Colocation vs Hyperscale: Two Data Centre Economics

Two Growth Models Under One Sector Label

“Data centre REIT” covers two businesses that differ in one thing above all: who has the upper hand when the contract comes up for renewal.

Retail colocation packs hundreds of customers into shared halls and sells them space, power, and above all connections to each other. Once a customer’s traffic runs through a dozen cross-connects to its suppliers and partners in the same building, moving out means rebuilding that web somewhere else. Contracts are short, and the landlord is the one who reprices them.

Hyperscale is the opposite trade. A whole hall, sometimes a whole campus, is pre-let to one very large tenant on a long contract, often years before the building is energised. The revenue is contracted and the counterparty is creditworthy, which reads as safety, and for the length of the term it is. But only a handful of companies in the world buy capacity at that scale, each of them can build its own, and each knows roughly what the landlord spent.

Equinix and Digital Realty are the clearest US-listed examples of the two poles, though neither is pure: Digital Realty sells colocation as well. They report on different earnings bases and headline different growth indicators, so translate the basis before screening either on the other’s numbers.

Equinix: Interconnection and MRR Compounding

Equinix’s retail footprint and interconnection mix show up in low churn by colocation standards and an AFFO-led earnings screen (footprint as of 31 December 2025). FY2025 metrics are in the table.

MRR is monthly recurring revenue, the contracted rent running through the estate each month. The churn figure is the share of it lost per quarter, and the word quarter does the work. Left at 2.4% a quarter, churn takes out roughly 9% of the base over a year, so it cannot be set beside a tower REIT’s 1–2% annual churn and read as a similar number. The churn guide sets out how far apart the two definitions sit.

What holds the churn down is interconnection. Revenue grows when customers cross-connect to each other inside an Equinix building, and every cable added raises what it would cost to leave. Growth compounds as existing cabinets take more cross-connects and new customers arrive to reach the ones already there.

Digital Realty: Bookings and Hyperscale Scale

Digital Realty leads on bookings for growth and Core FFO for earnings, with AFFO filed in the reconciliation behind it. FY2025 metrics are in the table.

Hyperscale growth is visible in the bookings line before it fully converts to AFFO. Power and construction timelines create a 12–24 month recognition lag (analyst convention drawn from backlog commentary, not a filed statistic). See data centre power constraint for why bookings can run ahead of energised capacity.

Bookings are also where the concentration sits. A hyperscale hall has one tenant, so at building level the concentration is total, and at portfolio level it depends on how many of the world’s handful of hyperscalers a landlord has signed. The disclosures to read are the top-customer table and the lease expiry schedule in the 10-K: they show how much rent renews in one negotiation, which is the risk a bookings headline hides.

Side-by-Side: How the Two Models Differ

Read the earnings-screen row carefully, because the two lines are not the same units. Equinix leads with AFFO. Digital Realty leads with Core FFO, which stops short of the AFFO deductions and runs $0.84 a share above its filed AFFO. Any multiple struck on the two headlines inherits that gap, and the illustrative data-centre band of roughly 22–28× is an AFFO band. The worked example below shows how large the distortion can be.

Worked Example: Normalising the Earnings Basis

An investor comparing “earnings yield” across two data centre REITs must match numerator and denominator. Take two illustrative companies at the same $100 share price, one that leads with AFFO and one that leads with Core FFO:

Matched on AFFO, the two earnings yields are almost identical: 3.55% against 3.6%. The Core FFO screen prints 4.0% and makes Company B look the cheaper of the pair, purely because Core FFO stops before the deductions AFFO makes. The gap between Company B’s two lines, about 11%, is the same size as the gap between Digital Realty’s filed Core FFO of $7.39 and AFFO of $6.55.

One caution before leaning on any of it. Nareit defines FFO, which is what makes P/FFO comparable across the sector. It has never defined AFFO, and says plainly that there is no consensus on what the adjustments should be. Each company draws the line between maintenance capital and growth capital itself, and growth capital never reaches AFFO. A data centre is where that line is hardest to hold: a hall stripped out and re-powered for denser racks can be booked as sustaining an existing asset or as building a better one, and the two answers give different AFFO from identical spending. So read each reconciliation before ranking on the multiple. Our FFO vs AFFO guide works through where the discretion sits.

Tower REITs trade in a lower band, roughly 15–22× AFFO. Part of that gap is simply that a steel mast costs almost nothing to sustain, so AFFO there is a closer proxy for spendable cash than it is here; AFFO for towers makes the comparison.

Contract Duration and Who Holds the Renewal

Colocation contracts are short, often a year or two, which sounds like the weaker position until you ask what a customer would have to do to leave. Rebuilding a dozen cross-connects in another building is expensive, slow, and risky for whatever runs across them. Short contracts plus high switching costs mean the landlord reprices often, and upward.

Hyperscale contracts run long, ten to fifteen years is common, and are signed before construction starts. That takes lease-up risk out of the development, which is worth a great deal. What it does not buy is pricing power. The tenant fills the hall, its alternatives include building for itself, and it can see roughly what the landlord spent. The safety lives in the contracted term and the risk in the renewal, which arrives in one piece.

Margins follow from that. Colocation carries the cost of serving many customers: sales, support, staff on site to plug things in. Hyperscale earns less per megawatt but is cheap to run once let, and its exposure sits in construction timing and power.

Equinix’s 392,300 cabinet capacity, of which 299,300 were billing at the end of 2025, is a utilisation read on a retail footprint: roughly three quarters full, with the rest saleable without building anything. Digital Realty’s $1.2B of bookings is the mirror image, capacity committed before it exists.

Valuation: Premium for Density vs Premium for Backlog

Density and backlog are different things to pay for. Interconnection lets colocation revenue reprice upward with little execution attached to it; a hyperscale backlog still has to be built, energised and, eventually, renewed.

Which raises the question a multiple cannot answer. A data centre REIT is a developer as much as a landlord, and what decides whether development creates value is the yield on cost, the stabilised net operating income divided by everything spent to reach it, set against the cost of the capital funding it. Build at 8% against capital costing 6% and each megawatt adds value. Build at 6.5% against the same 6% and the company is growing its asset base and its share count without leaving its owners better off. Bookings tell you the volume. Only the yield on cost tells you whether the volume was worth having.

The binding constraint on that volume is neither land nor demand. It is power, and more precisely the lead time to get a grid connection, which is measured in years and set by a utility. That is why signed leases can sit well ahead of billing capacity; power availability takes it further.

Long-duration data-centre cash flows are also rate-sensitive, because most of the value sits in cash a long way out. Planning 10-year Treasury 4.50% (10 June 2026 deck) is the conservative stress for multiples in the 22–28× band.

Two things to do before ranking Equinix against Digital Realty. Put both on AFFO, since the headline multiples are not in the same units. Then look past the contracted revenue to the renewal, because that is where the two models stop resembling each other.