Digital Realty’s $1.9B Data Center Backlog: AI’s Power Trade
AI is not making landlords obsolete. It is making the right landlords frighteningly powerful: $1.9 billion of annualised rent is already queued behind power, permits and 1.4 gigawatts of construction.
AI is not making landlords obsolete. It is making the right landlords frighteningly powerful.
Digital Realty finished June with a $1.9 billion annualised-rent backlog at 100% share, after signing $307 million of bookings in the second quarter and another $410 million across two hyperscale leases in July. That is not a cute technology trend. It is real estate income queued up behind concrete, cables, transformers and planning approvals.
And it is a very different game from buying a unit, putting in grey vinyl flooring, and calling yourself a property investor.
The core story: Digital Realty has sold tomorrow’s capacity
Digital Realty, listed on the NYSE as DLR, reported its second-quarter 2026 results on July 23. The headline number that matters is the backlog: $1.9 billion of annualised GAAP base rent at 100% share, or $1.4 billion at Digital Realty’s share.
That backlog exists because the company has already signed leases that have not fully commenced. In plain English: customers have committed to pay for capacity that still needs to be delivered.
The second quarter produced $307 million of annualised GAAP base-rent bookings at 100% share. Of that, $108 million came from Digital Realty’s smaller 0–1 megawatt and interconnection business. The bigger July signal was two hyperscale leases worth $410 million of annualised base rent at 100% share.
This is what people miss when they say AI is just software. The software sits somewhere. It needs enormous amounts of electricity, cooling, network connectivity and land in places where customers can actually use it.
Digital Realty had 310 data centres, about 3.1 gigawatts of in-place IT capacity and around 8.5 gigawatts of buildable capacity under active development or held for future development as of June 30. It also had roughly 1.4 gigawatts under construction.
That is property. But it is property with a moat most residential investors will never touch: power procurement, engineering expertise, supply-chain relationships, customer contracts, land banks and years of local approvals.
The September investor presentation makes the point bluntly: the company is positioning itself around AI workloads that require data, compute and connectivity to sit close together. Its future development capacity is measured not in apartments or square feet, but in gigawatts.
Why this is a proper real-estate story, not a tech-stock story
The market loves pretending every AI winner is a chip company. That is lazy thinking.
Nvidia sells the picks and shovels. Digital Realty sells the serviced land, buildings and power environment in which those picks and shovels can earn money. One is hardware. The other is a toll road with a very expensive on-ramp.
Digital Realty’s revenue reached $1.9 billion in the second quarter, up 29% from the same quarter a year earlier. Adjusted EBITDA was $978 million, up 19% year on year. Core FFO excluding net promote was $2.13 per share, while management raised its 2026 outlook for that measure to $8.15 to $8.20 per share.
More important than any one quarter, the company reported 25.4% cash rental-rate increases on renewal leases in the second quarter. That is what constrained supply looks like when demand arrives with a chequebook and a deadline.
A normal commercial landlord worries about filling vacant space. An AI-ready data-centre owner worries about whether it can get enough electricity and skilled labour to deliver contracted capacity on time.
That is a far better problem. It is still a problem.
The constraint is not demand. It is power.
The overlooked part of the AI property boom is that a data centre is not simply a warehouse with servers inside. The building is almost the easy bit.
The actual bottlenecks are grid connections, substations, transformers, water or cooling design, construction labour, equipment lead times and community or regulatory pushback. At the Bank of America Media, Communications & Entertainment Conference on September 9, Digital Realty flagged labour availability and growing resistance to data-centre development as operational risks.
This is why the best asset is not necessarily the cheapest acreage on the map. It is acreage with credible, deliverable power and a pathway through local government.
Digital Realty’s portfolio shows the value of location. Northern Virginia alone had 842 megawatts of IT load and 98.6% occupancy at June 30. Across the Americas, portfolio occupancy was 95.7%. You do not create that position by finding a cheap block on the outskirts and lodging a development application on Tuesday.
You build it over years, then spend more years making sure the grid, contractors and customers can all move at the same speed.
For founders, this should sound familiar. Distribution beats product quality when the product is merely good enough. In data centres, power access is distribution. If you own it, everyone else is pitching from the queue.
The second-order implication: AI will split property into winners and wallpaper
For years, investors were taught that property was property. Buy a decent building, in a decent suburb, with decent tenants, and let time do the heavy lifting.
That is no longer enough.
The AI build-out is accelerating a divide already underway between strategic real estate and generic real estate. Strategic property has a hard-to-replicate reason for existing: power, logistics access, zoning scarcity, connectivity, specialised build quality or proximity to an economic engine. Generic property has a brochure and a comparable sale.
Data centres sit at one extreme of that spectrum. They are purpose-built, capital-intensive and operationally demanding. That makes them difficult to own directly, but it also makes the best assets difficult to copy.
This is the real lesson for REIT investors. Stop treating the sector as one big dividend bucket. Office, regional retail, self-storage, apartments, logistics, towers and data centres do not share the same economics just because a database labels them “real estate.”
A listed REIT is a business with assets, leverage, management, development risk and tenant concentration. The asset class is not the investment thesis. The cash flow is.
Digital Realty had $18.6 billion of total debt at June 30, alongside a 4.7-times net-debt-to-adjusted-EBITDA ratio. That is manageable only if lease commencements, development execution and capital access remain strong. A backlog is valuable, but it is not cash in the bank until the facilities are delivered and tenants turn on.
The contrarian angle: do not confuse a brilliant theme with a cheap investment
Here is where investors get themselves into trouble: they discover an obvious structural winner, then pay any price for it.
Data-centre demand can be real. Digital Realty’s leasing numbers can be excellent. AI can require more infrastructure than most people appreciate. All three things can be true while a specific share price still offers a mediocre return.
The risk is not that AI disappears next week. The risk is execution and capital intensity.
Data-centre owners need to keep spending before they collect the rent. Digital Realty expects to put billions into development. Its future returns depend on delivering power, buildings and customer capacity on schedule, without construction costs exploding or local approvals stalling.
Then there is tenant concentration. Hyperscale customers are financially powerful, but they are also sophisticated buyers. They negotiate hard, build themselves, shift workloads and can change their own capital-spending plans. A long lease from a giant customer is attractive. Betting your entire portfolio on one theme is still how people get carried out.
So the contrarian conclusion is simple: the AI-property trade is not “buy every company with data centre in the presentation.” It is “find operators with real capacity, contracted demand, financial discipline and a credible route to power.”
That is harder. Good. Investing is supposed to be hard before it becomes rewarding.
What this means for you
If you are a saver or retail investor, do three things tomorrow.
First, separate your home from your investment portfolio. Your house may be useful, enjoyable and financially sensible. It is not a diversified real-estate strategy. Treat it accordingly.
Second, when you look at a REIT, read five numbers before you get excited by the dividend: occupancy, lease-expiry profile, development pipeline, debt maturity schedule and the gap between signed backlog and operating cash flow. If management cannot explain those in plain English, move on.
Third, hunt for constraints, not narratives. “AI” is a narrative. A permitted site with power, a signed customer, a viable build plan and a decent balance sheet is a constraint-backed business.
For operators, the lesson is even cleaner. The businesses that win big are often boring at the point where the bottleneck lives. In this case, the bottleneck is not a chatbot. It is megawatts.
I have made money by respecting boring infrastructure and lost money by getting seduced by exciting stories without checking who controlled the choke point. The story gets attention. The bottleneck gets paid.
Digital Realty’s $1.9 billion backlog is a reminder that the next property boom is not necessarily visible from the street. It may sit behind security gates, humming away beside a substation.
That does not make it risk-free. It makes it worth understanding properly before everyone else decides it was obvious.