If you value commercial real estate for a living, you've noticed: every broker pitch deck now has a data center slide, every landowner with 50 acres near a substation thinks they're sitting on a gold mine, and half the industrial developers I know are "pivoting to digital infrastructure." So I spent the weekend reading the actual research — JLL's 2026 global outlook, the Uptime Institute's operator survey, the IEA's energy-and-AI report, a DOE lab study, the CRE Finance Council's primer, and six more — and here's the honest version.
The numbers are almost hard to believe
JLL expects nearly 100 gigawatts of new data center capacity between 2026 and 2030 — roughly doubling the entire planet's existing capacity — growing at about 14% a year. They put the price tag at up to $3 trillion by 2030, with $1.2 trillion of that showing up as real estate asset value. Tenants will spend another $1–2 trillion just fitting out the buildings with servers.
The hyperscalers are spending like it: the top five put ~$443 billion of capex into the ground in 2025, up 73% in a year, and they're projected to hit ~$602 billion in 2026 — three-quarters of it aimed at AI infrastructure. (CRE Finance Council)
Power is the whole ballgame
Here's the number that matters for anyone touching dirt: U.S. data centers went from 60 terawatt-hours a year in the mid-2010s to 176 TWh in 2023 — 4.4% of all American electricity — and the DOE's Berkeley lab projects 325 to 580 TWh by 2028. Globally, the IEA has data centers at ~415 TWh in 2024, more than doubling by 2030. One large facility eats as much power as 100,000 homes; the biggest ones under construction match two million homes.
Translation for appraisers: the old location mantra is dead. A site's value now starts with a single question — can you get 100+ megawatts to it, and when? The IEA flags grid-connection delays as a live risk, and operators are increasingly designing for locational flexibility: go where the electrons are. Land near a constrained substation with a five-year interconnection queue is a different animal than land next to surplus generation. Underwrite the power first, the dirt second.
The buildings themselves are changing
Two shifts worth knowing. First, AI workloads were about a quarter of data center demand in 2025, mostly training; JLL expects inference to overtake training in 2027 and AI to be half of all workloads by 2030. Inference facilities look different — more distributed, closer to users, different redundancy math. Second, rack densities keep creeping up (10–30 kW is the new normal band; almost nobody's past 30 kW yet), liquid cooling is moving from exotic to expected, and efficiency gains have flatlined — average PUE hasn't budged in six years. (Uptime Institute)
For valuation, that's obsolescence risk in plain English. A 2018-vintage enterprise data center and a 2026 AI-ready facility are not the same asset class anymore, and the market is starting to price them that way. Check the cooling, check the floor loading, check the power density per rack — the building's birthday matters more than it used to.
The pushback is real
Three of the reports I read are basically warning labels. Local governments are writing ordinances specifically targeting data centers — noise, water use, and the oldest complaint in the book: "why are my electric bills going up?" An E3 study digs into data centers' role in rising retail electricity rates, which is the political fuel for all of it. And at the federal level, the Congressional Research Service is tracking policy that could swing either way.
If you're appraising entitled land for data center use, entitlement risk isn't theoretical anymore. A site that's "perfect except the county just passed a data center moratorium" carries a real, quantifiable discount. Talk to planning staff before you believe the broker's "shovel-ready" claim.
What could go wrong
The most interesting warning came from AlixPartners' survey of 400 industry insiders: the AI boom is real, but the silicon architecture underneath it is about to change, and they expect disruption and M&A across the industry. Translation: today's winners might not be tomorrow's tenants. The Uptime data backs the caution — two-thirds of operators can't find or keep enough qualified staff, and one in ten outages still causes serious disruption.
Nobody's calling for a bust. But "double the world's capacity in four years" is the kind of sentence that has preceded every commercial real estate correction I've ever appraised through.
The bottom line
Data centers are a real, scaled property type now — securitized, financed, and traded like one (~$57 billion in U.S. securitizations since 2021). The demand story is the strongest of any CRE sector. But the value isn't in the dirt, it's in the megawatts; the entitlement isn't guaranteed; and the buildings are evolving fast enough to strand yesterday's designs.
If someone brings you a data center assignment, start with three questions: where do the electrons come from, what does the county actually allow, and was this building designed for the workloads of 2026 or 2016? The reports agree on the growth. They just don't agree it'll be smooth.
Sources: JLL 2026 Global Data Center Outlook; Uptime Institute Global Data Center Survey 2025; IEA Energy and AI (2025); Lawrence Berkeley National Lab 2024 U.S. Data Center Energy Usage Report; CRE Finance Council Data Center E-Primer (Jan 2026); AFCOM State of the Data Center 2026; AlixPartners 2026 Data Center Market Outlook; CRS R48762; E3 2026 (electricity rates); GCC Local Ordinances 2025; IBISWorld Hyperscale Data Center Services in the US (Mar 2026).