How do companies in the data center market make money?
In our data center market deck, you will find everything you need to understand the market
SUMMARY
Companies in the data center market make money by turning scarce, deliverable power and connectivity into long-term recurring revenue, then adding higher-margin income from interconnection, services, development and investment partnerships.
The industry increasingly looks less like ordinary real estate. A building has limited value without committed electricity, cooling infrastructure, fiber and a realistic energization date, which is why control of usable megawatts has become one of the main sources of competitive advantage.
Customer size changes the economics dramatically. Smaller deployments can pay almost twice as much per megawatt as hyperscale customers, while hyperscalers compensate with contracts that can absorb an entire campus and run for more than a decade.
That makes retail colocation and hyperscale development almost different businesses. Retail operators maximize revenue from each kilowatt and diversify across many customers; wholesale operators sacrifice some pricing in exchange for scale, long leases and faster absorption.
Connectivity can make a mature urban data center much more valuable than its electricity alone would suggest. Once clouds, carriers, financial firms and hundreds of other customers are interconnected inside the same facility, moving elsewhere becomes increasingly inconvenient.
Electricity reimbursement can make reported revenue look larger without creating equivalent profit. The more important economics sit in controlling scarce power capacity and using it efficiently enough to support more computing equipment from the same grid connection.
AI has strengthened both sides of the business at once: rents and lease renewals are rising in constrained markets, but new facilities are becoming substantially more expensive, more power-intensive and more dependent on a small number of enormous customers.
Preleasing is removing a large part of traditional development risk. In the tightest markets, customers are committing to capacity months or years before opening, allowing developers to start billion-dollar projects with much of the future revenue already contracted.
Outside capital is becoming part of the operating model rather than just a financing tool. Joint ventures with pension funds, sovereign funds and private-equity firms let specialist operators build far more capacity than their own balance sheets could support while still collecting fees, investment returns and development upside.
The biggest risk is not simply that AI demand slows. It is committing billions to the wrong market, with uncertain power delivery, weak preleasing, obsolete infrastructure assumptions or one tenant carrying too much of the economics.
The strongest data center businesses today therefore combine three scarce assets: power that can actually be delivered, locations or ecosystems customers genuinely need, and access to enough capital to keep building without overstretching the balance sheet.
This market map, featured in our data center market deck, highlights top companies and startups in the data center market
What do data center companies actually sell?
Data center companies make money by selling guaranteed access to power, cooling, physical capacity and networks, usually through recurring contracts.
The easiest mistake is to think of a data center as a warehouse full of servers. In many colocation facilities, the operator does not even own the servers. Customers bring or control the computing equipment, while the data center company provides the environment required to keep that equipment running around the clock.
That environment has become increasingly valuable. A customer may need several megawatts of electricity, redundant power feeds, backup generators, industrial cooling, physical security and high-capacity fiber connections. Providing all of that reliably is much harder than constructing the building itself.
The business also changes depending on the customer. A large AI company may reserve 100 MW in a hyperscale campus and sign a contract lasting more than a decade. A bank may rent a few hundred kilowatts inside a highly connected urban facility. A telecom operator may pay mainly because it wants direct access to hundreds of networks already present in the building.
So when we talk about the data center business model, "rent" covers several different products. Wholesale operators mainly sell large blocks of powered capacity. Retail colocation operators sell smaller blocks at higher prices and often make additional money from connectivity. Developers can also make money by building facilities for institutional partners and sharing the investment returns.
These days, the scarce resource underneath almost all of those models is usable power. A piece of land becomes dramatically more valuable when an operator can prove that hundreds of megawatts will actually reach the site.
Why is everyone building data centers right now?
Data center demand is extremely tight today because cloud computing is still expanding while AI has added a second, much more power-hungry wave of demand.
The shift was already well underway before generative AI took off. Synergy Research Group estimated that hyperscale companies controlled 48% of worldwide data center capacity at the end of 2025, while another 20% sat in non-hyperscale colocation facilities. Traditional company-owned, on-premise data centers were down to 32% of capacity. In 2018, they still represented 56%.
AI is now increasing the amount of infrastructure required per customer. The International Energy Agency estimates that global data center electricity consumption reached around 485 TWh in 2025 and could approach 950 TWh by 2030. Electricity use from AI-focused data centers is expected to grow roughly threefold over that period.
What makes the current market unusual is that construction is already running at extraordinary speed and still struggles to catch demand. CBRE's latest Global Data Center Trends study found that inventory across the four biggest North American markets increased 33% in a year. Vacancy nevertheless fell to 0.3% in Northern Virginia, 1% in Atlanta, 1.8% in Dallas-Fort Worth and 2.2% in Chicago.
Northern Virginia alone added more than 1.1 GW of inventory in one year. Customers absorbed slightly more than that.
Developers can therefore add the equivalent of several large power plants' worth of computing capacity and still see available space shrink. That is not how a normal property boom usually behaves; aggressive construction tends to produce empty buildings fairly quickly.
For now, data center demand is running into the physical speed limit of electricity grids, substations, transformers, generators, cooling equipment and construction crews.
As this chart shows, and as featured in our data center market deck, search interest in data centers has increased significantly
What exactly does a data center customer pay for?
A data center customer can pay separately for capacity, electricity, network connections and operating services, so two operators with similar revenue can have very different businesses.
The main charge is usually colocation or rental revenue. A customer reserves racks, cages, rooms or a specified amount of electrical capacity and pays every month for access to that infrastructure.
Electricity is often billed separately. Digital Realty's latest quarterly accounts, for example, showed roughly $1.15 billion of rental revenue alongside about $353 million of utility reimbursements. That utility line is large, although much of it compensates the company for electricity it has already purchased.
Connectivity creates another stream. Customers can pay for physical cross-connects, virtual connections and direct access to cloud platforms or telecom networks. At Equinix, interconnection currently brings in hundreds of millions of dollars each quarter.
Operators may also charge installation fees, remote-hands fees, managed infrastructure revenue and other services. Large platforms increasingly earn development and management fees from data centers partly owned by outside investors as well.
The useful distinction is between revenue that mainly passes costs through and revenue that comes from scarce infrastructure. Electricity reimbursement can inflate reported sales without producing the economics of an extra dollar of rent or an additional network connection.
| What the customer pays for | What the data center company provides | Typical role in the business |
|---|---|---|
| Colocation / rent | Reserved racks, rooms or power capacity | Main recurring revenue |
| Electricity reimbursement | Power consumed by customer equipment | Large revenue line, often partly pass-through |
| Interconnection | Direct links to clouds, carriers and other customers | Recurring, high-value add-on |
| Managed services | Operations, remote hands and infrastructure support | Smaller service revenue |
| Development / management fees | Building or managing assets for investment partners | More capital-light fee income |
Why do giant AI and cloud customers get cheaper data center prices?
Large AI and cloud customers get a major volume discount on data center capacity, while smaller deployments generate far more revenue per megawatt.
Digital Realty's latest new leasing gives us a clean comparison. Across its portfolio, new deployments of up to 1 MW were signed at an average GAAP rent of about $280 per kW per month. Contracts above 1 MW averaged around $157.
At those prices, one fully leased megawatt would produce roughly $3.36 million of annual rent in the smaller-deployment category, compared with about $1.88 million for the larger one. The small-customer product was therefore earning almost 80% more per unit of capacity.
The hyperscale customer gives something back in exchange for that discount: size and duration. Across Digital Realty's first six months of 2026, new leases below 1 MW had an average term of 4.2 years, while leases above 1 MW averaged 12.2 years.
Operationally, the difference is huge. Finding hundreds of customers to consume 100 MW can take years and requires a significant sales operation. One hyperscaler can potentially take the whole 100 MW campus under a handful of contracts.
Wholesale and retail data centers therefore have different economics. Retail operators try to earn more from each kilowatt. Hyperscale operators accept lower pricing per kilowatt because one agreement can absorb an extraordinary amount of capacity for more than a decade.
A 100 MW contract at $157 per kW per month works out to roughly $188 million of annualized rent before considering the exact lease structure. At that scale, even a heavily discounted customer can be extremely valuable.
| Recent Digital Realty leasing | Up to 1 MW | Above 1 MW |
|---|---|---|
| New GAAP rent per kW/month | ~$280 | ~$157 |
| Approx. annual rent per MW | ~$3.36M | ~$1.88M |
| Average new lease term, H1 2026 | 4.2 years | 12.2 years |
| Basic model | Higher price per MW | Huge volume and longer commitment |
If you want more recent data on this point, please see our latest data center market report.
This chart, featured in our data center market deck, illustrates yearly venture capital funding for data center startups
Why is interconnection such a good business for Equinix-style data centers?
Interconnection makes connectivity-heavy data centers much harder to replace because customers are paying to reach the companies already inside the ecosystem.
A customer renting space in an Equinix facility might want a private connection to AWS, Microsoft Azure, Google Cloud, a telecom carrier, a financial exchange or another company sitting nearby. Equinix can charge recurring fees for those links.
The scale is now substantial. In its latest quarter, Equinix generated $453 million of recurring interconnection revenue. Colocation remained much larger at $1.77 billion, but interconnection already accounted for about one-fifth of recurring revenue.
Equinix also added a record 9,700 net interconnections during the quarter. The value of a connectivity-heavy facility grows as more networks and customers join it. A company that wants direct access to ten important counterparties has a strong reason to choose the building where those ten counterparties already operate.
That creates very different customer behavior from a giant AI training facility built around cheap electricity. A training cluster may be able to move farther from a major city if sufficient power and fiber exist. A financial company that needs ultra-low-latency access to exchanges, networks and cloud on-ramps has fewer substitutes.
Equinix's customer concentration reflects this structure. Its largest customer currently represents only around 2% of recurring revenue, while its 50 largest together represent roughly 36%. The business relies on a dense network of customers rather than a few enormous tenants.
A megawatt in a connectivity hub can therefore be worth much more than a megawatt at an isolated hyperscale site. The electricity may be identical. The surrounding ecosystem is not.
Do data center companies actually make money on electricity?
Data center companies can earn some money around electricity, but the real economic advantage comes from controlling access to power rather than adding a huge markup to every kilowatt-hour.
Digital Realty illustrates the accounting clearly. Its latest quarter included roughly $353 million of tenant utility reimbursements. That is a meaningful part of reported revenue, yet the company also has to pay utilities and other power suppliers to obtain that electricity.
Many contracts therefore pass some or all of the customer's power cost through to the tenant. Higher electricity prices can push reported revenue upward while producing far less incremental profit than an equivalent increase in base rent.
What operators can monetize much more effectively is power availability. Consider a developer that controls a site where the utility has committed 300 MW on an acceptable timeline. That commitment gives the operator something a competing landowner may be unable to obtain for years.
Power efficiency can also improve margins. A facility that wastes less electricity on cooling can support more computing equipment from the same incoming grid connection. For a campus constrained at 100 MW, squeezing more useful computing capacity from those 100 MW has real economic value.
Electricity therefore plays two very different roles. The energy consumed is often largely reimbursed by customers; access to scarce electrical capacity can determine the value of the whole development.
If you want more recent data on this point, please see our latest data center market report.
This chart, featured in our data center market deck, shows how Equinix is capturing share in data centers
Why do AWS, Microsoft and Google still rent data centers?
AWS, Microsoft, Google and other hyperscalers still lease enormous amounts of data center capacity because renting can get computing infrastructure online faster than building everything themselves.
Synergy Research Group estimates that almost 60% of hyperscale capacity is currently housed in facilities the hyperscalers built and own themselves. Roughly 40% is leased.
That leased share is enormous when hyperscalers already control almost half of worldwide data center capacity.
The reason is practical. A hyperscaler may eventually prefer to own a huge strategic campus, but finding land is only the start. It still needs grid capacity, substations, permits, fiber, construction crews and years of execution. An established data center developer may already have all of those pieces moving.
Location also changes the calculation. A cloud company expanding into a new country may want capacity quickly without building an entire local real-estate and operations organization from scratch.
Recent leasing shows how long these relationships can become. Goodman Group, for example, recently signed a 20-year lease with a major hyperscale customer for the first 50 MW phase of its large Greater Tokyo data center campus. Long contracts like that give the customer secure capacity and give the developer unusually visible future cash flows.
Hyperscalers therefore switch between ownership and leasing depending on the situation. They build when long-term control justifies the investment and rent when an external developer can deliver the right megawatts sooner.
Can a data center be mostly rented before it even opens?
Yes. Large data centers are currently being leased so early that much of the capacity under construction in major markets already has a customer attached to it.
CBRE found that 80% of the data center space under construction across the four largest U.S. markets was already preleased at the end of 2025. That leaves surprisingly little upcoming inventory available to customers that have not reserved capacity yet.
Digital Realty offers a company-level view of the same phenomenon. Its latest reported backlog reached $1.9 billion of annualized base rent from signed leases that had not yet started, measured at 100% ownership. Digital Realty's own economic share was $1.4 billion.
Some contracts are signed far ahead of delivery. In the first quarter of 2026, the average gap between Digital Realty signing a new lease and the customer's contractual start date reached 19 months. The latest quarter was shorter at nine months, partly because of the mix of projects signed.
For the developer, preleasing changes the risk dramatically. A company can commit billions to a campus knowing that a customer has already agreed to rent much of it once construction finishes.
Construction and power risk remain very real. A delayed substation can delay the customer's rent commencement, and a project that runs over budget can still produce a poor return. But the developer has removed one of the biggest unknowns before opening day: whether enough customers actually want the capacity.
That is why the best development sites today can attract commitments while they are still drawings, construction zones or partially completed buildings.
This chart, featured in our data center market deck, illustrates yearly funding for data center startups
How do data center developers make money without owning 100% of the building?
Data center developers can earn property income, development fees, management fees and investment gains while outside investors fund a large share of the project.
The scale of new AI campuses makes this increasingly important. A 500 MW development can require several billion dollars before anyone installs the GPUs. Even very large operators eventually run into balance-sheet limits if they insist on owning every project outright.
Digital Realty showed one version of the model when three facilities in one of its development joint ventures reached performance milestones. The company recognized $188 million of net promote income in its latest quarter. A promote gives the developer an extra share of investment profits once agreed return thresholds are reached.
Equinix uses joint ventures heavily for its xScale hyperscale business. Its large U.S. partnership with GIC and CPP Investments was designed around more than $15 billion of eventual capital. Earlier this year, Equinix transferred its Hampton data center campus into that venture for $459 million of consideration while keeping an economic interest in the partnership.
That gives Equinix several ways to win. It can develop the site, earn fees, recover capital by transferring assets into the venture and retain part of the future property economics.
Private capital likes the other side of the trade. Pension funds, sovereign funds and infrastructure investors want long-lived assets backed by recurring contracts with large technology companies.
Blackstone's numbers show how far this has gone. Its mid-year investment update valued its global data center platform, including projects under construction, at around $165 billion and said leased capacity at QTS had increased roughly fifteenfold since Blackstone acquired the company.
These partnerships let the specialist operator run a much bigger development machine than its own equity base could support. That is becoming a core part of the model, not a side arrangement.
If you want more recent data on this point, please see our latest data center market report.
How profitable is a mature data center once it fills up?
A mature data center can produce operating margins around 50% or higher, although those margins come after years of heavy upfront investment.
Equinix's latest quarter is a good example. The company generated $2.63 billion of revenue and $1.40 billion of adjusted EBITDA, producing a record 53% adjusted EBITDA margin. Its cash gross margin reached 70%.
That is a strong infrastructure business once the buildings are operating and customers are paying recurring bills. Equinix also says recurring revenue has represented more than 90% of total revenue during each of the past three years, giving the company unusual visibility compared with many technology businesses.
Iron Mountain's data center segment shows similar economics. In 2025, segment revenue reached $803 million and adjusted EBITDA reached $416 million, a margin of about 52%. Revenue had grown almost 30% from the previous year.
The latest operating momentum remains strong as well. Iron Mountain recently reported 110 MW of data center leasing year to date, including 75 MW signed immediately after the second quarter ended.
The catch is capital expenditure. Equinix currently expects roughly $5 billion to $6 billion of total capital spending this year, and its updated long-term plan calls for $5 billion to $7 billion annually between 2027 and 2029.
A 50% EBITDA margin therefore tells us the stabilized asset is attractive. It does not tell us how much money had to be spent to create that asset or how long investors waited before it filled.
The real test is the return on that invested capital after construction costs, financing costs and development time.
This chart, featured in our data center market deck, compares the main business model options for hyperscale data center operators
Why does a modern data center cost so much to build?
A modern data center costs roughly $10 million to $15 million per megawatt before the customer even buys its AI servers, and the cost keeps climbing.
JLL's 2026 Global Data Center Outlook estimates that average shell-and-core construction costs increased from $7.7 million per MW in 2020 to $10.7 million in 2025. JLL expects about $11.3 million per MW this year.
Some major markets sit well above the average. JLL estimates roughly $12 million to $14 million per MW in Chicago and London, $11 million to $12 million in Northern Virginia and as much as $14 million to $18 million in Tokyo.
Those figures exclude land and active IT equipment.
A 100 MW facility built at $11.3 million per MW therefore implies roughly $1.13 billion of building and infrastructure costs before the tenant fills it with servers.
Then comes the expensive part for AI customers. JLL estimates that the technology fit-out for AI infrastructure can reach $25 million per MW. At 100 MW, that can mean another $2.5 billion of computing hardware and associated equipment.
Liquid cooling adds another cost. JLL estimates that liquid-cooled facilities can carry about a 10% construction premium over traditional air-cooled designs.
So a large AI campus quickly becomes a multibillion-dollar project even though the data center operator may own only part of the total capital sitting on the site. The landlord builds the powered environment; the customer can spend an even larger amount filling that environment with computing hardware.
Is securing power now more important than securing prime land?
For large AI data centers, securing deliverable power has become more important than owning a prestigious location.
The shift is visible in where new projects are going. CBRE's latest research says electricity availability and grid constraints are increasingly pushing development away from traditional hubs toward Tennessee, West Texas and other markets where large blocks of power can be delivered.
The International Energy Agency explains why the mismatch is so persistent. A data center can sometimes be built and operational within two or three years, while new grid infrastructure often takes much longer to plan, approve and construct.
Operators are adapting directly to that problem. CyrusOne has contracted more than 1.1 GW of power for Texas developments through agreements involving Calpine and Constellation. One project near the Freestone Energy Center starts with a 380 MW agreement and includes an exclusive arrangement for another 380 MW phase.
Another CyrusOne project in Fort Worth is being built next to existing high-voltage infrastructure and a grid-scale battery site specifically to shorten the time required to obtain electricity.
The market is moving even farther upstream. The Wall Street Journal recently reported that Nvidia was discussing an investment in Cloverleaf Infrastructure, a company focused on securing power and powered land for data centers. Cloverleaf has reportedly facilitated more than 7 GW of powered-land sales and has a pipeline above 10 GW.
That is a revealing development. When the world's dominant AI chip supplier considers investing in companies that assemble electricity and land before the data center is even built, power procurement has clearly become part of the computing supply chain itself.
Prime locations still command premiums where latency and network density are crucial. For a 500 MW AI training campus, though, a site with a realistic energization schedule can beat a famous data center market where the utility cannot deliver enough power for years.
If you want more recent data on this point, please see our latest data center market report.
This chart, featured in our data center market deck, shows the revenue mix across customer segments in the data center market
Is AI making data center economics better or riskier?
AI is making existing data center capacity more valuable today while making the next generation of projects far more expensive and concentrated.
The pricing upside is already visible in lease renewals. Digital Realty's latest renewals increased 25.4% on a cash basis overall. The most dramatic repricing came from contracts above 1 MW, where renewed rents were 66.7% above the expiring cash rents.
We should treat that 66.7% carefully because one quarter's lease mix can produce an extreme result. Digital Realty itself described the period as exceptional. Still, it shows what can happen when an old contract expires inside a severely supply-constrained market.
Equinix is responding to the same demand by spending much more. After its latest results, the company raised its expected annual capital expenditure for 2027 to 2029 from $3 billion-$4 billion to $5 billion-$7 billion. It also raised its long-term revenue-growth outlook.
AI therefore improves the revenue opportunity and raises the amount of capital required to chase it.
The hardware adds another uncertainty. Rack densities are increasing, liquid cooling is becoming more important and AI clusters place unusual demands on electrical systems. Facilities being designed today have to remain useful through several generations of hardware that do not exist yet.
Customer concentration can also increase quickly when individual projects reach hundreds of megawatts. Signing one hyperscaler can fill a campus, but losing or renegotiating that tenant later creates a much bigger hole than losing a normal enterprise customer.
For now, scarce capacity gives landlords considerable leverage. The risk grows if developers extrapolate today's shortage too far into the future and build expensive campuses without enough contracted demand.
Which data center markets are actually the most attractive today?
The most attractive data center markets today combine scarce capacity with achievable power delivery; the city charging the highest rent does not automatically produce the best investment.
Singapore shows how scarcity can support extraordinary pricing. CBRE currently puts asking rents there at roughly $330 to $475 per kW per month, the highest range among the major Asia-Pacific markets it tracks. Vacancy is only about 2%, helped by tight limits on new greenfield capacity.
Frankfurt leads the large European markets on price, around $235 to $265 per kW per month, with vacancy near 5%.
In the United States, Chicago currently commands around $200 to $230 per kW per month for a typical 250-to-500-kW requirement. Northern Virginia sits around $190 to $235 despite being by far the world's biggest data center market.
Yet very low vacancy can hide a development problem: developers may be unable to get enough power to add new capacity. Northern Virginia had only around 10.8 MW of available supply in CBRE's latest study even after adding more than 1.1 GW of inventory in a year.
We can see the opposite problem in Querétaro, Mexico. Its inventory increased about 450% in one year, pushing vacancy from 0.9% to 10.6%. Demand still exists, but supply arrived faster than customers could absorb it.
That comparison is useful. "AI demand is booming" does not protect every building in every city. Developers still need to understand how much supply is entering their particular market, when the power will arrive, what customers require and how much rent those customers will pay.
| Market | Recent vacancy | Recent asking rent where available | What stands out |
|---|---|---|---|
| Singapore | ~2.0% | $330-$475/kW/month | Extreme supply constraint and premium pricing |
| Frankfurt | ~5.0% | $235-$265/kW/month | Tight European market |
| Chicago | ~2.2% | $200-$230/kW/month | High U.S. pricing, difficult power timelines |
| Northern Virginia | ~0.3% | $190-$235/kW/month | Huge scale, almost no vacancy |
| Querétaro | ~10.6% | Varies | Supply jumped much faster than absorption |
This chart, featured in our data center market deck, shows how hyperscale AI-ready campus technology has evolved over time
What can make a data center investment go wrong?
A data center investment can go badly when expensive construction, delayed power and a small number of huge customers collide.
Customer concentration is one obvious fault line. Equinix has a very diversified retail-heavy model: its largest customer represents roughly 2% of recurring revenue. Digital Realty has much more hyperscale exposure. At the end of 2025, its largest customer represented 11.7% of annualized recurring revenue, while its 20 largest represented about 51%.
Neither structure is automatically better. Hyperscale tenants can sign very large, long leases and often have excellent credit. The downside appears when one customer represents an entire building or a large part of a campus.
Power timing creates another risk. A developer can buy land and announce a future gigawatt of capacity, but that future capacity generates nothing until electricity actually reaches the site. Years of grid delays can leave capital tied up without rent.
Financing is becoming equally important. CyrusOne recently expanded two major loan facilities to around $8 billion of combined commitments. Equinix issued $2.4 billion of senior notes in the first half of 2026. The amounts involved show how dependent the next stage of data center growth has become on deep capital markets.
Oversupply can still happen as well. Querétaro's jump from 0.9% to 10.6% vacancy is a useful reminder that a strong global theme can coexist with a weak local supply-demand balance.
Then there is technical risk. A facility designed around today's rack densities, cooling architecture and electrical configuration may need substantial upgrades as AI hardware changes.
The projects that look most dangerous are speculative campuses combining uncertain power delivery with uncertain customers and aggressive assumptions about future AI demand. A fully preleased facility with committed power and a creditworthy tenant is a very different investment even if both projects are described as "AI data centers."
If you want more recent data on this point, please see our latest data center market report.
So how do data center companies make money today?
Data center companies make money today by turning scarce megawatts into long-term recurring revenue, with additional profit coming from connectivity, development and outside capital.
At the basic level, a data center operator secures land and electricity, builds electrical and cooling infrastructure, and rents that capacity to companies that need somewhere to run servers. Large customers buy megawatts at lower prices under very large contracts. Smaller customers usually pay much more per kilowatt.
Connectivity-heavy operators add another layer. Equinix can keep charging customers for direct connections to cloud platforms, carriers and other businesses already inside its facilities. That makes some mature data centers much harder to substitute than a generic building with available power.
Developers have also learned to monetize their expertise without financing every project themselves. They can build campuses with pension funds, sovereign wealth funds and private-equity firms, earn development or management fees, collect incentive profits and retain minority stakes in the finished assets.
Power procurement now sits underneath all three models. A company that controls a viable 500 MW power pipeline has something hyperscalers desperately need. A company with an empty building and no additional electricity has much less room to grow.
The model is fairly clear by now. The data center market has evolved into a business of securing scarce infrastructure and selling access to it repeatedly. Real estate still matters, but the most valuable operators currently combine three things that are difficult to reproduce quickly: deliverable power, dense customer connectivity and access to enormous amounts of capital.
AI has made those advantages considerably more valuable. It has also raised the price of making a mistake. The winners should be the operators that secure power early, lease capacity before committing too much capital and use outside investors intelligently rather than trying to own every megawatt themselves.
| Data center business model | Main way it makes money | What makes it attractive | Main risk |
|---|---|---|---|
| Retail colocation | Higher rent per kW | High pricing and diversified customers | Slower to fill huge campuses |
| Connectivity-heavy colocation | Rent plus recurring interconnection | Strong ecosystem and switching friction | Best locations are expensive and power-constrained |
| Hyperscale / wholesale | Large long-term MW leases | Huge contracts and long visibility | Lower price per MW and customer concentration |
| Data center development | Building and leasing new capacity | Development upside | Construction, power and leasing risk |
| Joint-venture platform | Property income, fees, promotes and retained stakes | Can grow without funding everything alone | More financial complexity |
In our data center market deck, we identify pain points entrepreneurs should prioritize
OUR METHODOLOGY
This analysis examines how data center companies actually generate revenue and returns today. We break the market into the mechanisms that drive the economics: leasing, power, interconnection, development, operating margins, construction costs, preleasing, customer concentration and the increasing use of joint ventures and outside capital.
We prioritized the freshest available evidence rather than relying on broad industry averages. Company filings and financial disclosures were used for operator-level economics, including recent rents, lease terms, utility reimbursements, interconnection revenue, backlog, margins, capital expenditure, customer concentration and development profits.
Broader market conclusions were checked against independent infrastructure and energy research. CBRE was used for vacancy, inventory growth, preleasing and pricing across major data center markets; the International Energy Agency for electricity demand and grid constraints; Synergy Research Group for the shift toward hyperscale capacity and the owned-versus-leased mix; and JLL for construction costs and AI infrastructure requirements.
Individual companies are used as case studies where their disclosures make a particular business mechanism unusually visible. Digital Realty provides useful evidence on the pricing difference between smaller deployments and hyperscale leases, utility reimbursements, development promotes and signed backlog. Equinix provides unusually clear data on interconnection economics, customer diversification, profitability and its xScale joint-venture model.
Exceptional quarterly figures are treated as evidence of current market conditions rather than permanent run rates. That applies in particular to extreme renewal pricing, very low vacancy rates and unusually large leasing quarters. Where several independent financial, operational and physical indicators point in the same direction, we give the conclusion more weight.
Key sources used for this analysis include: Synergy Research Group on global hyperscale capacity, the International Energy Agency on data center electricity demand, the IEA on power-infrastructure constraints, CBRE's Global Data Center Trends 2026, CBRE on preleasing and supply scarcity, Digital Realty's Q2 2026 results, Digital Realty's Q1 2026 results, Equinix's Q2 2026 results, Equinix on its U.S. xScale partnership with GIC and CPP Investments, Iron Mountain's Q2 2026 results, JLL's 2026 Global Data Center Outlook, CyrusOne and Constellation on large-scale Texas power procurement, Blackstone's 2026 Mid-Year Investment Perspectives, and Goodman's data center development track record.
This chart, featured in our data center market deck, shows the revenue mix by region across Europe, Asia, North America, Africa, and South America in the data center market
Related blog posts
