The first generation of data center hubs emerged near fiber networks and major metros, but AI has changed the scale of demand. Projects that once required 5-20 megawatts now request 100-300 megawatts at a time, making electrical certainty more valuable than downtown addresses. As Tier-1 markets reach capacity, operators are moving into exurban and rural corridors with access to transmission, water, incentives, and a cooperative permitting culture.

Why clusters form, and why they’re capped

From Georgetown’s campus and the Steers Center for Global Real Assets, it’s less than an hour northwest to “Data Center Alley” in Ashburn, Virginia, the world’s largest data center hub. On the Dulles Toll Road from Washington Dulles International Airport to downtown Washington, D.C, the pattern is hard to miss; empty fields have turned into rows of windowless shells, bracketed by substations and high-voltage lines.

U.S. Tier-1 clusters – Northern Virginia (Ashburn), Atlanta, Silicon Valley (Santa Clara), Dallas, Chicago, Phoenix, Hillsboro – formed where four ingredients overlapped:

• Deep fiber routes and peering points. 
  • Ashburn sits where backbone fiber networks converge. Santa Clara connects to Bay Area Internet Exchange Points(IXP) and subsea routes. Hillsboro links into trans-Pacific cables.
• Substation capacity and relatively cheap power
  • Utilities could deliver tens of megawatts at a time without grid upgrades as long as demand grew in known increments.
• Experienced permitting ecosystems used to large, power-dense buildings
  • Local governments had already seen large data-center proposals and knew how to evaluate them
• Large, aggregable land parcels near metro areas. 
  • Not downtown towers, but 10-200 acre blocks close enough to labor, airports, and fiber.

Once the first wave of data centers arrived, network density, utilities, and specialized labor followed. This created the perfect firestorm; more electrical and operational certainty allowed more data centers to enter the market, which then gave more certainty to those same vendors to keep providing in that market.

However, those T1-clusters are becoming capped. According to CBRE’s H1 2025 report, primary North American data-center markets hit a record-low 1.6% vacancy, with Northern Virginia the largest market and among the tightest.

JLL notes that pre-leasing sits around 73%, meaning most capacity is spoken for years before delivery, keeping vacancy restrictive through at least 2027.

In practice, that means:

• Substations are already heavily committed or need expensive upgrades
• Interconnection queues, utility permission to attach new power, can add 3-10 years with delays
• Large, well-located parcels are scarce or politically contentious

Hubs like Northern Virginia still matter, but they can’t absorb all of the AI era’s new demand. So, where is all this new data center development activity going?

The rural paradox: “middle of nowhere” = “center of everything”

On a geographical map, many next-wave data center sites look like they’re in the middle of nowhere: exurban cornfields, timber tracts, industrial parcels outside small towns. On a grid map, they’re on top of the action.

As the original cluster crowds out, developers are looking to more non-traditional markets to develop. They’re chasing the following ingredients:

• Large, contiguous land parcels where 100-1,000+ MW campuses can grow in phases
• Adjacency to transmission corridors instead of a 30-mile spur line
• Water strategies, whether that’s reclaimed wastewater, industrial effluent, or designs that minimize potable water use
• State and local incentives that recognize data centers as economic anchors
• Predictable permitting timelines and communities willing to partner rather than litigate

That’s why you see emerging hot spots in places like:

• Dallas-Fort Worth’s outer ring, where utilities, counties, and policy are aligned
• Iowa and other Mid-continent markets, where existing generation headroom allows multi-campus builds
• Exurban edges of Phoenix, Richmond, or Columbus, where land is cheaper but fiber can still be stitched in

“Rural” in this context doesn’t mean remote in the traditional sense, it means adjacent to wires, water, permits, and willing partners, even if the nearest central business district is over an hour away.

AI is the demand engine reshaping power and siting

The reason all of this matters now is simple: AI is hungry.

One generative AI training campus can request 100-300 MW on day one, the equivalent of powering 80,000+ homes. Instead of incremental expansion, hyperscalers want baseload commitments, often 10-20 years long. McKinsey estimates that data centers could account for 30-40% of all net new U.S. electricity demand through 2030, with roughly 400 TWh of added consumption ~23% CAGR.

Other analyses suggest global data-center power demand could more than triple by 2030 as AI workloads scale.

For real-assets investors and utilities, that translates into:

• Bigger, lumpier loads: single projects requesting 100 MW+ in one geography
• Longer commitments: 10-20-year contracts tied to AI clusters
• More stringent availability requirements: power is now as mission-critical as the building itself

As hyperscalers race to secure capacity, they’re less focused on a downtown ZIP code and more on one question: Can you energize 100-300 MW on a realistic timeline, and then double it?

Case study: Louisiana, when utility approvals redraw the map

Louisiana is a textbook example of how a power-first strategy can turn a “non-hub” state into an AI anchor.

In August 2025, Louisiana’s Public Service Commission approved Entergy Louisiana’s proposal to move forward on major generation and transmission investments as part of Meta’s planned mega-campus in Richland Parish. When completed, this will be Meta’s largest data center globally, with an announced investment of around $10 billion. 

A few things make this notable:

• The utility structured upgrades benefit the wider grid, not just Meta.
• Regulatory approvals arrived before physical buildout, not after.
• The energization timeline was made predictable enough to unlock private capital.

Here, Meta asked: “Which jurisdiction will guarantee the energy?” Louisiana answered with a credible, long-term power plan and political alignment.

That’s how a region that looks “middle of nowhere” on a hiring map becomes “center of everything” on an AI-power map.

Regional outlook: PJM, MISO, Northwest, Southeast

Zooming out, you can see how different grids shape the new geography:

• PJM Interconnection (Mid-Atlantic / Midwest):
  • Home to Northern Virginia and emerging Pennsylvania clusters
  • Constrained by interconnection backlogs and slow approvals 
  • Pennsylvania is a live test of pairing gas and legacy hydro assets with new AI clusters
• MISO (Midwest/parts of South):
  • Attractive for phased campuses
  • Wind, solar, and thermal generation balance
  • Iowa and neighboring states offer transmission adjacency + ample land
• Northwest:
  • Abundant hydroelectricity
  • Limited by policy constraints and transmission bottlenecks
• Southeast:
  • States like Georgia and Alabama view data centers as industrial anchors
  • Pair incentives with new substations and transmission
  • Local permitting cultures are more predictable

Investors should view this not as a list of states, but as a menu of grid philosophies and regulatory behaviors.

Bottom line: geography is strategy

The geography of AI infrastructure is undergoing rapid change.  

• Core hubs, such as Northern Virginia, will remain central to network density, but they are effectively sold out.
• The next wave is heading into exurban and rural corridors where land, power, water, and permitting align.
• AI’s power appetite makes it possible, and in many cases preferable, to build “in the middle of nowhere,” so long as that location is wired into a reliable grid.

From a car window on the Dulles Toll Road, those low, windowless buildings might look anonymous & misplaced. From a real-assets lens, they’re signposts for where power, policy, and AI demand intersect, and they’re quietly redrawing the map of the U.S. economy.