Data-Center REITs & Colocation
Data Center  Demand vs supply & the price of exposure · unit of demand: leased MW of data-center capacity
EQIXDLR
V2 · factsJun 2026
Sector scan: Data Centers & Infrastructure Group-level demand/supply Updated Jun 2, 2026 · data verified Facts only · no recommendation
Snapshot Product Demand Supply The gap The players The price Deep-dive next Sources

Snapshot — the group at a glance

This group sells the physical buildings where computers live. A data-center REIT (REIT = Real Estate Investment Trust, a landlord company that must pay out most of its taxable profit as dividends and in exchange pays little corporate tax) builds large, secure, climate-controlled, heavily-powered halls and then leases that space — really, the electrical power and cooling inside it — to the companies running AI and cloud workloads (a "workload" = a computing job, e.g. training or running an AI model). The product is measured in megawatts (MW) of leased, powered capacity (a megawatt = a million watts of electrical draw). Buyers are the hyperscalers (very large cloud operators: Amazon AWS, Microsoft Azure, Google Cloud, Oracle, Meta) and the AI labs (OpenAI, Anthropic, xAI). The two big US-listed pure plays (companies whose business is almost entirely this one product) are Equinix (EQIX) and Digital Realty (DLR).

~$60-80B est.
Approx. global colocation / wholesale DC revenue, ~2025 (not live-verified)
~15-25%/yr est.
Approx. capacity growth rate, AI-driven (not live-verified)
2 scan
US-listed pure-play public REITs (EQIX, DLR)
Power scan
The binding supply bottleneck
<3% scan
Vacancy in top-tier markets
+20-40% scan
Rent increase in constrained markets since 2023
Demand for powered data-center space is running ahead of supply: hyperscalers are pre-leasing capacity (signing leases for buildings that do not exist yet) years before it comes online, vacancy in good markets is below 3%, and rents are up 20-40% since 2023. What limits supply is not money or land but electricity — the multi-year wait to connect new sites to the power grid. In money terms, exposure to the two public landlords carries a high multiple of today's earnings, because the market is paying for years of contracted, escalating future rent rather than for current cash. Whether that price is worth paying is left to the reader.

The product & how money is made

The product is a finished, powered, cooled, secured slot in a data center, sold by the megawatt of electrical capacity (and, in the retail/colocation model, by the cabinet or cage — a "cabinet" or "cage" is a rented enclosure for a customer's own servers). The customer brings its own servers and GPUs (GPU = graphics processing unit, the chip used to train and run AI models); the REIT supplies the floor, the power feed, the cooling, the physical security, and the network connections.

Money comes in through long-term leases, typically 10-15 years, priced per kilowatt of committed power, with annual escalators (built-in rent increases, often 2-4% a year) scan. Because the tenant signs up for a decade and pays whether or not it uses every watt, much of the rent is effectively contracted and recurring — close to a take-or-pay arrangement (take-or-pay = you pay for the reserved capacity even if you don't actually consume it). Two sub-models exist:

Owner cash is what's left after the landlord pays to build (capex = capital expenditure, the upfront cost of land, power, and steel) and to operate (power, staff, maintenance). REITs report performance as FFO (Funds From Operations — net income with property depreciation added back, the REIT version of cash earnings) and most of that FFO is paid out to shareholders as dividends.

Source: 500-stocks scan, Data Center REITs section (lease structure, escalators, wholesale vs. retail).

Demand — how much the world will want this

Today (contracted / known): The scan states backlog (signed-but-not-yet-started contracts) at the major DC REITs has grown roughly 3-5x since 2023, vacancy in top-tier markets is below 3%, and every major AI lab (OpenAI, Anthropic, xAI, Google DeepMind) is already a tenant or co-builder scan. Hyperscalers are pre-leasing capacity years in advance — i.e. signing the lease before the building exists — which is a clear signal that buyers expect to be short of space.

Forward (forecast): Given AGI is arriving, demand for compute — and therefore for the powered floor space that compute sits on — rises rather than plateaus. The shift matters in two ways:

Who the buyers are: a short list of deep-pocketed giants — AWS, Azure, Google Cloud, Oracle, Meta — plus the AI labs. This concentration means demand is large and creditworthy, but also that losing or renegotiating one tenant moves the numbers.

✓ VERIFIED — the following figures were confirmed from primary sources after initial publication:

Remaining caveat: some market-size and growth-rate figures not listed above are directional estimates from general knowledge (model cutoff ~early 2026), not live-verified. Company-specific financials in the Players table are from the most recent public filings or earnings. For SEC-verified deep dives on individual companies, see Stock Reports.

Supply — how much can be made, and what limits it

Current capacity & expansion: EQIX operates a global footprint on the order of 250+ data centers across 70+ metros, and DLR operates on the order of 300+ facilities worldwide est. (footprint counts are from general knowledge, not live-verified); both are continuously building. They expand by buying land that already has, or can get, large power entitlements (the legal right to draw a given amount of grid power), then building halls in phases as tenants commit.

The binding bottleneck is electricity, not capital or land. The scan is explicit: supply is "severely" constrained because utility interconnection queues (the wait to get a new site connected to the power grid) run roughly 3-5 years in the key markets (Northern Virginia, Dallas, Phoenix), and land with power entitlements commands premium prices scan. An operator can have the money and the steel and still wait years for the grid to deliver power. This means that a REIT already holding entitled, powered land can build now while a new entrant cannot.

Market-share structure: Among US-listed public pure plays, supply is concentrated in two operators, EQIX and DLR, with the scan also flagging CBRE (development/management) and QTS (owned by Blackstone, limited public float — i.e. few shares trade publicly) scan. The largest builders of all are the hyperscalers themselves, who self-build much of their capacity — so the public REITs supply an important slice, not the whole, of total industry capacity.

Source: 500-stocks scan, Data Center REITs section (interconnection queues, vacancy, rent moves, named players). Footprint counts: general knowledge, not live-verified.

The gap — demand vs supply

On every observable signal in the scan, the product is short (demand exceeds available supply) right now:

SignalWhat it showsReading
Vacancy, top-tier markets<3% scanEffectively full
Rent change since 2023+20-40% scanRising fast = shortage
Backlog growth since 2023~3-5x scanFuture demand already locked
Pre-leasingYears ahead scanBuyers can't wait
Power lead time~3-5 yrs scanSupply can't catch up quickly

The arithmetic of the gap: demand is being added in months (a new AI model, a new fleet), but supply is added in years (the ~3-5 year power queue). As long as that mismatch holds — and on the AGI premise compute demand keeps climbing — the product stays short and pricing power stays with the landlord. (The persistence of the gap is a forecast, not a contracted fact.)

When could it flip to oversupply? A few forecast scenarios, none currently visible in the data: (1) grid power gets unblocked en masse — large new generation plus faster interconnection; (2) hyperscalers over-build their own capacity and stop leasing from third parties; (3) AI compute demand grows far slower than expected, or efficiency gains cut compute-per-task sharply; or (4) so much capacity is started in 2025-2026 that it all energizes together around 2028-2030 and lands at once. These are scenarios to watch, not present conditions.

The players — who captures the money

TickerWhat it makesExposure to this productRough sizePosition / edge
EQIXRetail/colocation data centers; interconnectionPure play (~100% of revenue)~$80-90B mkt cap est.Global interconnection leader; recurring cross-connect revenue; 70+ metros
DLRWholesale/hyperscale + colocation data centersPure play (~100% of revenue)~$45-55B mkt cap est.Large hyperscale landlord by capacity; land/power bank; global platform reach
CBRECommercial real-estate services; DC dev/mgmtDiversified (DC is a small slice)~$40B+ mkt cap est.Builds/manages DCs for others; a services provider, not a landlord
QTSHyperscale data centersPure play, but privateBlackstone-owned scanFast-growing hyperscale builder; limited public float (few shares trade publicly)

Note: the biggest "players" in raw capacity are the hyperscalers (AMZN, MSFT, GOOGL, ORCL, META), who self-build — but they are buyers here, not pure data-center landlords, and are covered in the Cloud Hyperscalers group. Among publicly investable pure plays, EQIX and DLR are the bridge to a company-level deep-dive.

Source: 500-stocks scan, Data Center REITs ticker list (EQIX, DLR, CBRE, QTS). Market caps: general knowledge, approximate, not live-verified.

The price of exposure

Plain money-in / money-out shape: this is a high-capex, capital-heavy business. The owner spends large sums upfront per megawatt, then collects contracted rent for 10-15 years. So owner cash arrives slowly and steadily; the value sits in the stack of future escalating rents rather than in this year's free cash.

What an owner pays today (all multiples below are general-knowledge approximations, est., not live-verified):

Neutral arithmetic, no verdict: the market is paying a multiple above current cash because supply is short, leases are long, and rents are rising — i.e. the buyer is paying today for a forecast of years of contracted, escalating future rent, not for current free cash. Whether that price is worth paying depends on whether the demand-over-supply gap persists, which the reader must judge.

What to deep-dive next

This is a factual pointer to where company-level analysis adds the most signal, not a recommendation.

Sources & confidence

Source: /Users/ravf/projects/work/.claude/worktrees/sector-hub/research/investments/500-stocks/04-data-centers-infrastructure.html (Data Center REITs section); general knowledge to ~early 2026 for market-size, share, and valuation-multiple figures, which are approximate and not live-verified. No prior report existed for this group.