Compute Factories Are Remaking U.S. Real Estate: A Field Manual for Developers and Builders
Across the heartland, farmland and factory shells are turning into hyperscale campuses that draw city-scale electricity. The dust and heat in West Texas frame a simple truth: AI runs on land, steel, and megawatts. OpenAI's Stargate campus in Abilene is one example - thousands of workers, a footprint like a small city, and a path to more than a gigawatt of capacity.
This isn't a trend piece. It's a site, a substation, and a procurement schedule. If you build, own, or finance large projects, the next five years will be decided by how fast you can deliver energized real estate.
Where the money is going (and why it matters to you)
The top hyperscalers are on pace for hundreds of billions in annual capex, with roughly three-quarters pointed at AI infrastructure. A growing share is funded with debt, not cash. Bond issuance has surged, and credit-default swaps on some names have widened - a sign that lenders smell risk.
That backdrop affects you at the jobsite. Debt-fueled schedules create pressure for faster delivery, firmer milestones, and ironclad change control. Expect more phasing, more take-or-pay structures, and more scrutiny on long-lead equipment.
The new map: who's building what
- OpenAI: Stargate in Abilene, TX - multi-phase, targeting 1+ GW. Ballpark cost per site: around $50B, with a multi-campus vision pushing totals into the hundreds of billions.
- Meta: Hyperion in northeast Louisiana - multi-million-square-foot AI campus, projected to draw more electricity than New Orleans.
- Google: A new campus on 1,100 acres in West Memphis, AR - pitched as the largest private capital project in state history.
- xAI: Colossus in Memphis, TN - a shuttered factory flipped in 122 days; expansion under way with a purchased power plant across the state line.
- Microsoft: Southeast Wisconsin - a multi-billion dollar AI data center slated to house hundreds of thousands of accelerators.
- Amazon: Project Rainier in rural Indiana - 1,200 acres converted for model training, running custom silicon for Anthropic.
Different logos, same constraint: power first, dirt second, everything else follows.
The constraint is electricity, not enthusiasm
Executives agree on the bottleneck. Cash is abundant. Transformers, substations, and interconnects are not. A single mature campus can top a gigawatt - enough to match a mid-sized city.
Utilities move on permitting clocks and regulatory cycles, not quarterly earnings. Interconnect queues are congested. High-voltage gear lead times bite. That's where deals win or die.
What the "circular" AI economy means for project risk
One company's capex is another company's revenue - and often another's equity or supply commitment. OpenAI's web of announced frameworks with chip suppliers, cloud partners, and landlords illustrates the loop. It works until it doesn't. If demand softens or financing tightens, stress can ricochet through shared dependencies.
For owners and GCs, that means you need binding offtake, credit support, and step-in rights - not just headlines. Treat "framework agreements" as intent, not certainty.
Site selection checklist for hyperscale AI
- Transmission: Proximity to 230-500kV lines, clear right-of-way, and available substation sites. Get a utility letter of intent early.
- Interconnect: Enter the queue now. Model N-1/N-2 contingencies and plan for curtailable loads.
- Power mix: Gas, renewables with firming, or bilateral PPAs. Consider on-site generation and heat reuse options.
- Water and cooling: Secure non-potable/reclaimed sources. Plan for direct-to-chip liquid cooling and water-neutral strategies.
- Soils and drainage: Overbuild for flash rain and expansive clays. Design to keep heavy equipment moving.
- Permitting: Fast-track jurisdictions, industrial zoning, and predictable environmental review. Lock tax incentives early.
- Logistics: 24/7 heavy haul access, laydown space, and near-port or rail options for chillers, gensets, and prefab modules.
- Workforce: Skilled trades within 60-90 minutes; housing and transit for a peak workforce measured in thousands.
Procurement: the long-lead gauntlet
- Order now: Grid transformers, switchgear, generators, chillers, dry coolers, and busway. Multi-source where possible.
- Substations: Co-develop with utilities; pursue standard designs to compress approvals. Pre-stage pads and oil containment.
- Modules: Use repeatable, prefabricated data halls and skids. Standardize to cut rework and QA lag.
- Supply security: Dual suppliers, bonded inventory, and transparent factory slots. Track Tier-2 constraints (resins, cores, semis).
Contracts and financing that survive cycles
- Phasing: Break into MW tranches; tie payments to energization, not just substantial completion.
- Credit: Parent guarantees, letters of credit, and take-or-pay for base capacity. Watch RPO vs. binding commitments.
- Risk share: EPC with liquidated damages, index-linked price protection, and clear change-order gates.
- Exit ramps: Step-in rights, cure periods, and remarketing rights if a tenant stalls or pivots chip strategy.
12-month playbook to get ahead
- Month 0-3: Lock land options on multiple sites. File interconnect applications. Order transformers and main switchgear.
- Month 3-6: Secure water rights and discharge permits. Finalize incentive packages. Release early works: mass grading, access roads, and duct banks.
- Month 6-9: Issue EPC. Kick off substation civils. Start steel, foundations, and prefab module fabrication.
- Month 9-12: Set transformers, pull feeders, and dry-fit cooling. Commission first feeders to support staged energization.
Operational realities you can bank on
- Demand is lumpy: Training spikes; inference is always-on. Design for base load with surge capacity.
- Chip cycles move: Layouts must flex for new accelerators, higher rack densities, and liquid-cooled retfits.
- Water scrutiny rises: Favor reclaimed sources and closed-loop systems to reduce community friction.
- Community optics matter: Tie incentives to local jobs, workforce training, and grid upgrades residents can feel.
Policy and power: watch these signals
- Federal incentives: Proposals to extend manufacturing credits and apply them to data center builds come and go. Track official updates from the CHIPS Program Office.
- Grid planning: Regional transmission expansion and queue reform will dictate where gigawatt campuses can plug in. Energy intensity of data centers is rising; see current analyses from the IEA.
The takeaway for developers, owners, and GCs
AI tenants move fast and speak in frameworks; your projects must move faster and be backed by hard commitments. Anchor sites near big wires, pull long-leads forward, and phase builds around energization. Price risk into contracts, and don't confuse a press release with revenue.
Some investors will get ahead of demand. Others will be late and lose deals to interconnect timing. The winners will be the ones who can turn dirt into dependable megawatts on schedule - again and again.
Skills and teams
You'll need staff who can speak AI buyer language and translate it into design, procurement, and commissioning. For practical upskilling by job function, see Complete AI Training.
Final word
"People will get burned on overinvesting," one AI leader said. "People also get burned on underinvesting." In this cycle, the safest position isn't the cheapest land - it's the site that can switch on first.
Your membership also unlocks:
