Why This Site is Prime for AI Data Centers

The numbers don't lie: AI infrastructure demand is outpacing every projection made just two years ago. Hyperscalers are signing land deals in markets they would have ignored in 2021, utilities are fielding interconnection requests that dwarf their entire existing load, and zoning attorneys who specialize in data centers are suddenly the most sought-after people in commercial real estate. When a site gets positioned for AI data center development, that's not a speculative bet — it's a calculated response to one of the most capital-intensive infrastructure buildouts in modern history.

This particular acquisition is worth examining closely because the path from raw land to operational AI facility is longer, more expensive, and more complex than most people outside the industry realize. Getting the fundamentals right at the site selection stage determines whether a project delivers or dies in permitting.

Zoning Is Where Projects Are Won or Lost

Before a single kilowatt of power gets drawn or a single server rack gets installed, a site has to survive zoning. This is the unglamorous part of data center development that rarely makes headlines — but it's where projects quietly fail.

Data centers, and AI-optimized facilities in particular, are an awkward fit for most zoning classifications. They're not manufacturing. They're not traditional office. They generate significant electrical demand, require large cooling infrastructure, and produce low employment density per square foot — all of which can trigger opposition from planning commissions that weren't anticipating this type of use when they wrote the underlying ordinances.

The smartest developers don't just ask whether a site is zoned correctly — they ask how defensible that zoning is when neighbors show up to oppose it.

Jurisdictions that have updated their zoning codes within the last five years with explicit data center classifications — and there are more of them than you'd expect, particularly in the Sun Belt and Mountain West — offer a meaningful advantage. They've already worked through the definitional questions: What constitutes a data center use? How are cooling towers treated as structures? Does the facility trigger industrial-level environmental review or something lighter?

For the site in question, the reference to zoning review in the acquisition signals that this work is either underway or explicitly anticipated. That's the right sequence. Locking up a site before confirming zoning viability is how developers end up owning expensive land they can't use.

What AI Actually Demands From Infrastructure

A standard enterprise data center built in 2015 might operate at 200-500 watts per square foot. An AI training cluster running dense GPU configurations — the kind needed for large language model development or frontier model training — can push 1,000 watts per square foot or higher. That's not a rounding error. It's a fundamentally different infrastructure problem.

Power isn't just about megawatts available; it's about how fast those megawatts can actually be delivered to a site.

The transmission and substation picture matters enormously. A site with 50 MW available on a 3-year timeline is, functionally, worth less than a site with 20 MW available in 18 months. AI workloads can't wait. The hyperscalers and the well-capitalized colocation operators who serve AI tenants are making decisions on 12-18 month build timelines, which means they're prioritizing sites where power delivery is already de-risked.

Beyond power, fiber diversity is non-negotiable. AI inference workloads require low-latency connectivity to serve end users, while training workloads demand enormous bandwidth for data ingestion. A site that sits on or near major fiber routes — or that can be reached with a relatively short fiber run to a major carrier hotel — clears a meaningful hurdle. Sites that require custom fiber builds of 20+ miles are workable, but they add cost and timeline risk.

Cooling is the third leg. Water availability and water rights have emerged as serious constraints in arid markets. The data center industry has been making genuine progress on power usage effectiveness (PUE) and water usage effectiveness (WUE), with hyperscalers publishing annual sustainability reports that track these metrics. But the underlying physics don't change: dense AI compute generates heat, and that heat has to go somewhere. Whether a site supports evaporative cooling, dry cooling, or liquid cooling configurations depends heavily on local climate and water access.

Site Plan Review: The Devil Is in the Details

Site plan review is the formal process through which local planning authorities evaluate whether a proposed development meets all applicable standards — setbacks, height limits, traffic impact, stormwater management, utility connections, and more. For a data center, this process has some specific wrinkles worth understanding.

Diesel backup generators are the most common flashpoint. Large data centers may have dozens of generators on site, each a permitted air emission source. In jurisdictions with aggressive air quality regulations — California's South Coast AQMD being the most notorious example — generator permitting alone can add 12-18 months to a project timeline. Some developers have responded by exploring alternative backup strategies, including battery energy storage systems (BESS), though most operators still consider diesel the most reliable option for the kind of extended outage scenarios that actually matter.

Traffic impact studies can be surprisingly contentious. Data centers have very low ongoing traffic — maybe a handful of employees per shift, some periodic equipment deliveries. But construction traffic is substantial. A large greenfield build might involve thousands of heavy truck trips during the foundation and structural phases. Planning commissions in areas that aren't accustomed to large industrial construction sometimes underestimate this, then overcorrect when neighbors complain.

The developers who move fastest through site plan review are the ones who over-prepare the application package, not the ones who negotiate with planning staff after the fact.

Engaging a local land use attorney before the application is filed — someone who knows the specific planning commission, understands their concerns, and has relationships with the relevant staff — is not a luxury. It's table stakes for projects of this scale.

The Growth Trajectory Is Real, and It's Accelerating

Grid operators across the country are publishing load forecasts that would have been dismissed as fantasy three years ago. PJM, which covers the mid-Atlantic and Midwest, has revised its 10-year demand growth forecast dramatically upward, driven almost entirely by data centers and electrification. MISO, ERCOT, and SPP are seeing similar dynamics. The utilities themselves are scrambling to build generation and transmission capacity that takes 5-7 years to permit and construct — while AI data center demand is materializing on 2-3 year timelines.

That gap is not going to close quickly. Which means sites that can plug into existing transmission capacity — rather than waiting for new lines to be built — carry a structural premium that will persist through at least the early 2030s.

The AI model itself is also evolving in ways that affect site requirements. Inference — running trained models to generate outputs — is growing faster than training as a share of compute demand, partly because inference happens every time a user interacts with an AI product, while training happens in discrete cycles. Inference workloads have different characteristics: lower peak power density, higher network throughput requirements, and stronger preferences for geographic distribution closer to end users. This is creating demand not just for hyperscale campuses in low-cost power markets, but for edge-adjacent facilities in secondary markets that previously weren't on the data center map.

Finding the Hidden Value in Site Development

The acquisition of a site for AI data center positioning is, in many cases, only the beginning of the value creation story. Local and state incentives can dramatically alter the economics — and they're more widely available than developers outside the sector typically realize.

Many states offer property tax abatements for data center investments that meet certain capital expenditure thresholds. Virginia's data center sales tax exemption is the most famous, but similar programs exist in Texas, Georgia, Nevada, and a growing number of Midwestern states actively competing for this investment. Some jurisdictions are now negotiating direct agreements — effectively a payment in lieu of taxes structure — that give developers certainty and give municipalities a predictable revenue stream.

Workforce development partnerships with community colleges and technical schools are another underutilized lever. Data centers employ relatively few people once operational, but the construction and commissioning phases are labor-intensive, and operators increasingly want a pipeline of local technical talent for ongoing operations. Jurisdictions that can credibly offer both tend to get better terms in incentive negotiations.

The site that gets acquired today and positioned correctly — clean title, viable zoning path, confirmed power access, and a realistic site plan review timeline — is the site that attracts a creditworthy tenant or buyer in 18-36 months. In an environment where AI infrastructure demand is structurally exceeding supply, that sequencing is how value gets created.

The developers who understand that are already moving. The ones still waiting for perfect certainty will find that the best sites are gone.

Explore the InfraSale Marketplace for prime AI data center sites!

[INTERNAL LINK: AI infrastructure demand]

[INTERNAL LINK: zoning classifications]

[INTERNAL LINK: site plan review process]