Site Selection

Key Takeaways

• Land cost is less than 1% of project cost, yet site selection is critical
• Greenfield (farmland) is preferred over brownfield (former industrial) despite policy goals
• Brownfield barriers: contamination liability, timeline uncertainty, infrastructure gaps
• Transmission corridor proximity matters more than land price

The Farmland Paradox

In October 2021, Meta announced plans for a $800 million data center in DeKalb, Illinois—on 381 acres of former cornfield. Six months later, Google selected a 350-acre agricultural site in Columbus, Ohio for a $600 million facility. Microsoft built its Iowa complex on soybean fields. Amazon's Virginia campuses replaced dairy farms.

The pattern is consistent and puzzling: the world's most technologically sophisticated companies, capable of building billion-dollar facilities with extraordinary precision, overwhelmingly choose to build on farmland. Meanwhile, old steel mills, former auto plants, and decommissioned industrial sites—the brownfield locations that policymakers actively incentivize for redevelopment—sit mostly unused.

Why would developers pay to demolish functioning agricultural operations when contaminated former industrial sites are available, often at lower cost and with government support for cleanup?

The answer reveals a fundamental mismatch between policy intent and market reality. And it starts with understanding what site selection actually optimizes for.

Land Economics at Scale

Consider the Saline Township project in Michigan: a proposed 1.4 gigawatt facility on 640 acres of farmland. The developer paid approximately $8 million for the land—roughly $12,500 per acre. The total project cost: $7 billion.

Land represents 0.11% of project cost.

Even if the developer had paid ten times as much—$80 million for the land—it would still be barely 1% of total cost. For comparison:

• Building construction: 15-20% of project cost
• IT equipment (servers, networking): 50-60%
• Power infrastructure (generators, substations): 10-15%
• Cooling systems: 5-8%
• Grid connection fees: 3-8%

This is why the standard real estate heuristic—"location, location, location"—doesn't apply to data centers the way it does to retail or housing. The developer isn't optimizing for land cost. They're optimizing for risk, timeline, and grid access.

Farmland at $10K/acre is expensive compared to brownfield at $5K/acre. But if the brownfield site adds two years to the project timeline or introduces $50 million in remediation costs or creates contamination liability that jeopardizes financing, the math changes entirely.

Brownfield Barriers

Former industrial sites face three compounding challenges that make them unattractive for data center development despite apparent advantages.

1. Contamination and Liability

Any industrial site operated before modern environmental regulations likely has soil and groundwater contamination—petroleum products, heavy metals, solvents, PCBs. Under CERCLA (the Comprehensive Environmental Response, Compensation, and Liability Act), the current property owner can be held liable for cleanup costs regardless of who caused the contamination.

Before purchasing any brownfield site, a developer must conduct Phase I and Phase II Environmental Site Assessments. Phase I reviews historical records and visible conditions ($5K-10K). If Phase I identifies potential contamination, Phase II requires soil and groundwater sampling ($30K-100K+).

But here's the problem: you often don't discover the full extent of contamination until you start excavating for construction. A Phase II assessment might identify contamination in sampled locations, but miss underground storage tanks, buried waste, or plume migration. When construction crews hit contaminated soil, work stops. Remediation begins. Timelines extend. Costs escalate.

More fundamentally, CERCLA liability is strict, joint, and several—meaning even if you discover contamination you didn't cause and couldn't have found during due diligence, you're potentially liable for the entire cleanup. For a $7 billion project seeking financing from pension funds and institutional investors, this legal exposure is often a deal-killer.

2. Timeline Uncertainty

Remediation timelines are notoriously unpredictable. A "simple" cleanup might take 18-24 months. Complex contamination can extend 3-5 years. And you typically can't begin construction until remediation is complete and state environmental agencies sign off.

Compare this to greenfield development:

• Greenfield - Phase I assessment (30 days) → Clear title → Begin construction
• Brownfield - Phase I (60 days) → Phase II (90 days) → Remediation plan approval (6 months) → Remediation work (18-36 months) → Verification (6 months) → Begin construction

For data center operators racing to deploy AI capacity, a 2-4 year timeline penalty is unacceptable. If you're Meta or Microsoft promising cloud customers GPU availability in 2026, a site that delays occupancy until 2028 has zero value.

3. Infrastructure Mismatch

Former industrial sites were built for a different era's power requirements. A 1960s auto plant might have had 20-50 megawatts of electrical infrastructure. A modern gigawatt data center needs 1,000 MW—twenty to fifty times as much capacity.

The existing transmission connections are typically obsolete for data center purposes. The buildings, designed for assembly lines and manufacturing equipment, have floor loading, ceiling heights, and HVAC systems incompatible with racks of servers. The road access was built for trucks, not the continuous stream of equipment deliveries data centers require.

You end up demolishing the existing buildings anyway, remediating the site, and building entirely new infrastructure. You've gained nothing from the "developed" nature of the brownfield except contamination liability and timeline risk.

Greenfield Advantages

Agricultural land offers exactly what data center developers need: predictability, clean title, and design flexibility.

Clean Title and Environmental Baseline - Farmland has a documented history of agricultural use. Phase I assessments are straightforward. There's no contamination from industrial processes, no buried tanks, no legacy liability. Environmental review can be completed in weeks.

Predictable Timeline - Once you secure land, permits, and grid approval, construction can begin immediately. The entire development process from site acquisition to energization can be completed in 2-3 years—half or less than brownfield timelines including remediation.

Design Flexibility - Starting with a blank slate means buildings can be positioned optimally for power distribution, cooling systems can be designed for the specific topology, and security perimeters can be established without working around existing structures. This isn't cosmetic—it affects operational efficiency and long-term costs.

Clear Security Perimeters - Data centers require controlled access and physical security. An isolated greenfield site allows creation of well-defined perimeters without adjacent buildings, easements, or shared access roads. Brownfield sites often have complicated property boundaries and neighboring uses that create security challenges.

The Grid Connection Imperative

But the single most important factor in site selection isn't land cost or contamination risk. It's proximity to transmission infrastructure.

High-voltage transmission lines follow specific corridors determined by geography, historical load centers, and grid topology. Substations—the critical nodes where voltage is transformed and loads connect—are even more geographically constrained.

A gigawatt facility might need to be within 2-5 miles of a 345 kV substation or transmission corridor. Every additional mile requires more interconnection infrastructure, higher costs, and longer timelines. Beyond a certain distance, connection becomes economically infeasible.

This creates a geographic sorting pattern: the best data center sites are near transmission corridors with available capacity. In much of the country, those corridors run through rural areas—which means through farmland.

Urban brownfield sites are typically far from transmission infrastructure capable of handling gigawatt loads. Cities are served by distribution networks designed to spread load across millions of small consumers. Transmission corridors bypass cities, connecting large power plants to regional load centers.

So the site selection calculus becomes: farmland along a transmission corridor versus urban brownfield requiring 10+ miles of new transmission construction. The farmland wins on timeline, cost, and risk—even before considering contamination.

Policy Attempts and Failures

Recognizing the benefits of brownfield redevelopment, many states have tried to incentivize data centers to choose former industrial sites over agricultural land. These attempts have largely failed.

Michigan, competing for the Saline Township project, offered an additional 11 months of sales tax exemption for brownfield sites versus greenfield. The project chose farmland anyway.

The calculation for developers is straightforward: 11 months of sales tax exemption might be worth $30-50 million on a $7 billion project. But if brownfield adds 2 years to the timeline, the delay cost in lost revenue and competitive position exceeds $500 million. If remediation introduces a 10% chance of $200M in surprise costs, the expected value is negative even without considering timeline risk.

For policy to actually shift behavior, the incentive differential would need to be massive—not 11 months of tax exemption, but perhaps full property tax abatement for 20 years, or state-funded remediation, or liability caps that eliminate CERCLA exposure. Even then, the timeline advantage of greenfield might dominate.

This creates a fundamental tension: the market systematically produces outcomes that policymakers don't want (agricultural land conversion) but economic logic strongly favors. And because land cost is such a small fraction of total project cost, financial incentives alone cannot overcome the structural disadvantages of brownfield sites.