The buyer's question is no longer whether artificial intelligence demand is real. The question is whether a scarce block of powered, cooled, connected capacity is cheaper through a global data-centre REIT campus than through a self-build program that may not secure power, transformers, permits, fibre diversity or operating staff on time. A cloud buyer looking at a 2026 deployment has to price more than steel and concrete. The buyer has to price the option value of committed utility power, the risk of late energisation, the liquidity cost of tying up billions of dollars before a server is installed, the network value of sitting near carriers and cloud on-ramps, and the probability that today's training-heavy workload mix turns into a more distributed inference estate.
Digital Realty is interesting because it sits exactly at that negotiation point. It is a public REIT, not a chip company, yet the AI infrastructure cycle is turning its most basic product, a megawatt of reliable data-centre capacity, into a scarce financial instrument. Its pitch is not simply that it owns buildings. Its pitch is that it can assemble land, power, cooling, interconnection, development capital and operating credibility at a speed and scale that customers would struggle to replicate internally. The economics are therefore less like ordinary property leasing and more like a capital-allocation test: can Digital Realty convert AI-density demand into rentable power and connectivity revenue while keeping enough balance-sheet discipline to avoid letting the boom consume the REIT?
The latest public numbers put the tension in plain view. In the first quarter of 2026, Digital Realty reported total bookings expected to generate $707 million of annualized GAAP base rent at 100% share; Digital Realty's own share was $423 million. Those bookings covered 312.8 megawatts at 100% share and 176.0 megawatts at Digital Realty's share, with a grand-total GAAP base rent of $183 per kilowatt at 100% share and $191 per kilowatt at Digital Realty's share. The same release put total backlog at $1.8 billion of annualized GAAP base rent at 100% share and $1.0 billion at Digital Realty's share, with a nineteen-month weighted average lag between lease signing and contractual commencement. The source documents are Digital Realty's Q1 2026 release and supplement: https://www.globenewswire.com/news-release/2026/04/23/3280372/0/en/digital-realty-reports-first-quarter-2026-results.html and https://investor.digitalrealty.com/static-files/953419cb-91ee-4485-8017-26ee0b29bb2a.
That is the hard-number spine for the article. A business with $1.6 billion of first-quarter revenue, $920 million of adjusted EBITDA, $18.0 billion of total debt and 4.7x net debt-to-adjusted EBITDA is not being valued only for its present rent roll. It is being valued for whether the $1.0 billion Digital Realty-share backlog turns into cash rent on time, whether the 176.0 megawatts signed in Q1 were leased at returns that clear a higher capital cost, and whether the company can fund $3.5 billion to $4.0 billion of 2026 development capex net of partner contributions without losing the investment-grade characteristics that make the model work. Digital Realty's investor page points to the same Q1 2026 materials and positions the company as a global provider across enterprise colocation and hyperscale: https://investor.digitalrealty.com/financials/quarterly-results.
The scarce input is committed power, not the corporate logo on the building
The simplest way to misunderstand Digital Realty is to treat it as a large landlord with a technology customer list. In the AI cycle, the scarce input is not merely square footage. It is the right to use a large amount of power in the right metro at the right time, inside a facility whose cooling plant can handle high-density racks and whose interconnection fabric gives customers optionality. A shell without utility commitment is not equivalent to a leased megawatt. A powered building in a weak connectivity location is not equivalent to a campus in Northern Virginia, Frankfurt, Paris, Singapore, Chicago, Dallas or an emerging gateway such as Accra. The rent is paid for the bundle.
Digital Realty's 2025 Form 10-K makes the scale of that bundle visible. It says the portfolio included 310 data centers as of December 31, 2025, including 89 held as investments in unconsolidated entities, spread across the United States, Europe, Latin America, Africa, Asia, Australia and Canada. The same filing says the portfolio contained about 57.6 million rentable square feet, including 9.7 million square feet under active development and 4.7 million square feet held for development, and estimated more than 3,500 megawatts of additional data-centre capacity on land and other space held for, or under, construction. The filing is here: https://www.sec.gov/Archives/edgar/data/1297996/000110465926015365/dlr-20251231x10k.htm.
Scale by itself does not prove pricing power. The useful question is whether Digital Realty can turn that scale into a shorter path to revenue for customers and a disciplined return on capital for shareholders. Its public site claims a footprint of 300+ data centers in 55+ metros and more than a decade of 99.999% uptime, while its locations page lists 55+ global metros, 300+ data centers, 500+ available clouds and 250+ Fortune 500 companies: https://www.digitalrealty.com/ and https://www.digitalrealty.com/data-centers. Those figures are marketing shorthand, but they also describe why a buyer with a global cloud or AI inference estate might prefer a platform lease over a one-off self-build: each new metro comes with contracting, compliance, cross-connect, power-procurement and operating friction.
The REIT's job is to internalize that friction and charge rent for lowering it. The customer's job is to compare that rent with the full cost of doing it alone. In a normal real-estate market, self-build can look cheaper if the buyer has patience and a strong internal development team. In the current market, the cost of delay changes the answer. If the customer misses a product window, cannot train a model on schedule, or cannot put inference capacity near users when demand appears, the paper savings from owning the building may be irrelevant. Digital Realty is selling a time-to-power claim, not just a floor plan.
Backlog is future rent with execution risk attached
Digital Realty's Q1 2026 backlog is the central evidence of that time-to-power trade. A $1.8 billion backlog at 100% share looks powerful because it is already signed but not yet commenced. It is also a warning label: the rent is not fully in the current income statement, and the nineteen-month weighted average lag means that capital, construction and customer readiness have to line up across several quarters. The company's Q1 release says the backlog was $1.0 billion at Digital Realty's share and that Q1 bookings carried a nineteen-month weighted-average lag: https://investor.digitalrealty.com/static-files/186019d2-aa98-4bf5-a1c6-15bb5af879ed.
That lag matters because AI demand can be both urgent and volatile. The immediate need for training clusters creates pressure for large contiguous blocks. Inference may be more distributed and latency-sensitive. Enterprise AI adoption may require smaller deployments near existing data estates, cloud on-ramps and compliance boundaries. A landlord that signs a massive long-term lease is not immune from technology risk; it has transformed that risk into counterparty credit, construction delivery, power delivery and residual-location risk. The backlog is therefore not a trophy. It is a financed promise to convert capital into rent.
The Q1 2026 rent table is more informative than the headline. At Digital Realty's share, 0-1 megawatt leases in all regions generated $78.954 million of annualized GAAP base rent across 26.6 megawatts, or $247 per kilowatt. Leases above 1 megawatt generated $324.482 million across 149.3 megawatts, or $181 per kilowatt. Interconnection added $18.611 million. The smaller-capacity and interconnection category is higher priced per kilowatt, but the hyperscale category supplies the bulk of megawatts and often carries much longer terms. The Q1 supplement shows the same split and also shows weighted-average lease terms of 4.2 years for new 0-1 megawatt leasing and 13.0 years for new greater-than-1-megawatt leasing in the quarter: https://investor.digitalrealty.com/static-files/953419cb-91ee-4485-8017-26ee0b29bb2a.
This split defines Digital Realty's strategy. The company wants the credit and scale of hyperscale leasing, because that can underwrite large power blocks and campus development. But the company also wants the denser economics of enterprise colocation and interconnection, because cross-connects and smaller deployments can make a campus more defensible and less commodity-like. AI demand can help both sides, but not in the same way. Training demand rewards massive, contiguous power. Inference and enterprise AI reward location, network density and private connectivity. If Digital Realty's portfolio is only a cheap megawatt supplier, it competes with every well-funded developer chasing power. If it is a connected platform, the rent has more layers.
The revenue outlook reinforces the point. Digital Realty raised its 2026 total revenue outlook to $6.65 billion to $6.75 billion and adjusted EBITDA outlook to $3.65 billion to $3.75 billion in April 2026, while raising Core FFO per share to $8.00 to $8.10. It also raised development capex net of partner contributions to $3.5 billion to $4.0 billion. This is the economics of accelerated demand: more growth visibility, more capital spending, more need to defend funding costs. The Q1 release lays out those outlook ranges: https://www.globenewswire.com/news-release/2026/04/23/3280372/0/en/digital-realty-reports-first-quarter-2026-results.html.
Rent per kilowatt is only half the margin story
The $191 per kilowatt GAAP base rent figure in Q1 2026 is useful, but it is not a complete unit-economics answer. Data-centre returns depend on the capital cost to build the megawatt, the cost and timing of grid connection, lease term, escalation, customer credit, maintenance capex, tax and insurance, the share of utility reimbursement that is pass-through rather than margin, and the value of interconnection around the initial lease. A 149.3 megawatt greater-than-1-megawatt booking at $181 per kilowatt can be attractive if it is long duration, pre-leased, efficiently financed and attached to a strategic customer. A smaller deployment at a higher per-kilowatt rent can be more profitable per unit but may not absorb a large campus on its own.
Digital Realty's Q1 occupancy table gives another part of the picture. The portfolio total/weighted average occupancy was 90.1% at 100% share and 89.4% at Digital Realty's share as of March 31, 2026, excluding capacity under active development and held for development. Northern Virginia was 98.6% occupied at 100% share, Chicago 95.9%, Dallas 93.2%, Frankfurt 93.5%, Singapore 93.3%, while London, Johannesburg, Hong Kong and some other markets sat lower. The same table shows portfolio data-center count of 309. That mix tells investors that scarce-power markets can be effectively full while some metros still have leasing or repositioning work to do. The supplement is the source: https://investor.digitalrealty.com/static-files/953419cb-91ee-4485-8017-26ee0b29bb2a.
The Q1 renewal data is also economically important. Digital Realty reported rental rate increases on renewal leases of 5.0% on a cash basis and 6.3% on a GAAP basis in Q1 2026. In the supplement's renewal detail, total renewed cash rent per kilowatt or square foot rose 4.8% for data-center renewals in the quarter, and 6.1% over the last twelve months. In a capital-intensive REIT, renewal spread is a quiet test of scarcity. If customers can easily move workloads, rent growth is fragile. If migration risk, cross-connect value and power scarcity bind customers to the site, the landlord has more pricing room. But aggressive renewal pricing can also push cost-sensitive enterprise customers to reconsider architecture over time.
The operating margin question therefore has two layers. First, can Digital Realty earn a spread between stabilized development yield and its weighted cost of capital? Management's 2026 outlook assumes average stabilized yields of 10.0%+ on development, while long-term debt issuance was expected at 4.0% to 4.5% in the April outlook. Second, can the company keep that spread after recurring capex, leasing costs, utilities, taxes, insurance and cooling investments? The answer will not be visible in one quarter. It will show up in commencements, Same-Capital cash NOI, renewal spreads, development yields and leverage over several years.
Development is a funding problem before it is an engineering problem
Digital Realty's Q1 2026 investment activity reads like a map of the industry's power hunt. The company acquired an 873-acre parcel in greater Atlanta for $95 million, expected to support over one gigawatt of IT capacity. It acquired 30 acres in the Portland metro for $50 million, expected to support 160 megawatts, near another assemblage expected to support up to 85 megawatts. It acquired Telepoint in Sofia for EUR 66.5 million, or $76.6 million, and two land parcels totaling more than 90 acres near Milan for EUR 56.5 million, or $65.1 million. It also disclosed Malaysia transactions around TelcoHub 1 and Cyberjaya. These facts are in the Q1 release: https://www.globenewswire.com/news-release/2026/04/23/3280372/0/en/digital-realty-reports-first-quarter-2026-results.html.
That list is not random. Atlanta, Portland, Milan, Sofia and Cyberjaya represent different versions of the same search: find markets where power, land, subsea or terrestrial routes, cloud demand and customer expansion can be combined before competitors bid away the return. The greater Atlanta parcel is an explicit power-scale bet. Milan points to Mediterranean connectivity and European demand. Sofia is an interconnection acquisition rather than just a land bank. Malaysia gives Digital Realty a connected campus angle near regional cloud and network growth. The company is buying future rent options, not merely locations.
But options cost money. Digital Realty ended Q1 2026 with about $18.0 billion of total debt outstanding, including $17.2 billion of unsecured debt and about $0.8 billion of secured and other debt. Net debt-to-adjusted EBITDA was 4.7x; debt-plus-preferred-to-total enterprise value was 22.7%; fixed-charge coverage was 4.9x. The company also sold 7.3 million shares under its ATM equity issuance program since December 31, 2025, at a weighted average price of $179.30 per share, raising about $1.3 billion of net proceeds. Those figures are in the Q1 materials: https://investor.digitalrealty.com/static-files/186019d2-aa98-4bf5-a1c6-15bb5af879ed.
The balance sheet is why Digital Realty's private-capital strategy matters. On March 30, 2026, the company announced the final close of its inaugural U.S. hyperscale data center fund, securing $3.25 billion of total equity commitments from global institutional investors. The fund focuses on ownership and development of hyperscale data centers across major U.S. Tier I metros, including Northern Virginia, Santa Clara, Dallas, Atlanta, Charlotte and New York. Digital Realty retained a 20% interest and serves as manager: https://investor.digitalrealty.com/news-releases/news-release-details/digital-realty-announces-final-close-325-billion-us-hyperscale.
That structure is economically important because it lets Digital Realty participate in large hyperscale demand without carrying 100% of every development dollar. The company can earn its share of asset economics, retain customer control, manage development and operations, and use outside equity to preserve capacity for the next site. The tradeoff is that some asset-level upside is shared. For a REIT, that may be a rational price for scale if the alternative is overleveraging the balance sheet in the middle of a demand boom.
The June 2026 Blackstone transaction shows the other side of the same discipline. On June 29, 2026, Digital Realty agreed to purchase Blackstone-affiliated funds' blended 64% equity interest in three fully leased Northern Virginia data centers containing 288 megawatts of total IT capacity, valuing the assets at $7.8 billion at 100% share. Total consideration to Blackstone was $3.5 billion, including $1.2 billion of cash and $2.3 billion in Digital Realty shares. Each of the three data centers carries 96 megawatts; the portfolio is 100% leased to three distinct investment-grade hyperscale customers; the leases average 15 years, with a blended AA- customer credit rating and 3.6% annual rent escalators. The announcement is here: https://investor.digitalrealty.com/news-releases/news-release-details/digital-realty-announces-purchase-blackstone-interest-three.
The market should read that deal as a capital-allocation test, not simply a growth headline. In 2023, Digital Realty and Blackstone announced a $7 billion hyperscale development joint venture expected to support about 500 megawatts across Frankfurt, Paris and Northern Virginia, with Blackstone taking an 80% interest and Digital Realty retaining 20%: https://www.digitalrealty.com/about/newsroom/press-releases/123237/digital-realty-and-blackstone-announce-7-billion-hyperscale-data-center-development-joint-venture. In 2026, Digital Realty is buying back control of a selected, fully leased Northern Virginia portion. That can make sense if the lease duration, escalators, customer credit and initial stabilized capitalization rate above 6.5% are better than the marginal opportunity elsewhere. It can be dangerous if buybacks of developed assets consume too much liquidity and reduce flexibility. The discipline is in choosing when full ownership is worth more than balance-sheet optionality.
Interconnection is the second rent roll
If Digital Realty were only selling powered shells, hyperscale customers could pressure margins over time. Interconnection is what gives parts of the estate a second economic layer. Digital Realty says its global platform includes more than 234,000 cross-connects and 55+ metros on its investor pages, while its June 2026 investor presentation cites 4,350+ network instances, 234K+ cross-connects and 30+ metros with over 1,000 cross-connects each: https://investor.digitalrealty.com/ and https://investor.digitalrealty.com/static-files/3fa86ccd-e071-4b5f-85d7-5885df9e4b80. Cross-connects are not glamorous, but they are one of the reasons customers stay. The cost of moving a workload is not just truck rolls and server relocation; it is the cost of rebuilding network adjacency.
Digital Realty's ServiceFabric page frames this as a private virtual connectivity layer for AI, hybrid cloud, partners, model providers, services and locations. It says ServiceFabric connects across 800+ data centers and 350+ cloud on-ramps: https://www.digitalrealty.com/platform-digital/connectivity/service-fabric. The company's Internet Exchange page describes a neutral, privately owned and managed exchange that lets participants aggregate ISPs, content providers, enterprises and others on one peering fabric: https://ix.digitalrealty.com/. PeeringDB gives an external view of the same ecosystem, listing Digital Realty facilities, networks and exchanges under the organization page: https://www.peeringdb.com/org/8592.
For AI, the interconnection value depends on where the workload sits. Training can be more location-flexible if the customer can tolerate distance from end users and has enough dedicated network capacity. Inference is more sensitive to latency, data sovereignty, data gravity and enterprise integration. A hospital, bank, media platform or national enterprise may not want every inference call to cross a proprietary public-cloud path. A campus with private connectivity to clouds, carriers, partners and existing data stores can therefore carry higher strategic value than a remote low-cost power site. This is why the "rentable power" argument should not be detached from the "rentable network optionality" argument.
The Q1 2026 booking detail hints at this. The 0-1 megawatt category plus interconnection contributed $98 million at Digital Realty's share, while the greater-than-1-megawatt category contributed $324 million. The larger category won the headline because it absorbed the megawatts. The smaller category and interconnection preserve price density and ecosystem value. If the company can keep both engines working, Digital Realty becomes less exposed to a binary choice between cheap hyperscale capacity and expensive enterprise colocation. It can place large customers on campuses where network density and enterprise demand deepen over time.
Ghana shows why small connected nodes can matter
Digital Realty's directory row in this assignment flags Accra ACR2 in Ghana, and Ghana is a useful counterweight to the Northern Virginia megawatt story. ACR2 is not a 96 megawatt hyperscale building. Digital Realty's Accra page describes 11.8k square feet, or 1.1k square meters, of colocation space, one data center, N+2 cooling, PCI-DSS and ISO 27001 certifications, and a position as a gateway to Western Africa: https://www.digitalrealty.com/data-centers/emea/accra. The ACR2 facility page places it at Bank Street and Prof. Atta Mills High Street in Accra, within close proximity to submarine cable landing points and metro fibre infrastructure, with an 11,800 square foot two-story building, N+1 UPS redundancy and N+2 cooling redundancy: https://www.digitalrealty.com/data-centers/emea/accra/acr2.
Public reporting around the launch filled in more operating detail. Data Center Dynamics reported that Digital Realty's Ghana facility would provide 1.7 megawatts of capacity and about 1,100 square meters of space: https://www.datacenterdynamics.com/en/news/digital-realty-to-establish-its-first-data-center-in-ghana/. Ghana's Ministry of Communication, Digital Technology and Innovations described ACR2 as a state-of-the-art data centre in Accra and said it would support Ghana's digitalisation agenda: https://moc.gov.gh/2025/10/31/digital-realty-opens-state-of-the-art-acr2-data-centre-in-accra/. Digital Realty's own Ghana press-release page is here: https://www.digitalrealty.com/about/newsroom/press-releases/123355/digital-realty-expands-presence-in-west-africa-with-first-data-center-in-ghana.
The strategic point is not the size of ACR2. It is the location. Digital Realty's Accra metro page says the subsea 2Africa cable lands in its Accra data center, positioning Ghana as a gateway to West Africa. LINX announced in June 2025 that Digital Realty's ACR2 would be an access point for LINX Accra, a new interconnection fabric, and cited six subsea cable systems as a driver of Ghana's connectivity growth: https://www.linx.net/news/linx-accra-digital-realty/. PeeringDB lists Digital Realty Accra ACR2 as a facility with local exchange presence: https://www.peeringdb.com/fac/14469.
For Digital Realty, a small African connectivity hub does not move consolidated revenue like a large U.S. hyperscale lease. But it can strengthen the global platform thesis. The company already has African exposure through Teraco in South Africa, where Teraco's homepage describes 8 locations, 650 clients, 27,000 interconnects and 228 MW of IT load: https://www.teraco.co.za/. It also has iColo operations in Kenya and other African markets through prior transactions. If AI and cloud demand become more distributed, regional gateways matter because data cannot always be hauled back to the same few global hubs. They matter for banks, content platforms, public-sector workloads, mobile operators and cloud access in markets where local latency, cable diversity and regulatory comfort are part of the buying decision.
The uncertainty is that emerging-market demand can be uneven. A 1.7 megawatt facility is not proof that Ghana will rapidly absorb a full hyperscale campus. Power reliability, local enterprise budgets, cloud-region strategy, currency risk, subsea cable resilience and carrier competition all shape the outcome. The more disciplined interpretation is that ACR2 is a strategic option on West African interconnection, not a standalone answer to Digital Realty's global growth. It is valuable if it draws networks and enterprises into the Digital Realty fabric and if that fabric becomes a reason to buy in other metros.
AI density raises the engineering threshold and the political cost
The AI demand story is ultimately an electricity story. The International Energy Agency's Energy and AI executive summary says data-centre electricity consumption is set to more than double to around 945 TWh by 2030, with AI as the most important driver alongside other digital services: https://www.iea.org/reports/energy-and-ai/executive-summary. The IEA's energy-demand chapter examines scenarios in which total data-centre demand can vary substantially depending on efficiency, adoption and technology assumptions: https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai. For a REIT, that range is not academic. It determines how many substations, transformers, generators, chillers, water systems, grid upgrades and power-purchase arrangements the company has to underwrite.
Digital Realty's June 2026 investor presentation argues that average rack density rose from 7 kW in 2021 to 27 kW per rack in 2025 and that future AI and HPC racks may require advanced cooling modes, including direct liquid to chip and rear-door heat exchangers: https://investor.digitalrealty.com/static-files/3fa86ccd-e071-4b5f-85d7-5885df9e4b80. The claim is directionally consistent with the market: AI clusters pack more power into fewer racks, generating heat loads that older facilities were not designed to absorb. A building that was attractive for enterprise colocation five years ago may not be ready for dense AI without material electrical and mechanical upgrades.
This is where Digital Realty's scale can be an advantage and a liability. Scale gives purchasing power, design experience and a multi-market view of customer requirements. It also gives public visibility. Local communities, utilities and regulators increasingly ask whether data centres are crowding out residential load growth, industrial projects or decarbonisation goals. Long lead times for power equipment can delay commencements even after a lease is signed. Water use can become a political issue in markets where evaporative cooling is contentious. Noise, diesel backup, tax incentives and land use can all become local flashpoints.
Digital Realty's sustainability record is therefore not decorative; it is part of the licence to grow. In May 2026, the company said its 2025 Impact Report showed 93% global renewable energy coverage, 1.7 GW of large-scale renewable energy capacity contracted, 205 sites matched with 100% renewable and emissions-free energy, 75% of sites operating without evaporative cooling, 1.38 global PUE and 1.31 EMEA PUE. It also said six data centers totaling 1.8 million square feet and 196 MW-IT achieved sustainable-building certifications, with an average design PUE of 1.20 for those six delivered data centers: https://www.globenewswire.com/news-release/2026/05/27/3302380/0/en/digital-realty-publishes-2025-impact-report-highlighting-sustainability-progress.html.
Those numbers are useful, but they need careful interpretation. Renewable energy coverage is not the same as every facility consuming carbon-free electricity every hour. PUE is a facility-efficiency measure, not a measure of absolute energy use. Water performance can improve while total water use still rises if the portfolio grows. Digital Realty's 2025 report highlights are evidence of operating focus, not a guarantee that every new high-density campus will face smooth community approval. The watchpoint is whether sustainability metrics continue improving as AI density and total power load rise.
The market backdrop is scarcity pricing, but scarcity attracts capital
The broader data-centre market supports Digital Realty's pricing argument. CBRE's North America Data Center Trends H2 2025 report describes record demand, low vacancy, supply growth and power supply challenges in major markets: https://www.cbre.com/insights/books/north-america-data-center-trends-h2-2025. JLL's 2026 Global Data Center Outlook says AI represented about a quarter of all data-centre workloads in 2025 and could represent half by 2030, with inference expected to become the primary driver as the market evolves: https://www.jll.com/en-us/insights/market-outlook/data-center-outlook. These are not Digital Realty-specific sources, but they frame the environment in which the company is leasing capacity.
Scarcity is favorable for incumbents until it is not. When vacancy is low and power is delayed, customers pre-lease earlier and accept longer terms. That supports development yields and rent growth. But scarcity also invites capital from infrastructure funds, private credit, utilities, hyperscalers, sovereign wealth funds and specialist developers. Some customers may decide that owning power generation, signing direct utility arrangements, or partnering with private developers gives them more control than leasing from a public REIT. Digital Realty's advantage is its operating history and platform breadth. Its risk is that the highest-growth part of the market becomes too capital hungry for public shareholders to fund at the desired pace.
The credit-rating page underscores why the company cannot ignore that risk. Digital Realty lists long-term issuer ratings of BBB+ with stable outlook from S&P, Baa2 with positive outlook from Moody's, and BBB with stable outlook from Fitch: https://investor.digitalrealty.com/financials/credit-ratings. Investment-grade access is part of the competitive moat. If the company can borrow long-term capital at attractive spreads while smaller developers depend on more expensive capital, it can win sites and customers. If leverage rises or development delays erode confidence, the cost of capital advantage narrows precisely when the market requires more funding.
That is why the Q1 balance sheet metrics deserve as much attention as bookings. Net debt-to-adjusted EBITDA at 4.7x is a stronger starting point than many investors would have expected at this stage of the AI capex cycle. Total debt of about $18.0 billion is still large. The June Blackstone purchase adds strategic control but also requires cash and equity issuance. The $3.25 billion hyperscale fund adds outside capital but shares economics. Digital Realty is trying to thread the needle: use third-party capital when it helps scale, use the public balance sheet when control of fully leased assets is worth it, and keep the credit story intact.
Customer concentration is manageable only if credit and duration hold
Data-centre landlords often advertise customer count, but economics concentrate in large buyers. Digital Realty's Q1 supplement shows its top 20 customers generated $2.503 billion of annualized recurring revenue, representing 51.9% of the portfolio, with a weighted-average remaining lease term of 6.0 years. The definition includes monthly contractual base rent and interconnection revenue under existing leases as of March 31, 2026, multiplied by 12: https://investor.digitalrealty.com/static-files/953419cb-91ee-4485-8017-26ee0b29bb2a. Concentration is not automatically negative if the customers are investment-grade cloud, network, financial or technology buyers with long-term commitments. It is negative if one or two demand narratives dominate capital allocation.
The 2026 AI cycle increases both sides of the concentration trade. Hyperscale customers can sign huge leases that derisk development. They can also demand lower per-kilowatt rents, specific technical designs, custom phasing and strict delivery obligations. A fully leased 96 megawatt building is a powerful asset if the tenant credit is strong and the rent escalates at 3.6% annually, as in the Blackstone portfolio Digital Realty agreed to buy. It is less flexible than a multi-tenant interconnection building if technology requirements change or if the customer has bargaining power at renewal.
Digital Realty's better answer is not to avoid hyperscale. The market is too large, and the company's land and power inventory are too valuable. The better answer is to keep layering connectivity-rich enterprise demand, regional gateways, cloud on-ramps, Internet exchange activity and private virtual connectivity around the power campuses. That makes the platform more than a financing wrapper for cloud megawatts. It also gives the company multiple ways to monetize the same metro over time.
The evidence is mixed but constructive. Digital Realty's interconnection revenue contribution in Q1 bookings was modest relative to hyperscale rent, yet the company has a broad cross-connect base and a ServiceFabric product that speaks to distributed AI and hybrid-cloud use cases. The Sofia acquisition of Telepoint is small next to Northern Virginia, but it adds a highly connected Southeast Europe hub; Digital Realty announced the acquisition in March 2026: https://www.digitalrealty.com/about/newsroom/press-releases/20096/digital-realty-enters-bulgaria-with-acquisition-of-highly-connected-interconnection-hub-in-sofia. Accra is small, but it ties West African subsea and exchange development to the platform. These assets matter if the next phase of AI is not just bigger training clusters but more distributed inference and regulated enterprise deployment.
The economic judgment
Digital Realty's current advantage is that it can make power rentable. That sounds simple, but in 2026 it is a rare capability. The company can show customers a global campus map, a real operating record, investment-grade funding, interconnection depth, a private-capital strategy and enough leasing evidence to prove that buyers are committing before delivery. The Q1 2026 bookings number, the $1.8 billion total backlog, the 176.0 megawatts signed at Digital Realty's share, and the $191 per kilowatt GAAP base rent are not abstract indicators. They are proof that customers are paying for future capacity before it is in service.
The company's vulnerability is that the same demand that strengthens pricing also raises the capital burden. A backlog that equals 23% of in-place annualized rent, as Digital Realty's Q1 presentation describes, is valuable only if projects commence, power is delivered and customers occupy on schedule: https://investor.digitalrealty.com/static-files/6de94c2d-af96-43ca-ab81-45ab3dc11eb6. Development capex of $3.5 billion to $4.0 billion in 2026, recurring capex plus capitalized leasing costs of $400 million to $425 million, and large land and asset transactions mean free cash flow will remain under pressure. The REIT has to fund growth before all of the rent arrives.
For a shareholder or customer, the right question is not "Is AI good for Digital Realty?" It clearly is, in demand terms. The better question is whether Digital Realty can preserve the spread between development yield and capital cost while maintaining customer relevance across both hyperscale and interconnection-heavy deployments. If the company chases every megawatt, returns can dilute. If it refuses to fund enough capacity, customers may go elsewhere. If it leans too far into third-party capital, it may sacrifice upside. If it buys back too many developed assets, it may reduce flexibility. The strategy is a balancing act because the product is capital intensive and time sensitive.
The strongest evidence in Digital Realty's favor is that management has not treated AI demand as a reason to abandon financial structure. The company raised equity through the ATM program, closed a $3.25 billion hyperscale fund, used joint ventures, retained investment-grade ratings, and still selectively moved to increase ownership of fully leased Northern Virginia assets where lease term, escalators and customer credit appear attractive. That does not remove execution risk, but it suggests the company understands the distinction between demand and value creation.
The biggest uncertainty is the future cost of power and grid delivery. Data-centre demand can be visible while utility execution is not. A site can be fully leased but still depend on substations, transmission upgrades, equipment availability, environmental approvals and political consent. AI customers can be urgent but also demanding, and the technology mix can change. Digital Realty's rent roll is contracted in leases; its development economics still pass through local power markets and capital markets.
Bottom line and watchpoints
Digital Realty should be read as a balance-sheet and power-conversion story. The company is not merely riding AI enthusiasm. It is trying to turn scarce power, cooling and interconnection into long-duration rent without letting development spend outrun capital discipline. The Q1 2026 numbers show the promise: $707 million of annualized bookings at 100% share, $1.8 billion of total backlog, $1.6 billion of quarterly revenue, $920 million of adjusted EBITDA, 90.1% portfolio occupancy at 100% share, 4.7x net debt-to-adjusted EBITDA, and a 2026 revenue guide of $6.65 billion to $6.75 billion. The same numbers show the pressure: nineteen months between signing and commencement, billions of dollars of development capex, a need for external capital, and a market where power is the binding constraint.
The first watchpoint is backlog conversion. Investors should track how much of the $1.8 billion total backlog commences in 2026 and 2027, whether delays rise, and whether commenced rent carries the expected margin. The second is development yield versus capital cost. Management's 10.0%+ stabilized-yield assumption is attractive only if construction, power and financing costs stay inside the model. The third is leverage after the June Blackstone transaction and any further asset purchases. The fourth is renewal pricing, especially whether 0-1 megawatt and enterprise customers continue accepting higher cash rents as AI raises power density across the estate. The fifth is interconnection growth, because cross-connects, exchange participation and ServiceFabric adoption are what keep Digital Realty from being judged only as a commodity megawatt supplier.
The sixth watchpoint is local power politics. Digital Realty's 93% renewable energy coverage, global PUE reporting and water-efficiency metrics help the licence-to-grow argument, but they do not eliminate regional opposition or grid constraints. The seventh is geographic option value. Northern Virginia remains central, but Accra, Sofia, Milan, Malaysia and the African platform show that smaller connected nodes can matter if cloud, financial, content and public-sector demand becomes more regional. The final watchpoint is customer control. Hyperscale leases can underwrite megawatts, but Digital Realty needs enough product breadth and connectivity value to keep the platform strategically relevant when the AI workload mix shifts.
The economic conclusion is therefore conditional but strong. Digital Realty is one of the few public vehicles with the scale, credit access, operating footprint and interconnection base to monetize the AI data-centre cycle without starting from zero. Its edge is the ability to make future power bankable for customers and financeable for investors. Its risk is that the price of turning that future power into rentable capacity keeps rising. The company wins if it converts backlog into cash rent faster than power, construction and funding costs erode returns. It disappoints if the AI boom becomes a capital sink rather than a rent-growth cycle.