America is not overbuilding. It is building the wrong asset class.
Over the past twelve months, U.S. data center construction spending reached a record $75 billion. Core market vacancy sits below 10 percent. Absorption cycles run four to six months. Asking rents still have another 15 to 20 percent of upside in several primary corridors. In the same country, residential and office product is stacking into an oversupply that already exceeds 18 months of inventory in a growing list of secondary metros. Some submarkets have gone past 24 months. Same labor pool. Same capital markets. Two completely different balance sheets.
Markets don't make polite distinctions between real estate sectors. They simply price future cash flow. Right now, those cash flows are moving from housing into compute.
The Ledger Is Physical
I have spent the last decade reading physical infrastructure through financial statements, and the last five years tracking token flows through the same lens. The lesson stays constant: scarcity, not narrative, sets final prices. In 2025, I watched spot Bitcoin ETF inflows change institutional behavior within weeks. The $75 billion data center construction print is a slower version of the same signal, but it is larger and more structural. Capital is not fleeing real estate. It is fleeing the wrong kind of real estate.
Data centers are not classic CRE. The old rule was location, location, location. The new rule is interconnection, interconnection, interconnection. Tenants are not families; they are AI clusters, cloud regions, edge nodes, and compute workloads. Demand follows power capacity and fiber distance, not school districts and transit lines. That is why California, Texas, and Virginia dominate the absorption stats. Those states are not the most populated or the cheapest. They are the places where megawatts are still reachable. The scarce asset model changed, and the official real estate metrics have not fully caught up.
Sentiment is the invisible ledger of value. During the AI capex surge, sentiment has shifted to data centers faster than most builders can add capacity. The inventory data shows why that repricing is justified. A vacancy rate below 10 percent is healthy. A four- to six-month absorption window, at current leased-up levels, is effectively zero slack. In residential real estate, four to six months of supply is equilibrium. In data centers, it means the landlord sets terms, not the tenant.
The Balance-Sheet Divergence
The second signal lives not in the property but in the financing stack. Apply the same three-red-line stress test that Asian regulators apply to developers. Equinix and Digital Realty generally carry asset-liability ratios below 60 percent, net gearing below 80 percent, and cash-to-short-term-debt coverage above 1.5 times. Their construction debt is long-dated. Their effective costs sit in a range of 4 to 6 percent. Operating cash flow keeps turning positive, and signed leases convert at high rates. There is no equity-call risk that keeps analysts awake at night.
Now run the same test on the traditional homebuilders. Lennar, D.R. Horton, and KB Home are not near default, but their working capital cycles are stretching. Land options are becoming less flexible. Sales conversion has weakened under 7 percent mortgages. Financing costs on marginal projects are running closer to 8 to 12 percent. Inventory write-downs are beginning to appear in lower-tier submarkets, and the market is quietly making them carry the cyclical risk of the whole economy.
This credit spread is the most important daily technical I follow. It tells you that fixed-income desks have already made a sector call. They are treating data center REITs as contracted infrastructure and homebuilders as discretionary consumer cyclicals. The distinction is fair. A data center lease is a cash-flow contract. An unsold house is a cash-flow expense that begins the moment the foundation is poured.
Policy Has Already Picked a Side
The policy layer reinforces the split. State governments now treat data centers the way they once treated auto plants. They offer tax abatements, fast-track approvals, and whitelist permitting for large power users. The federal government is not imposing a nationwide housing rescue, because the inflation fight is still controlling the rate path. The result is an asymmetric incentive structure: data center projects get streamlined regulatory treatment, while residential projects wait for mortgage rates that are not going to move quickly.
Based on my audit work across both traditional and digital asset companies, I have come to believe that a signed grid interconnection is now better collateral than a clean title to raw land. Land without power is passive. Power without land can still be contracted. The data center company that controls a substation slot has solved the true bottleneck, and the market prices that optionality into the equity.
Speed is the only currency that never depreciates. For data center developers, speed means bringing megawatts online before the AI demand curve flattens. For traditional homebuilders, speed is the enemy: faster completion just adds finished inventory to a market that cannot absorb it. The strategic response has been predictable. Large landlords are pivoting toward build-to-suit data centers. Residential developers are pivoting toward build-to-rent or shedding land. The old playbook is no longer a template. It is a warning.
Contrarian: The Choke Point Is Not Concrete
Now the part the mainstream story ignores. The conventional view is simply AI good, housing bad. That misses the real shared constraint, which is electricity.
Every new data center campus is a demand event that the grid was not designed to serve. Interconnection queues in the U.S. are already long. Transformer lead times stretch for years in some regions. ERCOT sends scarcity signals through wholesale prices, while Virginia is bumping against substation capacity limits. State whitelist policies make data centers go to the front of the line. That is efficient for AI capital. It is not neutral for everyone else.
Data centers are, in effect, consuming the same grid capacity that residential electrification and small business expansion need. The crowded-out cost does not show up on the data center operator's income statement. It shows up in delayed residential service upgrades, higher utility bills, and tighter peak-demand margins. The hidden negative externality in the $75 billion boom is not carbon. It is queue position. Power capacity is becoming a zero-sum resource, and data center lessors have larger balance sheets than most homebuyers. They will win the auction.
There is another missing ledger item that most coverage refuses to address: crypto mining. Some of the same data center shells that now host AI workloads still host digital asset mining, especially when energy prices collapse or network demand lulls. Mining hardware gives the facility a unique form of load flexibility. The operator can curtail when power is expensive and mine when price is low. The same physical asset is therefore a technology lease and an energy option. The official narrative often treats this as irrelevant, but it matters for revenue stability in periods of AI capex digestion.
I came to crypto through software engineering and stayed through market crashes and ETF approvals. The deepest lesson of decentralized finance is that trust is code, not character. The same lesson now applies to physical infrastructure. A data center with a signed interconnection agreement is more trustworthy than a borrower with a good credit score and no grid capacity. Collateral is increasingly defined by access to power, not by ownership of land.
What To Watch Next
The forward-looking trade is not forever long data centers and short homebuilders. The next two quarters will reveal who can convert construction commitment into energized supply. Watch data center construction spending for a monthly year-over-year growth rate above 10 percent. Watch vacancy for a drop below 8 percent, which would force another leg of rent increases. Watch housing starts for a decline below 3 percent year over year, which would signal accelerated price clearing in residential. Watch the 30-year mortgage rate, and the moment it drops below 6 percent, the homebuilding equity will trade before the real estate data confirms it.
That is the paradox. The data center boom is not separate from the housing crisis; it is partly causing it. As long as the grid is capped, digital infrastructure growth will keep imposing an energy tax on traditional property. The next critical insight will not come from a ribbon-cutting photo. It will come from a transformer order book or a utility interconnection filing.
Markets don't wait for perfect information. Speed is the only currency that never depreciates. The same applies to information: the market already knows the story. The edge now belongs to whoever can read the power map before the lease is signed.