Data Centre REITs and the Power Constraint

The Slow Part Is the Wire, Not the Building

A data centre is a shed full of racks, and sheds are quick. What is not quick is the electricity: a queue position with the transmission operator, a substation, transformers on multi-year order books, and often new lines across land somebody else owns. Buying the site and putting up the shell runs in quarters. Getting it energised runs in years, and the second timetable does not care about the first.

That asymmetry is the whole story for a data centre REIT. Leases can be signed faster than power can be delivered, so the growth you can see in the filings sits ahead of the growth you can bank. A bookings headline is a promise; an energisation date is a cash flow.

Why a National Demand Figure Tells You Nothing

Every article on this subject opens with national electricity demand climbing after years of flat consumption, and the trend is real. It is also close to useless for judging a REIT, because electricity constraints are not national. They are a property of one substation, one queue, one utility’s capital plan.

Two campuses in the same country, both with planning consent and both fully let, can be years apart on power. What separates them is their position in an interconnection queue, and how much reinforcement the local grid operator has to build before that position clears. So the question to ask about a development pipeline is not how fast demand is growing. It is how much of the pipeline holds a signed interconnection agreement with a date attached, and how much is land with a hope.

Bookings, Megawatts and What Neither Means

Digital Realty (DLR) reported FY2025 bookings of $1.2B in annualised rent, signed at 100% share. That is rent contracted, not rent collected. FY2025 AFFO per share was $6.55; Core FFO of $7.39/sh is the headline earnings metric management leads with. Market capitalisation was roughly $64.4B in June 2026.

Equinix (EQIX) ran 280 data centres across 77 metros and 36 countries at 31 December 2025, with about 3 GW of developable capacity on land it controls. FY2025 revenue was $9,217M, AFFO $38.33/sh, and average MRR churn (monthly recurring revenue lost to departing customers) 2.4% a quarter. Market capitalisation was roughly $102.9B in June 2026.

One gloss before going further, because it is where capacity figures get overstated. IT load is the power reaching the servers. Facility load is what the building pulls from the grid, and it is larger, because cooling and power conversion take their cut. The ratio between the two is PUE, power usage effectiveness: a site at 1.3 draws 1.3 MW from the grid for every 1 MW that reaches the racks, and lower is better. Leases are generally written on IT load while grid connections are sized on the facility total, so the megawatts a REIT sells and the megawatts a utility has to supply are different numbers. Treating them as one overstates how much sellable capacity a given connection buys.

Company	Growth leading indicator	Supply-side indicator
DLR	$1.2B FY2025 bookings (100% share)	Megawatts under construction, with energisation dates
EQIX	Interconnection and colocation demand	~3 GW developable land under control

Neither column is billing capacity. Bookings are rent that has not started; developable capacity describes what land and secured power might one day support, which is a different claim from a connection agreement in hand.

The Lag Between Signature and Rent

Rent on a signed data-centre lease typically starts 12 to 24 months after signature. That is a working rule of thumb drawn from how these companies talk about their backlogs, not a disclosed statistic, so treat it as a shape rather than a schedule.

The lag is physical, not administrative: permitting, construction, fibre, commissioning, and the grid connection. Leases signed in one year may not fully show up in the next year’s AFFO if the connection slips, and bookings can run well ahead of AFFO when tenants pre-let years of future capacity. Neither gap is a warning sign on its own. It becomes one when bookings keep climbing and the energised megawatt count does not.

Worked Example: Sizing a Bookings Number

Take Digital Realty’s FY2025 bookings and put them in per-share terms, which is the only form in which a growth number can be compared to earnings.

Step	Value	How
FY2025 bookings signed	$1.2B annualised rent	Filed, 100% share
Assume half energises in the first wave	$600M	Half of $1.2B
Shares outstanding	~355M	$64.4B ÷ $181.41
First-wave rent per share	~$1.69	$600M ÷ 355M
FY2025 AFFO per share	$6.55	Filed

Set beside $6.55 of AFFO, $1.69 of incoming annualised rent looks transformational. It is not, and understanding why is the point of the exercise. Rent is a top line. Out of it come the site’s operating costs, of which power is much the largest, then interest on the capital that built it, then maintenance capex. What survives to AFFO is a fraction of the $1.69, and the 100% share basis includes capacity in joint ventures that Digital Realty does not wholly own. A bookings figure tells you the direction of travel. It does not tell you the earnings.

Nor does it tell you the date. Half of that $600M arriving a year late is not a lost lease, it is a deferred one, and the gap between those two readings is most of the argument over what these shares are worth.

Why a Demand Wave Shows Up in Leases First

Large cloud and AI tenants take capacity years before they have machines to put in it, because the alternative is queueing behind whoever booked it first. Their commitment therefore lands in a REIT’s bookings line long before it lands in same-store revenue. That is the mechanism behind every gap you will see between a bookings chart pointing sharply up and an AFFO line that is not.

Equinix compounds mainly through interconnection density and low customer churn; Digital Realty through large hyperscale lettings. The two businesses meet at the same substation. See colocation vs hyperscale for the split.

Scarcity Is Not the Same Thing as a Moat

The tempting conclusion is that constrained power hands incumbents pricing power, so the multiple takes care of itself. Half of that is right. Where an operator has capacity already energised, scarcity does let it charge more, and a tenant who needs 30 MW this year cannot wait for a cheaper landlord.

The other half is that the constraint prices the operator’s own inputs too. When a grid is short, connection charges, network reinforcement costs and the electricity itself all rise, and utilities recover those costs from the people connecting. High rents on live megawatts sit alongside a higher cost for every future megawatt, and the second effect grows as the first one attracts capital. Scarcity also does not last: a shortage this visible pulls in generation, storage and transmission spending, and behind-the-meter supply from tenants themselves.

So the durable asset is narrower than “exposure to a power shortage”. It is secured, energised capacity, plus queue positions and connection agreements that competitors would now have to wait years to replicate. A land bank without an interconnection agreement is not that, however many gigawatts it is described in.

What the Multiple Is Paying For

Data-centre REITs trade at premium multiples largely because the growth is long-dated. Getting the comparison right needs one piece of care, because the two companies headline different earnings measures.

Company	Price (10 Jun 2026)	On AFFO	On the headline metric
EQIX	$1,043.60	~27.2× ($38.33/sh)	Same, AFFO is the headline
DLR	$181.41	~27.7× ($6.55/sh)	~24.6× Core FFO ($7.39/sh)

Comparing Equinix’s 27.2× P/AFFO to Digital Realty’s 24.6× P/Core FFO makes Digital Realty look two and a half turns cheaper. On the same AFFO basis it is fractionally more expensive. The basis switch moves the number by more than the gap between the two companies, which is worth remembering before ranking anything on a multiple somebody else has calculated. The ~22–28× range often quoted for the sector is a rough guide from a handful of large names, not a valuation rule.

Two things can then go wrong. A multiple this dependent on distant cash flows compresses when long rates rise, and our planning deck uses a 4.50% 10-year Treasury as the conservative case. The second risk is specific to this sector: even with the multiple unchanged, AFFO growth disappoints when connections slip. A bookings number with no energisation date behind it carries that risk whatever its headline size.

What to Track on Earnings Calls

Pair every demand disclosure with a supply one:

• Bookings or new leasing: dollar volume, and what management says about the backlog behind it.
• Megawatts under construction, with energisation dates, from the development tables in the supplement. This is the number that converts.
• How much of the pipeline holds a signed interconnection agreement. Developable capacity without one is land.
• Whether capacity figures are quoted on IT load or facility load, since the two differ by the PUE.
• Commentary on connection charges and power prices, which show whether scarcity is helping the rent line faster than it is hurting the cost line.

The Digital Realty and Equinix profiles carry the FY2025 operating detail. Churn and stickiness for colocation customers sit in the churn guide.