Virginia and rural America more broadly suggests the friction is moving beyond concentrated markets.
What You Can Do Before It Is Obvious
The window before these pressures arrive as hard constraints is narrowing, but it remains open for operators who move now rather than at the next planning cycle.
A portfolio-level grid mix audit against location-based Scope 2 methodology is the first move. EPA eGRID subregional data allows existing and planned facilities to be mapped against actual generation mix. If the portfolio is disproportionately connected to fossil-heavy subregions, that exposure surfaces before regulators, investors, or auditors ask for it—and it clarifies where REC purchasing is substituting for CFE matching rather than complementing it.
Interconnection sequencing that prioritizes clean capacity corridors ahead of demand is the second. The research pattern suggests that operators who wait for the build-out to mature before seeking clean interconnection will find queue positions have moved even further ahead of them. Queue positions secured now in markets with improving renewable capacity factors carry optionality that is not recoverable once a facility is sited on a fossil-adjacent grid.
The third is active engagement with state data center energy legislation as it develops. The research cites MultiState tracking of 2026 state bills on energy and tax issues for data centers. Operators with regulatory monitoring programs in place can shape provisions before they harden—a significantly more efficient use of influence than responding after rules are finalized and physical assets are already committed.
Sources
- Nature — Urban infrastructure and fossil fuel industrial legacy drive US data center siting (Link)
