The Definitive 2026 Data Center Development & Investment Guide

Executive Summary

In the span of about three years, data centers went from a quiet, specialized corner of commercial real estate to the single most sought-after asset class in the country. The reason is simple: artificial intelligence turned computing into an industrial activity, and industrial activity needs an enormous, physical home — land, steel, water, fiber, and above all electricity. The companies racing to build that home are some of the most creditworthy tenants in the world, and they are signing leases measured not in square feet but in megawatts.

Why data centers became the fastest-growing real estate asset class

Traditional real estate is valued on location, rent, and cap rates. Data center real estate broke that model. A hyperscale operator does not care whether a parcel has a pretty view or sits on a popular retail corridor. It cares whether the site can deliver tens or hundreds of megawatts of reliable power, on a defined timeline, with fiber connectivity and water for cooling. When those boxes are checked, the land commands prices that would be unthinkable for the same dirt sold as industrial, agricultural, or residential ground. A 500-acre former coal plant that might trade as scrap-and-salvage industrial land can, with the right power infrastructure, be worth many multiples more as a data center campus. That repricing — driven by power, not place — is what makes the asset class explosive.

AI's impact on power demand

The defining shift is electrical. A conventional cloud or enterprise data center hall might run a rack at 5–10 kW. A rack packed with AI accelerators for training large language models can draw 40–130 kW or more, and next-generation systems push higher still. Multiply that across thousands of racks and a single AI campus can demand 100 MW to over 1 GW — the power footprint of a mid-sized city. For the first time in decades, electricity demand growth in the United States is being driven materially by a single new category of customer. Utilities that spent twenty years planning around flat load are now fielding interconnection requests that dwarf anything in their history.

Why power matters more than land

This is the sentence that reorganizes everything: "The first question isn't how many acres. It's how many megawatts." Land is abundant and relatively cheap. Deliverable, reliable, near-term power is scarce, expensive, and slow. You can entitle and grade a site in months; bringing a new high-voltage interconnection and substation online can take years and tens of millions of dollars. The result is that the value, the risk, and the competitive moat in data center development all migrate toward power. Whoever controls — or can credibly deliver — megawatts controls the deal.

Why brownfields are becoming strategic assets

Because power is the bottleneck, the most valuable parcels are often the ones that already sit on top of heavy electrical infrastructure. That describes the old industrial economy almost perfectly: retired coal-fired power plants, shuttered steel mills, idle paper mills, and large manufacturing complexes. They come with existing substations, high-voltage transmission, industrial zoning, water rights, rail, and large contiguous acreage. A retired coal plant may have a 230 kV or 500 kV switchyard and an interconnection that a greenfield developer would wait half a decade to obtain. The infrastructure of the 20th-century industrial economy is becoming the foundation of the 21st-century AI economy.

The opportunity — for every player at the table

The chapters that follow unpack each of these in depth, starting with the basics — what a data center actually is — and building toward the megawatt math, the brownfield thesis, the underwriting framework, and a 150+ question reference FAQ.

Chapter 1 — What Is a Data Center?

A data center is a purpose-built facility that houses computing hardware — servers, storage, and networking — together with the power, cooling, and connectivity needed to keep that hardware running continuously. At its simplest, it is a building optimized to convert electricity into computation and to remove the resulting heat. Everything distinctive about data center real estate flows from those two jobs: feeding power in, and getting heat out.

Physically, a data center is organized around the white space (the raised-floor or slab area where IT racks live) and the gray space (the supporting mechanical and electrical plant — switchgear, uninterruptible power supplies, generators, chillers, and cooling distribution). A facility is described less by its square footage than by its critical IT load in megawatts — the amount of power available to the computing equipment itself, separate from cooling and overhead.

Types of Data Centers

Not all data centers are the same animal. The category drives the tenant, the lease structure, the power density, and the kind of land and power a developer needs to assemble.

Hyperscale

Hyperscale facilities are the giants — campuses built by or for the largest cloud and AI operators. They are designed for standardization, rapid replication, and immense scale, often planned in phases that grow from tens of megawatts to several hundred or more. Hyperscale tenants are the most powerful force in the land market because their power appetite is effectively unbounded and their credit is exceptional.

Colocation

A colocation ("colo") provider builds the shell, power, and cooling, then leases capacity to others. Retail colocation serves many smaller tenants who rent racks or cages; wholesale colocation leases large, dedicated blocks of power (often whole data halls) to single large customers — frequently the hyperscalers themselves, who lease wholesale capacity to supplement what they build.

Enterprise

An enterprise data center is owned and run by one organization for its own workloads. Once the default model, enterprise has steadily migrated to cloud and colo, but banks, hospitals, government, and certain manufacturers still run their own for control, latency, or compliance reasons.

Edge

Edge data centers are small facilities placed close to end users — in or near metro areas, at the base of cell towers, or inside existing buildings — to reduce latency for applications like streaming, gaming, autonomous systems, and real-time AI inference. They trade scale for proximity.

AI Training Facilities

AI training facilities exist to train large models. They concentrate thousands of GPUs into tightly networked clusters, producing the highest power densities in the industry and enormous cooling demands. Because training is somewhat latency-tolerant, these facilities can be sited wherever large blocks of cheap, reliable power exist — which is exactly why power-rich regions and brownfields matter.

AI Inference Facilities

Inference facilities run already-trained models to answer user requests. Inference is more latency-sensitive than training, so these tend to sit closer to population centers and often blend into colocation and edge footprints. As AI usage scales, inference is expected to become the larger long-run share of compute.

The major entities you'll encounter

Behind these names sit the colocation developers (Equinix, Digital Realty, QTS, CyrusOne, Vantage, Switch and others) who build much of the physical capacity, plus the chipmakers — above all NVIDIA — whose hardware sets the power density that every building must be engineered around.

Chapter 2 — Why Data Centers Are Exploding

Data center demand has compounded for two decades on the back of the internet, mobile, and cloud computing. What changed recently is the arrival of a second growth engine — artificial intelligence — layered on top of an already-growing base. The combination has produced demand that is straining the power grid itself.

AI Demand

AI is the accelerant. Training and running large models is extraordinarily compute-intensive, and that compute translates directly into power and physical space.

ChatGPT and the consumer AI moment

The public launch of ChatGPT turned AI from a research curiosity into a mass-market product overnight, triggering a global race to build the infrastructure behind it. Every major technology company responded by committing tens of billions of dollars to compute capacity, and that capital lands as concrete, steel, and megawatts.

LLMs (large language models)

Large language models are trained on vast datasets using enormous clusters of accelerators running for weeks or months. Each new generation of model tends to demand more compute than the last, and training is only half the story — once trained, models must be served continuously.

GPU clusters

Modern AI runs on GPU clusters (and related accelerators) wired together with high-speed networking so thousands of chips behave like one machine. These clusters are the densest, hottest, most power-hungry hardware ever deployed at scale, and they are the reason rack densities jumped from single-digit kilowatts to many tens of kilowatts.

Inferencing

Inferencing — actually using a trained model to generate answers — is where AI meets the user. As adoption grows, aggregate inference demand can exceed training demand, and because it is latency-sensitive, it pulls capacity toward population centers and the edge.

Cloud Computing

Underneath the AI wave, ordinary cloud computing keeps growing. Businesses continue migrating workloads off their own servers and into AWS, Azure, Google Cloud, and Oracle. Every SaaS app, streaming service, and corporate system that moves to the cloud adds steady baseline demand for data center capacity independent of AI.

Digital Transformation

Across every industry, "digital transformation" — moving operations, records, and customer interactions online — generates data and the need to store, process, and protect it. The long migration from paper and on-premises systems to digital and cloud is nowhere near finished and adds durable, recurring demand.

Autonomous Vehicles

Self-driving systems generate and consume staggering amounts of data — training perception models, processing sensor feeds, running simulations, and updating fleets. As autonomy advances, both the training (centralized) and the real-time inference (edge) components push data center demand higher.

Robotics

Industrial and consumer robotics, increasingly powered by AI, depend on models trained in data centers and, often, on low-latency inference nearby. The same compute backbone behind language models underlies the coming wave of physical automation.

Defense Applications

National defense and intelligence increasingly rely on AI, simulation, and large-scale data processing — much of it in secure, often domestically sited facilities. The strategic importance of compute has made data centers a matter of national security as well as commerce.

Healthcare

Genomics, medical imaging, drug discovery, and AI-assisted diagnostics are all compute-heavy. Healthcare's digitization and its adoption of AI add another durable demand stream, often with strict data-residency and security requirements that favor specific facilities and locations.

Chapter 3 — The New Currency Is Power

If you remember one chapter, make it this one. The entire data center land market reorganizes around a single resource: electricity. Power is the currency. Everything else — acreage, zoning, fiber, even water — is negotiable or solvable. Power is the constraint that makes or breaks a site.

Why Land Doesn't Matter Without Power

A perfect 1,000-acre parcel — flat, dry, well-zoned, fiber overhead — is worth very little to a data center developer if it cannot get power for seven years. Conversely, a modest 80-acre site adjacent to a large substation with available capacity can be worth a fortune. This inverts the traditional real estate hierarchy, where location leads. In data centers, the binding question is delivery of megawatts, and that is why the industry repeats the same line:

Land is a commodity input; reliable near-term power is the scarce asset. A developer's real job is often less about real estate and more about securing an electrical interconnection — navigating the utility, the queue, the substation, and the transmission system. Whoever solves power wins.

Understanding Power Capacity

Data center scale is measured in megawatts (MW) and, increasingly, gigawatts (GW). Here is a rough sense of what different capacities mean.

Utility Constraints

The utility is the gatekeeper. Whether — and when — a site can be powered depends on the local utility's available capacity, its transmission system, its interconnection process, and its willingness and ability to build new infrastructure. A site in a constrained service territory may face multi-year waits or be told that large new load simply cannot be served near-term. Understanding the serving utility is step one of any site evaluation.

Substations

A substation steps high-voltage transmission power down to levels a data center can use, and is the physical point where a campus connects to the grid. Proximity to a substation with available capacity — or the ability to build a new one and tie into nearby transmission — is one of the most valuable attributes a site can have. Substation buildouts are expensive and slow, so existing capacity is gold.

Transmission Lines

High-voltage transmission lines (often 115 kV, 138 kV, 230 kV, 345 kV, or 500 kV) are the highways that move bulk power. A site near high-capacity transmission has a far shorter, cheaper path to large-scale power than one that would require miles of new line. The voltage and spare capacity of nearby lines materially affect how much load a site can ultimately support.

Interconnection Queues

To connect large new load (or new generation) to the grid, a project must enter the utility's or grid operator's interconnection queue and undergo studies that determine what upgrades are needed and who pays for them. These queues have become badly congested, and timelines of several years are common. A site's effective value is heavily influenced by its position in — or ability to bypass — the queue.

Behind-the-Meter Generation

Because the grid is slow, many developers are turning to behind-the-meter generation — building power on-site rather than waiting for the utility. This can dramatically compress timelines and is one of the defining trends of the current cycle.

Natural Gas

On-site natural gas generation (turbines or reciprocating engines) can bring large blocks of power online relatively quickly where gas pipelines exist. It is a leading bridge solution for AI campuses that cannot wait for grid interconnection, though it carries emissions and permitting considerations.

Nuclear

Nuclear power — including restarting retired reactors, co-locating at existing nuclear plants, and future small modular reactors (SMRs) — has become a serious strategy for delivering large, firm, carbon-free power to data centers. Several major operators have signed nuclear-related deals, signaling that the largest players will pay for firm power and are willing to underwrite new generation.

The big power players you'll encounter

Chapter 4 — How Developers Evaluate Data Center Land

Once power is established as the lead criterion, developers run every prospective site through a multi-factor screen. A parcel that passes the power test still has to clear a long list of physical, regulatory, and economic checks. Here is the working framework, factor by factor.

Site Selection

Site selection is a process of elimination. Developers and their engineers score candidate parcels against a weighted set of criteria, weeding out anything with a fatal flaw — no near-term power, in a floodway, unbuildable topography, hostile zoning — before spending real money on diligence. The factors below are the core of that screen.

Acreage

Data center campuses need room — for the buildings themselves, for generators and cooling plant, for substations and switchyards, for security setbacks, and for future phases. A single large building might sit comfortably on 20–50 acres, while a multi-phase hyperscale campus can absorb 200–1,000+ acres. Contiguity matters: large, unbroken tracts are far more valuable than the same acreage split by roads, easements, or parcels.

Topography

Flat, gently graded land is ideal. Steep slopes, rock, and irregular terrain raise earthwork and foundation costs and can sterilize portions of a site. Developers favor parcels that minimize cut-and-fill and allow efficient building pads and equipment yards.

Flood Zones

Data centers are mission-critical and cannot tolerate flooding. Sites in FEMA floodways or high-risk flood zones are heavily disfavored; even partial flood exposure complicates insurance, financing, and tenant approval. Clean, well-drained ground above flood risk is a baseline requirement.

Environmental Issues

Contamination, wetlands, endangered-species habitat, and historical/archaeological constraints can all delay or block development. On brownfields this cuts both ways — the same industrial history that delivers power infrastructure can also bring remediation obligations. Phase I and, where warranted, Phase II environmental assessments are standard.

Fiber Access

A data center is useless without connectivity. Proximity to long-haul fiber routes and multiple carriers reduces the cost and risk of bringing in the bandwidth a facility needs. Sites along established fiber corridors — often following highways, rail, or pipeline rights-of-way — have a real advantage.

Water Availability

Many cooling designs consume significant water. Access to municipal water, wells, or surface water — and the legal right to use it — can be decisive, especially for large campuses in dry regions. Even air-cooled or closed-loop designs benefit from water access as a backstop. Water is covered in depth in Chapter 7.

Utility Access

Beyond electricity, sites need the full utility picture: gas (increasingly, for on-site generation), water and sewer, and telecom. The presence, capacity, and cost of extending these services factor into both feasibility and budget.

Labor Availability

Construction requires a large skilled workforce — electricians, mechanical trades, and specialized contractors — and operations require technicians. While data centers employ relatively few people once running, the construction phase is labor-intensive, and regions with deep trade labor pools build faster and cheaper.

Tax Incentives

States and localities compete hard for data centers with sales-tax exemptions on equipment, property-tax abatements, and other incentives. These can swing project economics meaningfully, and the strength of a jurisdiction's incentive program is a genuine site-selection factor. Incentives also signal a community's willingness to host the use.

Zoning

The right zoning — typically industrial or a purpose-built data center designation — and a cooperative local government dramatically de-risk a project. Sites that are already appropriately zoned, or in jurisdictions with clear approval pathways, are worth more than those requiring rezoning, variances, or contentious public hearings. Entitlement certainty has real value.

Chapter 5 — Brownfields Become Gold Mines

This is the most counterintuitive — and most lucrative — idea in the entire guide. The retired, rusting industrial sites that communities often write off as liabilities are frequently the single best data center sites in their region. Why? Because they already sit on top of the one thing everyone else is fighting for: power.

What Is a Brownfield?

A brownfield is a previously developed industrial or commercial site that is idle, underused, or abandoned, often with some real or perceived environmental contamination. Classic examples include retired power plants, closed factories, old steel and paper mills, rail yards, and chemical or manufacturing complexes. They contrast with greenfields — undeveloped land with no prior industrial use.

Why Data Centers Love Brownfields

For most uses, a brownfield's industrial baggage is a drawback. For data centers, the very features that made a site industrial are exactly what a hyperscale campus needs. The match is almost uncanny.

Coal Plant Redevelopment

Retired coal-fired power plants deserve their own spotlight. When a coal plant shuts down, its generation goes away — but its grid interconnection does not. The high-voltage switchyard, the transmission ties, the cooling-water access, and often rail and large acreage remain. That interconnection was sized to export hundreds of megawatts to the grid; a data center can run that flow in reverse, importing large load through infrastructure that already exists. This is why coal-plant sites have become some of the most fought-over parcels in the country.

The old industrial economy becomes the infrastructure for AI

There is a poetic symmetry here. The mills and plants that powered the 20th-century industrial economy are being reborn as the foundation of the 21st-century AI economy. The transmission lines built to move power from a coal plant now move power to a compute campus. The rail spur that brought in coal can bring in transformers and gear. The water that cooled turbines now cools servers. Examples of brownfields finding new life as data centers include:

• Retired coal plants — high-voltage switchyards, water, rail, large land.
• Steel mills — massive electrical service, industrial zoning, heavy foundations, rail.
• Manufacturing facilities — large power, utilities, and contiguous acreage near workforce.
• Paper mills — abundant water rights, significant power, and large rural tracts.

Case Study Framework — Former Coal Plant

Consider a hypothetical that mirrors deals happening across the country. (Illustrative figures, not a specific transaction.)

Why this may be worth more as a data center than as industrial land

Sold as generic industrial land, a 500-acre former coal plant might trade at industrial-acre pricing, discounted for demolition and remediation. But to a hyperscale developer, the interconnection and substation alone can be worth more than the land — because the alternative is a five-to-seven-year wait and a nine-figure infrastructure spend on a greenfield. When power is the binding constraint, a site that delivers power on day one commands a premium that has nothing to do with traditional land comps. That gap — between industrial-land value and powered-data-center value — is the brownfield opportunity in one sentence.

How to Determine the Highest & Best Use of an Obsolete Industrial Site

When a factory, mill, or plant goes dark, the owner faces a deceptively simple question: what is this property actually worth now? The answer depends on its highest and best use — the legally permissible, physically possible, financially feasible, and maximally productive use of the site. For a growing number of obsolete industrial properties, that use is no longer manufacturing, warehousing, or scrap-and-salvage land. It may be a data center, an industrial outdoor storage (IOS) yard, an enterprise IOS (EIOS) campus, or a logistics facility — and the gap between the old use and the best use is where value is unlocked.

The four tests of highest and best use

Appraisers and developers screen every candidate use against four filters. A use only qualifies as "highest and best" if it passes all four.

1. Legally permissible

What does the zoning, entitlement, and deed allow — and what could realistically be rezoned or variance-approved? Industrially zoned brownfields already clear this hurdle for data centers, IOS, and heavy logistics, which is a major head start.

2. Physically possible

Does the site's acreage, topography, soils, flood exposure, and existing infrastructure support the use? A 500-acre flat parcel with a substation supports a data center; a 15-acre paved yard near an interchange supports IOS.

3. Financially feasible

Does the use generate enough value to justify the cost of conversion, including demolition and remediation? Power infrastructure and paving that already exist tilt feasibility dramatically.

4. Maximally productive

Of the uses that pass the first three tests, which produces the greatest value? This is where a powered brownfield's data center value can eclipse its industrial value many times over.

Common highest-and-best-use outcomes for obsolete industrial sites

The discipline is to test the site against each use rather than defaulting to its prior life. An obsolete factory is not "a factory that no longer works" — it is a bundle of power, acreage, access, and entitlements that some buyer values more than you might expect. Determining which buyer, and which use, is the whole game.

10 Signs Your Industrial Property Is Worth More Than Its Current Use

Many owners of aging industrial real estate are sitting on far more value than their current rent roll or tax assessment suggests — because the market is now pricing power, access, and acreage rather than the building. If your property shows several of the signs below, it may be worth ordering a fresh highest-and-best-use analysis before you renew a low lease, accept a scrap-land offer, or let it sit idle.

1. It sits near high-voltage transmission or a substation. Proximity to deliverable power is the single most valuable attribute a site can have in the data center era. Lines at 115 kV and above, or an on-site or adjacent substation, are a major flag.
2. It has large existing electrical service. Heavy manufacturing, mills, and plants drew enormous power. That existing service — and the interconnection behind it — can be worth more than the land itself.
3. It's a retired power plant or heavy-industrial site. Retired coal plants, steel mills, and paper mills often retain switchyards, transmission ties, water rights, and rail — exactly what data center developers pay premiums for.
4. It offers large, contiguous acreage. Unbroken tracts of 50, 200, or 500+ acres in single ownership are increasingly scarce and are prized for multi-phase campuses.
5. It has water rights or a permitted intake. Cooling water — municipal, well, surface, or reclaimed — is a genuine entitlement that raises value for compute and heavy industrial reuse.
6. It's on or near major fiber routes. Long-haul fiber and multiple carriers along nearby highway, rail, or pipeline corridors make a site far more attractive for connectivity-dependent uses.
7. It has direct highway, interchange, or rail access. Strong access supports data centers (heavy equipment delivery) and is the core driver of industrial outdoor storage (IOS) and enterprise IOS (EIOS) value.
8. It's already zoned industrial. Existing heavy-industrial or data-center zoning sidesteps the rezoning fights and public opposition that kill or delay greenfield projects — entitlement certainty has real, bankable value.
9. It sits in an Opportunity Zone or an incentive-rich jurisdiction. Federal Opportunity Zone treatment plus state and local tax incentives can stack on top of the site's physical advantages.
10. It's paved, fenced, and stabilized. A paved, secured yard with truck circulation is turn-key for IOS/EIOS demand — one of the fastest-growing industrial niches — without any vertical construction.

None of these signs is a guarantee, and each must be verified rather than assumed. But if your property checks three or more boxes, the odds are high that its highest and best use — and its market value — is materially above its current use. That gap is the opportunity.

How to Sell an Industrial Campus to Institutional Buyers

Selling a large industrial campus — a former plant, mill, or multi-building complex — to an institutional buyer (a hyperscaler, a data center developer, a REIT, or an infrastructure fund) is a fundamentally different exercise from selling to a local user. Institutional buyers underwrite risk, not charm. They pay premiums for certainty and discount heavily for ambiguity. The seller's job is to remove ambiguity before going to market.

1. Establish the highest and best use first

Before pricing or marketing, determine whether the campus is worth more as a data center site, an IOS/EIOS yard, logistics, or continued industrial use. The buyer universe, the value, and the entire sale strategy flow from that answer. Marketing a powered brownfield as generic industrial land can leave enormous value on the table.

2. Assemble the power and infrastructure story

For a data center buyer, power is the deal. Document the serving utility, available capacity, transmission voltages and proximity, substation condition, interconnection rights, and any behind-the-meter or gas options. A credible, evidenced power narrative is the difference between industrial-land pricing and powered-land pricing.

3. De-risk the diligence items up front

Institutional buyers will find every problem eventually; sellers who surface them early control the narrative. Commission a Phase I (and Phase II where warranted), an ALTA survey, title work, zoning confirmation, and environmental status before listing. Clean, organized diligence compresses timelines and defends price.

4. Confirm entitlements and community posture

Existing industrial zoning, a cooperative local government, and available incentives materially raise value and lower a buyer's execution risk. Where possible, secure or document the approval pathway so the buyer isn't pricing in a rezoning fight.

5. Build a professional data room

Package everything — surveys, environmental reports, utility correspondence, tax and incentive detail, title, and infrastructure records — into an organized data room. Institutional diligence teams move faster and bid higher when information is complete and credible.

6. Reach the right buyer universe

The buyers for a powered campus are a small, specialized group: hyperscalers' real estate teams, data center developers, infrastructure funds, and the brokers who serve them. Reaching them — rather than the local industrial market — is what produces competitive, institutional-grade offers.

7. Structure the deal to match buyer needs

Institutional buyers often prefer options, phased takedowns, or long due-diligence periods that let them confirm power before closing. Sellers who understand and accommodate these structures — while protecting themselves with deposits and milestones — attract more and better bids.

Chapter 6 — Data Centers and Opportunity Zones

Few people connect these two topics, which is exactly why it is an opportunity. Opportunity Zones are designated economically distressed areas where investors can receive significant federal capital-gains tax benefits for long-term investment. Many of them happen to sit in exactly the kind of rural and post-industrial places where powered land and brownfields are found.

Opportunity Zone benefits

The Opportunity Zone program lets investors defer and potentially reduce tax on capital gains by rolling them into a Qualified Opportunity Fund that invests in property or businesses within a designated zone. The headline benefit is that appreciation on a qualifying investment held for the long term (generally ten years) can be excluded from federal capital-gains tax. The exact rules, dates, and percentages have evolved and must be confirmed with a tax advisor, but the structural incentive — favorable treatment for long-term investment in distressed areas — is the point.

Data center development in an Opportunity Zone

Data centers are capital-intensive, long-hold assets — precisely the profile the Opportunity Zone program rewards. A developer building a campus in a designated zone may pair an extraordinary operating asset (long-term, investment-grade-backed cash flow) with a powerful tax structure (deferral and potential exclusion of gains). When a brownfield with existing power also sits inside an Opportunity Zone, the stack of advantages — power, zoning, incentives, and tax treatment — can be remarkable.

Long-term capital gains

Because the largest Opportunity Zone benefit accrues to investments held for roughly a decade, the program aligns naturally with the long duration of data center leases and ownership. Investors sitting on large unrealized capital gains — from stock, a business sale, or other real estate — have a potential pathway to redeploy those gains into compute infrastructure on tax-advantaged terms.

Rural development

Many Opportunity Zones are rural, and so are many of the best power-rich data center sites — near generation, transmission, and water, away from congested metros. This overlap means data center investment can deliver on the program's original purpose (capital and jobs into distressed communities) while solving the industry's core problem (finding powerable land). It is one of the few topics where the tax policy, the community benefit, and the asset economics genuinely point the same direction.

Chapter 7 — Water Requirements

After power, water is the resource that most often makes headlines and shapes community reaction. Data centers generate enormous heat, and removing that heat efficiently has historically involved water. How much water a facility uses depends entirely on its cooling design.

Air-Cooled Facilities

Air-cooled designs reject heat to the atmosphere using fans and chillers, consuming little or no water for cooling. They trade water for electricity — air cooling is generally less energy-efficient than water-based cooling, so a dry design lowers water use but can raise power use. In water-scarce regions, this trade is increasingly worth it.

Liquid Cooling

Liquid cooling brings coolant directly to the chips (via cold plates) rather than cooling the room air. It is far more effective at handling the extreme heat of AI hardware and is becoming standard for high-density GPU clusters. Depending on the system, it can use water more efficiently and enable much higher rack densities than air cooling allows.

Immersion Cooling

Immersion cooling submerges servers in a non-conductive fluid that carries heat away directly. It enables the highest densities and excellent efficiency, and closed-loop immersion can dramatically reduce water consumption. It is newer and adds operational complexity, but it is a leading direction for AI-era facilities.

Water Usage Concerns

Traditional evaporative cooling (cooling towers) can consume large volumes of water — a point of real public concern, especially in arid regions or during drought. Water Usage Effectiveness (WUE) has joined Power Usage Effectiveness (PUE) as a metric operators track and disclose. The industry is moving toward designs — air-cooled, closed-loop liquid, immersion, and reclaimed/non-potable water — that cut potable water draw.

Municipal Challenges

For a town or utility, a large data center's water demand can rival that of a sizeable population. This raises legitimate questions about capacity, drought resilience, and fairness to other users, and it has become a flashpoint in some communities. Developers increasingly address it head-on by using non-potable or reclaimed water, closed-loop systems, or dry cooling, and by being transparent about consumption. A site's water rights and the local utility's capacity are genuine site-selection factors — and a place where doing right by the community and doing right by the project align.

Chapter 8 — Fiber Is the New Railroad

In the 19th century, towns lived or died by the railroad. In the data center era, the equivalent is fiber-optic connectivity. Power lets a data center run; fiber lets it matter. A facility with abundant power but poor connectivity is a generator looking for a purpose.

Dark Fiber

Dark fiber is fiber-optic cable that has been laid but is not yet "lit" with active equipment — spare capacity in the ground. Access to dark fiber lets an operator light its own dedicated, high-capacity connections without waiting for a carrier to build new routes. Sites near abundant dark fiber can scale bandwidth quickly and cheaply.

Long-Haul Networks

Long-haul fiber networks carry traffic between cities and regions over long distances. Proximity to multiple long-haul routes gives a data center diverse, high-capacity paths to the rest of the internet. These routes often parallel highways, rail lines, and pipeline corridors — which is why sites along established infrastructure corridors tend to have good connectivity.

Carrier Hotels

A carrier hotel is a major interconnection facility where many networks meet and exchange traffic. Being on a route to one or more carrier hotels — or being near a major interconnection point — improves a site's connectivity options and reduces the cost of reaching many networks at once.

Redundancy

Mission-critical facilities need redundant, physically diverse fiber paths so that a single cut — a backhoe, a storm — cannot sever the site. Multiple carriers entering from different directions, along separate rights-of-way, is the gold standard. Connectivity redundancy sits alongside power redundancy as a core reliability requirement.

Latency

Latency — the delay for data to travel to and from the facility — is critical for some workloads and nearly irrelevant for others. AI training is largely latency-tolerant, so training campuses can sit far from population centers in power-rich areas. Inference, real-time applications, financial trading, and interactive services are latency-sensitive and pull capacity closer to users. Matching a site's connectivity and location profile to the intended workload is part of good underwriting.

Chapter 9 — Understanding Megawatts (Definitions)

Answer engines and search both reward clear, self-contained definitions. This chapter is a glossary of the power and reliability concepts that govern data centers, each written as a direct answer to a direct question.

Chapter 10 — The Best Markets for Data Centers

Data center activity concentrates in markets that combine power availability, connectivity, land, incentives, and proximity to demand. Here are several of the most important US markets and what defines each.

Ashburn, Virginia

Ashburn and the surrounding Loudoun County corridor — "Data Center Alley" — is the largest and densest data center market in the world. Decades of fiber infrastructure, a deep ecosystem of operators, and proximity to a major internet interconnection point made it the default. Its challenge today is its own success: power and land have grown scarce and expensive, and grid strain has pushed new development outward — which is precisely what is opening up emerging markets.

Dallas, Texas

Dallas–Fort Worth offers a large, growing market with abundant land, an independent and flexible grid (ERCOT), business-friendly policy, and strong connectivity. Texas's energy resources and pro-development stance have made it a magnet for new capacity, including power-intensive AI projects.

Phoenix, Arizona

Phoenix has become a top-tier market thanks to large available land, relatively low natural-disaster risk, strong incentives, and aggressive utility support — balanced against water scarcity, which makes cooling design and water strategy especially important there.

Atlanta, Georgia

Atlanta has surged as a major Southeast hub, with competitive power, strong fiber, attractive incentives, and central regional geography. It exemplifies the shift of growth toward Southern markets with available power.

Nashville, Tennessee

Nashville anchors a fast-growing Tennessee market, benefiting from TVA power, central US geography, strong fiber routes, business-friendly policy, and a growing tech presence. It is part of the regional story explored in depth in Chapter 11.

Memphis, Tennessee

Memphis combines TVA power, abundant water from the Memphis Sand aquifer, a central logistics location, available industrial land, and major fiber and rail infrastructure. It has drawn significant attention — including very large AI projects — as a rising power-rich market.

The throughline: as established markets like Ashburn hit power and land limits, demand spills into power-rich, land-rich, incentive-friendly regions — and the Southeast and the TVA footprint are major beneficiaries.

Chapter 11 — Why Tennessee Is Becoming a Data Center Market

Tennessee sits at the convergence of nearly every factor that matters: large, competitively priced power; abundant water; central geography; strong fiber; and a business-friendly climate. For anyone working in powered land and brownfields, it is one of the most compelling regions in the country.

TVA (Tennessee Valley Authority)

The Tennessee Valley Authority is the federal power utility serving Tennessee and parts of Alabama, Mississippi, Kentucky, Georgia, North Carolina, and Virginia. TVA operates a large, diversified generation fleet (nuclear, hydro, gas, and more), historically offers competitive rates, and has actively courted large industrial and data center load. A single utility covering a broad, power-rich region simplifies the development conversation and is a major reason the area attracts compute.

Power availability

TVA's generation base and transmission network give the region meaningful capacity to serve large new load — the scarcest resource in the industry. Combined with retired and retiring coal plants in the footprint (with their existing interconnections), Tennessee offers both grid power and brownfield power opportunities.

Water availability

The region is comparatively water-rich — major rivers, reservoirs, and in West Tennessee the Memphis Sand aquifer. For an industry where water can be a constraint elsewhere (notably the desert Southwest), reliable water is a genuine competitive advantage for cooling.

Central geography

Tennessee's central US location offers low latency to a large share of the population and proximity to multiple major markets — useful for both content distribution and latency-sensitive inference. It also sits on major logistics and transportation corridors.

Fiber routes

Major long-haul fiber routes cross the state, following its interstate and rail corridors and connecting Southeast and Midwest markets. Good connectivity plus good power is the combination data centers seek.

The four metros

Chapter 12 — How Investors Underwrite Data Center Land

Underwriting data center land is fundamentally an exercise in risk decomposition. The headline numbers — price per acre, price per megawatt — are downstream of a series of risks that determine whether a site can actually become a revenue-producing campus. Sophisticated investors price each risk explicitly.

Power Risk

The dominant risk. Can the site obtain the required megawatts, at an acceptable cost, on the needed timeline? This means examining the serving utility's available capacity, transmission proximity, substation access, interconnection queue position, and the credibility of any power commitment. A site without a clear, credible path to power is not a data center site at any price — it is a speculation on getting power. Underwriting assigns this the heaviest weight.

Entitlement Risk

Will the project get the zoning, permits, and approvals it needs, and on what timeline? Sites already zoned industrial or for data center use, in cooperative jurisdictions, carry low entitlement risk. Sites requiring rezoning, variances, or contentious hearings carry more. Entitlement certainty has direct value because it removes a slow, binary risk.

Utility Risk

Beyond raw power availability, this captures the reliability, rate structure, and behavior of the utility — rate increases, large-load tariffs, the terms of interconnection agreements, who pays for upgrades, and the utility's track record of delivering on commitments. Utility risk also includes water and gas service for cooling and on-site generation.

Construction Risk

Can the campus be built on budget and on schedule? This covers site conditions (topography, soils, environmental remediation), supply chain for critical gear (transformers and switchgear have had long lead times), labor availability, and contractor capacity. Brownfields add demolition and remediation to the construction-risk column.

Tenant Risk

Who will lease or buy the capacity, and how creditworthy and committed are they? Hyperscale tenants are exceptionally creditworthy, but their site requirements are demanding and their decisions can move. Underwriting considers the depth of demand in the market, the likelihood of securing a tenant, lease term and structure, and the credit behind the cash flow.

Exit Risk

How and to whom does the investment ultimately monetize — a sale of powered land to a developer, a stabilized leased asset sold to an institutional buyer, or a long-term hold? Exit risk weighs liquidity, the buyer universe, cap-rate sensitivity, and how durable the asset's advantages (power, location, tenant) will be at sale. The more transferable the site's core advantages, the lower the exit risk.

Chapter 13 — Data Center Economics

Data center economics revolve around a handful of metrics that translate physical capacity into financial returns. Understanding them lets you compare deals, gauge development profit, and see why the asset class attracts so much capital. (Figures are general industry ranges and vary widely by market, design, and time.)

Cost per MW

Cost per megawatt is the headline development metric — the all-in cost to build a megawatt of critical IT capacity, including building, power, and cooling infrastructure. Industry figures commonly fall in the range of roughly $7M–$15M per MW for shell-and-core through fully fitted capacity, varying with tier, density, cooling type, and location. Because capacity is leased by the megawatt, cost per MW versus rent per MW drives the return.

Cost per rack

Cost per rack expresses cost at the rack level. As rack densities climb (from ~10 kW toward 100 kW+), the cost and complexity per rack rise — especially for the power distribution and liquid/immersion cooling that high density requires. Density changes the math: a high-density hall fits more compute (and more revenue) into the same building footprint, but at higher per-rack infrastructure cost.

Cost per square foot

Cost per square foot is the traditional real estate metric, but it is secondary in data centers because value tracks power, not floor area. A small, ultra-dense AI hall can be worth far more than a large, low-density one. Cost per square foot is still used for the building shell, but megawatts are the true unit of account.

Yield on cost

Yield on cost (development yield) is stabilized net operating income divided by total development cost. It measures the return the developer creates by building. Comparing yield on cost to the cap rate at which the stabilized asset can be sold reveals the value created — the heart of the development case.

Development spreads

The development spread is the gap between yield on cost and the market exit cap rate. If a developer builds to, say, an 8% yield on cost and the stabilized asset trades at a 6% cap rate, that 200-basis-point spread is the profit margin of development, capitalized into value. Wide spreads — driven by scarce powered land and strong tenant demand — are exactly what is attracting developers and capital to the sector.

Illustrative share of total data center development cost by category. Power and cooling together often rival or exceed the building itself — a reminder that a data center is an electrical asset wearing a building.

Chapter 14 — The AI Infrastructure Buildout

The current cycle is best understood as the construction of an entirely new layer of industrial infrastructure — "AI factories" — at a pace and scale rarely seen. The numbers being committed by the largest players dwarf prior data center cycles.

AI factories

The term AI factory captures the idea that these facilities are industrial plants that manufacture intelligence — taking in electricity and data and producing trained models and inference at scale. Unlike traditional data centers optimized for many small workloads, AI factories are purpose-built around dense accelerator clusters and the extreme power and cooling they demand. They represent a new building typology, not just a bigger data center.

GPU clusters

At the heart of every AI factory are GPU clusters — thousands of accelerators networked to train and serve models. The size of these clusters keeps growing, and each generation tends to demand more power per chip and per rack. The cluster's power and networking requirements dictate the building's electrical and cooling design, which is why hardware roadmaps drive real estate decisions.

Inference facilities

Inference facilities serve trained models to users. As AI moves from training breakthroughs to mass deployment, inference demand scales with usage and is expected to become the larger long-run share of compute. Because inference is more latency-sensitive, it distributes capacity toward population centers and the edge, complementing the remote, power-rich training campuses.

Future demand

Projections for AI-driven power and capacity demand are large and rising, and while specific forecasts differ and will change, the direction is consistent: continued, substantial growth in compute, power, and the physical infrastructure to house it. The binding constraint on that growth is increasingly power and grid capacity — which loops directly back to the central thesis of this guide.

The key players

Around these sit the hyperscalers (AWS, Microsoft, Google, Meta, Oracle) who build or lease the capacity, the colocation developers who deliver much of it, and the utilities and power developers racing to feed it. The AI buildout is, at bottom, a power-and-real-estate story — which is why landowners, developers, and investors who understand powered land are positioned at the center of it.

Chapter 15 — Data Center FAQs (150+)

A comprehensive, answer-engine-optimized question bank covering land, power, brownfields, water, fiber, economics, markets, and process. Each answer is self-contained so it can stand alone in search and AI results.

Land & Site Basics

Power, Megawatts & the Grid

Reliability & Design

Brownfields & Coal Plant Conversion

Chapter 16 — Future Trends Through 2035

The next decade will be defined by the race to feed compute with power. The trends below are already underway and are likely to intensify as AI scales and the grid struggles to keep pace.

Nuclear-powered data centers

Expect nuclear to move from novelty to mainstream strategy for the largest operators. Co-location at existing nuclear plants, restarts of retired reactors, and long-term power agreements give hyperscalers large, firm, carbon-free power that the congested grid cannot quickly provide. The willingness of major players to underwrite nuclear signals how valuable firm power has become.

Small modular reactors (SMRs)

SMRs — smaller, factory-built reactors — are a leading bet for delivering firm power co-located with data centers. Commercial timelines remain uncertain and the technology is still maturing, but the level of interest and investment suggests SMRs could become a meaningful part of the power mix for compute in the latter half of the decade.

Natural gas generation

In the near term, on-site natural gas is the pragmatic bridge. Where pipelines exist, gas can bring large blocks of power online far faster than the grid, and behind-the-meter gas plants are already powering AI campuses. Gas will likely remain a workhorse bridge fuel even as nuclear and renewables scale, with carbon and permitting considerations shaping its role.

Edge AI

As inference grows and latency-sensitive AI moves into applications, edge AI — distributed compute close to users — will expand alongside the giant remote training campuses. Expect a two-tier geography: massive, power-rich training factories in rural and post-industrial areas, and a proliferating layer of smaller inference and edge sites near population centers.

Autonomous systems

Autonomous vehicles, drones, and robotics will drive demand on both ends — centralized training and simulation, and distributed low-latency inference. The compute backbone behind physical automation will be a growing, durable source of data center demand through 2035.

National security implications

Compute has become strategic. Expect continued attention to domestic capacity, supply-chain security for chips and equipment, the resilience and security of the grid serving compute, and the treatment of data centers and powered land as matters of national interest. This elevates the strategic value of US powered land and the infrastructure that serves it.

Bonus: Checklists & Site Selection Scorecard

Practical, printable diligence tools. Use them as a starting framework — every site, utility, and jurisdiction differs, so adapt and verify with qualified professionals.

Data Center Land Checklist (50 points)

A broad screen for evaluating a prospective data center parcel across power, site, connectivity, water, regulatory, and economic factors.

1. Serving electric utility identified
2. Available power capacity confirmed with utility
3. Distance to nearest high-voltage transmission
4. Voltage and spare capacity of nearby lines
5. Proximity to a substation with capacity
6. Interconnection queue position/timeline understood
7. Estimated cost to deliver target MW
8. Behind-the-meter generation feasibility
9. Natural gas pipeline access (for on-site gen)
10. Target deliverable MW vs. project need
11. Total acreage adequate for program
12. Land is contiguous and unbroken
13. Room for future phases/expansion
14. Topography flat to gently sloping
15. Geotechnical/soils suitability
16. Outside FEMA floodway/high-risk flood zone
17. Stormwater management feasible
18. Seismic and natural-disaster risk acceptable
19. Phase I environmental complete
20. Phase II environmental (if warranted)
21. Wetlands/waters delineation
22. Endangered-species/habitat review
23. Historical/archaeological review
24. Contamination/remediation scope (if any)
25. Long-haul fiber proximity
26. Multiple carriers available
27. Dark fiber availability
28. Physically diverse fiber paths possible
29. Latency profile fits intended workload
30. Municipal/well/surface water access
31. Water rights and volume confirmed
32. Reclaimed/non-potable water option
33. Sewer capacity available
34. Zoned industrial or data-center use
35. Rezoning/variance needs identified
36. Local jurisdiction supportive
37. Setback and height requirements checked
38. Noise/aesthetic ordinances reviewed
39. State/local tax incentives available
40. Sales-tax exemption on equipment
41. Property-tax abatement potential
42. Road access and weight limits adequate
43. Rail access (for heavy equipment)
44. Construction labor availability
45. Operations labor availability
46. Proximity to demand/market fit
47. Title clean / encumbrances reviewed
48. Easements and ROWs identified
49. Control structure (option/land bank) in place
50. Exit/monetization path defined

Brownfield Redevelopment Checklist (30 points)

Specialized diligence for converting a retired industrial site — coal plant, mill, or factory — into a data center campus.

1. Prior industrial use documented
2. Existing electrical service capacity
3. On-site substation present and condition
4. Switchyard present and condition
5. Existing transmission ties and voltage
6. Interconnection rights transferable/intact
7. Utility willingness to re-serve the load
8. Usable capacity of existing infrastructure verified
9. Demolition scope and cost estimated
10. Salvage value of existing structures
11. Phase I environmental complete
12. Phase II sampling completed
13. Contamination extent characterized
14. Remediation plan and cost
15. Regulatory remediation status/program
16. Liability protections (e.g., applicable programs)
17. Asbestos/lead/hazardous materials survey
18. Existing water intake/rights
19. Cooling-water feasibility
20. Rail spur condition and access
21. Existing zoning (industrial) confirmed
22. Large contiguous acreage confirmed
23. Soil/foundation suitability (heavy industrial)
24. Existing roads and site access
25. Gas service availability
26. Fiber proximity to the site
27. Community/political support
28. Incentives for redevelopment available
29. Timeline advantage vs. greenfield quantified
30. Powered-value vs. industrial-value gap modeled

Utility Due Diligence Checklist (50 points)

A deep dive on the single most important factor — power — and the utility relationship behind it.

1. Serving utility and service territory confirmed
2. Regulated vs. deregulated market
3. Grid operator / RTO / ISO (or TVA, etc.)
4. Current available capacity for new load
5. Planned capacity additions
6. Generation mix serving the area
7. System reliability history
8. Nearest transmission line voltage
9. Spare capacity on nearby lines
10. Distance to interconnection point
11. Substation proximity and capacity
12. New substation feasibility and cost
13. Interconnection process and steps
14. Interconnection queue position
15. Study timeline (feasibility/system impact)
16. Estimated network upgrade costs
17. Cost allocation (who pays upgrades)
18. Large-load tariff terms
19. Energy rate ($/kWh) and structure
20. Demand charges
21. Rate escalation history/risk
22. Minimum take/contract terms
23. Power factor/quality requirements
24. Redundant feeds available
25. Dual-substation feasibility
26. Behind-the-meter generation allowed
27. Standby/backup tariff terms
28. On-site gas turbine feasibility
29. Gas pipeline capacity and pressure
30. Renewable/PPA options
31. Nuclear/SMR co-location potential
32. Curtailment/interruptible options
33. Utility's data center track record
34. Utility staffing/responsiveness
35. Letter of intent / capacity reservation
36. Conditional power commitment terms
37. Milestone and deposit requirements
38. Transformer/switchgear lead times
39. Construction coordination with utility
40. Metering and SCADA requirements
41. Easements for utility infrastructure
42. Future capacity expansion path
43. Water utility capacity (for cooling)
44. Water rate and availability
45. Sewer capacity
46. Stormwater/discharge permits
47. Environmental/air permits for generators
48. Regulatory approvals for new load
49. Stakeholder/community considerations
50. Contingency if power is delayed/denied

Data Center Site Selection Scorecard

A weighted worksheet to compare sites objectively. Score each factor 1–10, multiply by the weight, and sum. Power-related factors carry the most weight by design — reflecting that power is the binding constraint.

Tip: any score of 1–2 on "power availability & timeline" is usually a kill criterion regardless of total — a site you cannot power is not a data center site.

Glossary & Keyword Index

Quick definitions of the core terms in this guide, useful for orientation and for search.

Keyword index: data center land, powered land, data center development, data center investment guide, data center site selection, how much power does a data center need, data center land requirements, brownfield redevelopment data center, coal plant conversion data center, TVA data center sites, hyperscale development, AI infrastructure real estate, data center opportunity zone, data center due diligence, data center power availability, megawatts, gigawatt, interconnection queue, substation, switchyard, behind-the-meter generation, N+1 redundancy, 2N redundancy, colocation, edge computing, powered shell, water usage, dark fiber, cost per MW, yield on cost, development spread, AI factory, GPU cluster, small modular reactor, Ashburn, Dallas, Phoenix, Atlanta, Nashville, Memphis.