Data Centers Are the New Industrial Real Estate Asset Class

For about fifteen years, the smartest boring money in real estate bought warehouses. E-commerce needed shelves near cities, Amazon needed same-day fulfillment, and every logistics box you could put near a highway interchange filled up and raised its rent. Prologis became a giant on that trade. It worked because the demand curve was structural, not cyclical.

That trade is over. Not dead, but done growing the way it grew. In 2025 US logistics and warehouse rents actually fell 4.5% year over year, with coastal markets down 7.6%, and industrial vacancy drifted up to around 7.4%. The warehouse boom digested itself. Meanwhile a different kind of industrial box, the data center, did the exact opposite, and the gap between the two is the whole story.

Here is the way I think about it. Every generation of industrial real estate is really a bet on where physical scarcity meets a structural demand wave. Railroads made warehouses near tracks valuable. Interstates moved that value to highway interchanges. E-commerce moved it again, to the edge of dense metros. The AI buildout is doing the same thing one more time, except the scarce input this time is not location. It is power. That single substitution changes almost everything about how you underwrite the asset.

The numbers that flipped the trade

Start with occupancy, because occupancy is where a real estate thesis lives or dies. CBRE clocked colocation vacancy across the primary US markets at 1.4% at the end of 2025. That is a record low, and it is a number you basically never see in real estate. Northern Virginia, the largest data center market on the planet, ran at roughly 0.5%. For comparison, that same year warehouse vacancy sat around 7.4%. One asset class has nothing available. The other has a comfortable cushion.

Scarcity like that shows up in rent, fast. The average asking rate for a 250-to-500 kW wholesale colocation requirement rose 6.6% in 2025 to a record $196.25 per kW/month. Notice the unit. Data center leases are priced per kilowatt, not per square foot, because power is what the tenant is actually renting. And at the large end, where the AI money lives, pricing for 3-to-10 MW deployments jumped 12.5% year over year. The biggest boxes are getting the steepest rent hikes, which is the signature of a genuine supply shortage rather than a broad rising tide.

Total supply in the primary North American markets reached 9,432 MW at the end of 2025, a 36% jump in a single year. Read that carefully. Supply grew 36% and vacancy still hit a record low. That only happens when demand is outrunning a frantic buildout. In warehouses, a 36% supply surge would have crushed rents. In data centers it barely kept the lights on.

Where the demand actually comes from

The demand is not vague AI enthusiasm. It is capital budgets you can read off earnings calls. The five largest US hyperscalers, Microsoft, Alphabet, Amazon, Meta, and Oracle, have collectively guided to somewhere between roughly $600 billion and $725 billion of capital expenditure in 2026. That is up about 36% over 2025, and around 75% of it is pointed at AI infrastructure. This is not a survey of intentions. It is money already committed by companies that have to answer to shareholders for it.

Goldman Sachs Research put the power demand on a curve: US data center electricity demand rising from 31 GW in 2025 to 41 GW in 2026 and 66 GW by 2027. That more than doubles data centers' share of US peak summer power demand, from 4.1% to 8.5%. When a single category of building goes from four cents on the electricity dollar to eight and a half in two years, it stops being a niche and becomes a load-bearing part of the grid. That is a real estate demand signal dressed up as an energy statistic.

The catch: you cannot build what you cannot power

Here is where the shine comes off, and where the actual opportunity hides. The scarce input is power, and the power system is jammed.

As of the end of 2025, more than 2,060 GW of generation and storage capacity was sitting in US grid interconnection queues, per Lawrence Berkeley National Laboratory. That is roughly double the entire existing US power plant fleet, waiting in line to connect. And the line moves at a crawl. Projects that went operational in 2025 had spent an average of about eight years in queue. Eight years. In technology terms that is multiple hardware generations. You could commit to a data center campus today and not have grid power for it until the GPUs you planned around are museum pieces.

Texas makes the mismatch almost comic. Across the ERCOT grid, 143.5 GW of data centers were seeking to connect as of October 2025. ERCOT's all-time peak demand, set in August 2024, was 85.9 GW. In other words, the data centers lining up want to draw more than 1.6 times the most electricity Texas has ever used at once, on a grid that already sweats through its summers. Not all of those projects are real. That is exactly the point. When demand for power exceeds the entire historical peak of a major grid, the interconnect becomes the asset, and the building attached to it is almost an afterthought.

The supply constraint is not only about generation. It is local politics too. Northern Virginia led every primary market with 1,102 MW of net absorption in 2025, and yet Dominion Energy's power-batching system and Loudoun County's removal of by-right data center zoning are stretching delivery timelines and choking new supply. The tightest, most valuable market in the world is getting harder to build in, not easier. That is bullish for whoever already holds powered capacity there, and brutal for latecomers.

Goldman Sachs estimates that only about 60% of the data center capacity scheduled for the next year will actually come online on time, dropping to roughly 50% over the following two years, mostly because of power delivery. Sit with that. Half the promised supply is likely to slip. Every slipped megawatt is a rent increase for the megawatts that already exist and are already energized.

The economics are industrial in scale, not size

The capital intensity here is on a different planet from warehouses. A standard data center cost roughly $10.7 million per MW to build in 2025. An AI-optimized facility, with dense GPU racks, liquid cooling, and heavier power and backup, runs $15 to $20 million-plus per MW. Do the multiplication on a gigawatt campus and you land at $45 to $55 billion for a single site. That is not a building. That is infrastructure with a nameplate the size of a mid-cap company.

That price tag is why the buyer list is short. This is not an asset class a syndicator assembles from a dozen accredited investors. It is hyperscaler balance sheets and the largest private-capital platforms, and the private side is moving hard. Blackstone committed more than $25 billion to a Pennsylvania digital and energy hub in 2025, a project it expects to catalyze $60 billion more. Its BREIT vehicle put $5.8 billion into pre-leased data center developments in 2025. Its QTS platform carries a portfolio valued at over $70 billion with a development pipeline exceeding $100 billion. When Blackstone builds an energy hub, not just a data center, it is telling you the power is the product.

Where this connects to the equities

If you want exposure through public markets rather than direct development, the pure-play landlords are already printing the results. Digital Realty grew 2025 core FFO per share 10.1% to $7.39 on revenue up about 10% to roughly $6.1 billion, and guided 2026 core FFO to $7.90 to $8.00. Equinix grew 2025 annualized gross bookings 27% to $1.6 billion and guided to 9% to 10% revenue growth in 2026. Those are the income statements of landlords with pricing power and a waiting list.

I want to be clear about the boundary, because it matters. The physical asset class and the stock are two different lenses on the same thing. I have written separately about the REIT and equities angle in the data center REITs piece, and I would rather not repeat it here. The point of this piece is the asset itself: what makes a powered site scarce, why rents behave the way they do, and where the actual bottleneck sits. How you buy a slice of it through a ticker is a downstream question. Get the asset thesis right first.

How I would actually think about it

A few things I keep coming back to.

• Power is the moat, and it is durable. An eight-year interconnection queue is not a problem you outspend in a quarter. Whoever holds energized capacity today has a lead measured in years, and the queue makes that lead widen, not shrink.
• The rent curve tells you where the shortage is worst. Large 3-to-10 MW deployments rose 12.5% while smaller ones rose 6.6%. The biggest AI tenants are paying up the most, which means the scarcity is most acute exactly where the capital wants to go.
• Location still matters, just for a new reason. Northern Virginia at 0.5% vacancy, and getting harder to build in, is the clearest example of a market where the existing owners win by default. Land near cheap, available power and cooperative permitting is the new interstate interchange.
• The supply slippage is a feature for incumbents. If half the promised capacity slips, the megawatts already running get more valuable every quarter the buildout falls behind.
• Watch the tenant, not just the box. The whole thesis rests on hyperscaler capex staying near $600 to $725 billion. That is the load-bearing assumption. Everything else is commentary.

The bottom line

The warehouse trade taught a generation of investors that industrial real estate rewards whoever sees the next structural demand wave early and controls the scarce input feeding it. That lesson still holds. The input just changed. It used to be well-located dirt near consumers. Now it is power delivered to a site, and power is stuck behind a queue twice the size of the entire US generating fleet.

Data center rents rose 6.6% to a record while warehouse rents fell 4.5%. Vacancy hit 1.4% while warehouses sat near 7.4%. Power demand is on track to more than double by 2027. And the biggest, most sophisticated pools of capital on earth are pouring hundreds of billions into it while the grid strains to keep up. That is not a rotation inside industrial real estate. It is a new asset class being born under the old label, and the thing that makes or breaks it is the least glamorous input imaginable: an electrical connection, and how long you have to wait for one.