The Landlord of the AI Boom Owns Buildings You've Never Seen
The company you have probably never heard of owns the buildings your favorite apps actually live in. Here is how a REIT born from bankruptcy auctions became the landlord of the AI boom.
Almost nobody outside the industry can name it, yet a large share of the modern internet has a return address inside one of its buildings. Digital Realty Trust rents the floors, the power and the private cables that cloud platforms, banks, hospitals and streaming services rely on. It does not write the apps or design the chips. It owns and operates the roughly 300 data centers - across more than 25 countries and six continents - where all of that work physically happens.
The framing matters because the cloud is often described as if it were weightless. It is not. Every search, transaction and AI prompt runs on a machine that sits in a rack, in a room, in a building, drawing real electricity and throwing off real heat. Digital Realty's entire business is the ownership and operation of those buildings, leased back to the companies that fill them with servers. In legal terms it is a real estate investment trust, traded on the New York Stock Exchange under the ticker DLR and a member of the S&P 500. In practical terms it is one of the most strategically located companies in technology.
01From bankruptcy auctions to a global platform
The origin is unglamorous, which is part of the appeal. In the early 2000s the private equity firm GI Partners assembled a portfolio of data center properties, much of it bought cheaply from distressed sellers in the wake of the dot-com bust. In November 2004 that portfolio was spun out through an initial public offering as Digital Realty Trust, one of the first REITs focused exclusively on data centers. It began life with a couple of dozen buildings. Two decades later the count is roughly 300.
The founding cast came from finance, not from software. Michael Foust, who led the company from its inception, and Richard Magnuson of GI Partners built the early portfolio on a contrarian read: that distressed data center real estate was worth far more than the market had priced it, provided someone operated it properly. That instinct - buy the hard physical asset, run it well, collect the rent - has stayed remarkably constant even as the tenants changed from dot-com survivors to AI hyperscalers.
Growth on that scale did not happen by building alone. Digital Realty spent a decade buying its way onto new continents: interconnection specialist Telx in 2015, wholesale operator DuPont Fabros in 2017, Latin American builder Ascenty in 2018, and - the largest of them - the roughly $8.4 billion acquisition of Europe's Interxion, completed in 2020. Each deal added metros, customers and network density that would have taken years to grow organically.
- 2004IPO on the NYSE. Spun out of GI Partners with ~21 buildings.
- 2015Telx. ~$1.9B - interconnection and colocation depth.
- 2017DuPont Fabros. Large-scale wholesale for hyperscalers.
- 2018Ascenty. Entry into Brazil and Latin America.
- 2020Interxion. ~$8.4B - a leading EMEA operator overnight.
02The two words the whole pitch rests on
Digital Realty's strategy is built around a phrase it helped popularize: "data gravity." The idea is that large data sets behave a little like mass. The more data a company accumulates in one place, the more applications, services and additional data are pulled toward it, because it is slow, expensive and risky to move data around. Follow that logic and a conclusion appears: instead of hauling data to the applications, put the applications - and the cloud connections - next to where the data already sits.
That conclusion is, conveniently, an argument for renting space from Digital Realty. The company turned the concept into a research tool, the Data Gravity Index, which attempts to quantify the "gravitational" intensity of enterprise data across global metros, and into a solution methodology it calls Pervasive Datacenter Architecture, or PDx. Underneath the branding is a genuinely useful observation about how distributed infrastructure actually behaves.
03What customers actually buy
Strip away the vocabulary and Digital Realty sells three things that every server on earth needs: space, power and connections. It packages them through PlatformDIGITAL, its global platform brand, which frames the network of facilities as a single meeting place where a customer can deploy infrastructure once and reach everything else.
At one end of the range is colocation - a company renting anything from a single rack to a private cage inside a shared facility. At the other end is scale, or "hyperscale," where a cloud provider takes an entire building or a purpose-built campus measured in megawatts. In between sit powered base buildings (a shell with power that the customer fits out themselves) and turnkey space that is ready to occupy. The highest-margin layer is interconnection: the cross connects, metro links and cloud on-ramps that let a customer's gear talk privately to thousands of networks and to AWS, Microsoft Azure, Google Cloud and Oracle without touching the public internet.
04Who rents the space
The customer list runs past 5,000 and splits into three broad camps. First, the hyperscale cloud and content providers, who take the largest footprints and are, in a neat twist, both tenants and interconnection partners - their cloud on-ramps live inside Digital Realty's buildings. Second, the networks and carriers that make the interconnection ecosystem worth joining. Third, the enterprises: banks, insurers, hospitals, manufacturers, media companies and government agencies that need their own infrastructure to sit close to the clouds they depend on. A single campus can hold a hyperscaler in one hall and a Fortune 500 bank in the next.
05The business model, in plain numbers
As a REIT, Digital Realty makes money the way a landlord does: recurring rent for space and power, plus fees for interconnection, under contracts that run for years. Because it is a REIT it must distribute most of its taxable income to shareholders as dividends, which shapes how it funds growth. Rather than putting every new campus on its own balance sheet, it increasingly builds through joint ventures with large infrastructure investors, keeping the model capital-light while still expanding the footprint.
The headline financials sit in the billions: total revenue in the mid-single-digit billions of dollars a year, and a market capitalization in the tens of billions. But for a data center REIT the number investors watch most is not revenue - it is Core FFO per share, the cash-flow measure above, which grew about 10% in 2025 with management guiding to further growth in 2026.
06Where it fits, and who it competes with
Digital Realty sits at the infrastructure layer of the technology stack, beneath the clouds and applications that get the attention. Its closest rival is Equinix, which pioneered the interconnection-heavy colocation model; the two are often described as the twin giants of digital infrastructure, with Digital Realty historically weighted toward larger, wholesale and hyperscale space and Equinix toward dense interconnection. Beyond that pair the field includes NTT Global Data Centers, CyrusOne, Vantage, QTS (owned by Blackstone), Iron Mountain and CoreSite (owned by American Tower).
The differentiator Digital Realty leans on is the combination of global scale and connectivity under one roof: a customer can standardize on a single platform across dozens of metros, then use interconnection to knit those locations together and reach the major clouds privately. Its distinctive assets are less about any one building and more about the network of them and the exchange that connects them.
07The AI wave, and the power bill
The current moment is a favorable one for a company that owns buildings full of power and cooling. Training and running AI models demands dense, energy-hungry racks, and the hyperscalers are competing to secure capacity years in advance. Digital Realty has responded with a development pipeline it puts at roughly 5 gigawatts of buildable capacity and, through 2025, announced multi-phase campuses topping 250 megawatts each in Dallas, Tokyo and Frankfurt, backed by infrastructure-fund partners. Its chief executive, Andy Power - who ran the company's finances as CFO and then president before taking over in December 2022 - has spent recent quarters arguing that demand, not oversupply, is the binding constraint.
That growth carries an obvious cost in electricity, and the company has tied itself to sustainability commitments to match: it was an early signatory of the Climate Neutral Data Center Pact and has adopted science-based emissions targets. For a business whose product is, at bottom, well-run power and cooling at scale, energy strategy is not a side project. It is the operating model.
What ties it all together is a specific kind of expertise that rarely makes headlines: the operational discipline of keeping hundreds of buildings running without interruption. Power redundancy, cooling design, physical security, capacity planning and the logistics of moving megawatts onto a grid are the daily craft here. The company employs several thousand people to manage that craft across time zones. It is an unglamorous skill set, and it is precisely the one the AI era needs most.