Read time: ~5 minutes
TL;DR
Thesis: The market still prices industrial real estate by the square foot. In AI infrastructure, the scarce input is grid access, so power-rich legacy sites are systematically mispriced relative to new construction.
The number that matters: Five years; the median interconnection wait for projects completed in 2023. The entire premium is a function of that queue.
Biggest risk: The queue clears. FERC interconnection reform, utility buildout, or on-site generation scaling faster than expected would compress the scarcity premium, and an AI capex slowdown would kill the demand side outright.
Bottom line: Until interconnection timelines normalize, value accrues to whoever already controls the electrons: grandfathered hookups, on-site generation, and the 128-week switchgear backlog.
For the past century, commercial real estate has been priced by a single metric. Whether you were valuing a Class-A office tower in Manhattan or a logistics hub in Ohio, the entire financial model hinged on price per square foot.
The market is still applying that framework to the AI buildout. Developers are putting up million-square-foot warehouse shells on the assumption that if they build the box, hyperscalers will show up to fill it with server racks.
They won't.
Square footage without power is now a liability. Space is abundant, electricity is finite, and real estate valuations have stopped tracking the size of the floorplan. They track the capacity of the substation.
Welcome to the Megawatt Premium.
The thermodynamic reality of compute
During the cloud boom of the 2010s, a standard data center rack drew 5 to 10 kilowatts. A 10-megawatt facility was considered large, and the local municipal grid could support it without much drama.
AI training broke that math.
A single rack of Nvidia's latest silicon routinely draws 35kW to 120kW. Aggregate tens of thousands of those chips into one training cluster, and the demand curve goes vertical. The industry has moved from planning 100MW campuses to designing facilities measured in gigawatts. One gigawatt is roughly the electricity demand of a mid-sized American city, and you can’t call the local utilities provider to order one.
The interconnection chokepoint
The chips are no longer the constraint. The wires are.
PJM Interconnection, the largest grid operator in the country, and ERCOT, which runs the Texas grid, are both buried in connection requests. According to Lawrence Berkeley National Laboratory's Queued Up data, the typical project completed in 2023 spent a median of five years in the interconnection queue between its initial request and commercial operation.
Capital allocators cannot wait half a decade to deploy compute. The technology cycle moves too fast. A GPU purchased today will be obsolete before the utility finishes laying the high-voltage lines.
So the sharpest operators are bypassing the grid expansion process entirely.
A valuation premium is now attached to legacy industrial assets that already hold significant power rights.
Developers are buying decommissioned coal plants, bankrupt aluminum smelters, and shuttered textile mills. They have no interest in the buildings. They strip the structure to the concrete pad and keep the grandfathered electrical hookup because the utility permit is the entire asset.
These sites now trade on a dollar-per-megawatt multiple. The square footage doesn't appear in the model.
The institutional playbook
For family offices and institutional allocators, the application layer of AI is a venture bet. The physical bottleneck is a toll booth. Three ways capital is playing it:
1. Land-banking legacy grid connections. Syndicates are acquiring heavy industrial sites in secondary markets, with mandates limited to properties that hold existing high-voltage substations, the water rights needed for liquid cooling loops, and zoning permissive enough to allow continuous noise generation. The properties get entitled for data center use, then flipped to Tier-1 operators.
2. Private credit for "bring your own power" infrastructure. Operators tired of waiting are building on-site natural gas turbines and industrial-scale battery storage, funded through private debt. These microgrids sever the dependence on the delayed municipal grid, and lenders are financing the electrical equipment that makes each site energy-independent.
3. The switchgear supply chain. Even with land and power secured, you still need transformers and switchgear to step the voltage down to the racks. Wood Mackenzie's Q2 2025 survey puts the average lead time for a power transformer at 128 weeks, with generator step-up units at 144 weeks, and prices up 77 percent since 2019. Private equity is rolling up the fragmented middle-market manufacturers that build these components.
The bottom line
The next decade of infrastructure is a race to secure energy. Firms that keep evaluating compute through the lens of traditional commercial real estate will misprice the entire market, while the operators who control the interconnects, the switchgear, and the cooling will own the toll roads of the digital economy.
The software gets the headlines. The substation gets the pricing power.
Bedrock Capital Research publishes independent research for informational and educational purposes only. Nothing in this memo is investment, legal, or tax advice. Authors may hold positions in assets discussed.