Meet the ‘mini Erin Brockoviches’ fighting off eminent domain—and the shadow of Amazon—as a data center hands out $10,000 to their neighbors

Amazon-linked Project Hazelnut, a proposed 1,300-acre, 15-building Pennsylvania data-center campus backed by a $165 million local investment proposal, faces mounting community, zoning and utility-line opposition. A Luzerne County judge denied a project appeal in May over missing zoning approval and inadequate public input, while PUC proceedings on an associated 11.2-mile high-voltage line were temporarily halted. The setback highlights execution, permitting and power-access risks to Amazon's broader $20 billion Pennsylvania AI infrastructure investment, even as the company says it has no definitive agreement for the site.

Analysis

The investable issue is not one Pennsylvania campus; it is a widening permitting-and-interconnection discount on hyperscaler capacity plans. For AMZN, delays matter less through the next quarter than over 6-18 months: constrained regional AWS capacity can defer AI workload revenue while land, transmission and pre-development capital earn no return. The more material read-through is to the power stack: regulated utilities with approved, contracted load retain upside, while developers relying on speculative hyperscaler demand face a higher probability of stranded transmission and land assets.

Local control creates an uneven competitive outcome. Incumbent data-center operators with existing powered campuses and entitled land—EQIX, DLR and private CoreWeave-type capacity—should gain pricing power if new greenfield supply is delayed; however, their own expansion multiples are vulnerable where power delivery dates are uncontracted. PPL Corp. (PPL) faces modest project-specific earnings risk but a potentially meaningful regulatory-precedent risk: rejected cost recovery or prolonged siting litigation would raise the required return and timeline for distribution/transmission upgrades across its Pennsylvania service territory.

Consensus still treats AI infrastructure as a capital-expenditure availability story. The bottleneck is increasingly social license plus local zoning, which can make announced capex materially less convertible into commissioned megawatts. AMZN's transparency shift reduces reputational friction but does not solve water, grid, and eminent-domain objections; the relevant catalyst is whether management discloses usable AWS capacity additions and regional power commitments rather than aggregate capex.

Near term, this is not sufficient to trade AMZN outright: the project remains non-definitive and its financial exposure is immaterial versus consolidated capex. Over 1-3 months, track Pennsylvania PUC proceedings, county zoning outcomes and comparable local project cancellations as leading indicators of construction-start slippage. A reversal would be binding utility interconnection agreements, locally negotiated community-benefit packages, or state legislation preempting municipal restrictions.

Key Decisions for Investors

• Maintain AMZN core exposure but do not add on this headline; use the next earnings call to demand evidence of commissioned AWS capacity and utilization versus announced capex. Reassess bearish exposure only if management signals material regional capacity deferrals or capex rises without corresponding AWS revenue/guidance support.
• Prefer a 6-12 month long DLR / short AMZN relative-value watch trade only after verifying DLR's incremental capacity is already powered and permitted. Thesis: scarcity supports wholesale lease rates; invalidate if DLR's signed backlog, power-delivery dates, or development yields deteriorate.
• Avoid adding PPL into unresolved PUC siting decisions. Set an alert for a final adverse eminent-domain or cost-recovery ruling; a broader precedent against transmission buildout would warrant underweighting PPL versus regulated peers with cleaner load-growth pipelines such as AEP or SO.
• Monitor equipment suppliers with high data-center exposure—VRT, ETN and GEV—for order-to-revenue conversion rather than bookings. A cluster of permitting delays over the next two quarters would favor trimming names priced for uninterrupted 2027-28 build schedules; strong backlog conversion would falsify the bottleneck thesis.