How Data Centers Differ From Other Property Types
A data center is a building that houses the hardware that holds and moves digital information. Its income depends on power capacity and specialized operations as well as on the building, which separates it from an office or warehouse. A buyer first needs to know which interests the purchase includes, since enterprise and colocation facilities produce different claims on cash. The second question is how much of the stated capacity operates today and how much is still to be built.
Who owns the income: enterprise and colocation structures
A data center is a building containing hardware that holds and moves digital information. That definition describes the use of the building and says nothing about who collects the income. The ownership structure answers that question, and it varies more here than in most property types.
An enterprise data center is typically constructed and used by one organization. When the builder is also the only user, no third party pays rent, so a buyer of the real estate alone is buying a lease that still has to be created.
A colocation data center is owned and managed by a service provider that leases physical space to customers and supplies power and related services. The provider earns service revenue from many customers. A buyer collects that stream only when the acquisition includes the operating business. A buyer of the building alone collects whatever rent the operator pays the owner.
The wholesale model arose as an alternative to companies building their own facilities.
One label, several different claims on cash.
The valuation consequence is direct. A cap rate is one year of net operating income divided by price, so it fits a stream of rent from a lease. Service revenue from an operating business carries different costs and different risks. Capitalizing it at a property cap rate assigns value to receipts the buyer may have no contractual right to receive.
Megawatts set the capacity of the asset, and status matters
Megawatts of power capacity measure what a facility can support, and square feet measure only the shell. Pricing by floor area looks at the shell and ignores the constraint that tenant needs actually press against.
Capacity also has a status. A platform can report close to 500 megawatts of operating capacity and a 775 megawatt backlog to be built over the next several years, which brings the planned total to nearly 1,275 megawatts. The backlog is larger than the operating base, so the headline total says little about current income. Operating megawatts can be tied to rent being paid. Backlog megawatts depend on the build schedule and on whoever eventually occupies them.
Power risk sits inside the real estate risk. Data centers combine real estate with technology and power infrastructure risks, including rapidly changing tenant needs and potential power constraints. A building that is physically intact can still fail to support the capacity its income model assumes.
Power can also work in an owner’s favor. Data centers can act as a flexible grid resource, leaving the grid and running on backup power when the grid operator directs it. That feature depends on the power arrangement and on what the lease permits, because the tenant often controls the load and the backup generation, and it exists only where the site has the equipment and the permissions.
Some market participants argue that few brokers have the hard technical skills to understand how these assets work, including land assemblage and powering the land. For an underwriter, site control and a power path are part of the asset, and a lease alone describes neither.
“Data centers present a clear example of how investors are misreading fundamental value drivers.”
Why the sector sits between real estate and infrastructure
Investor interest in data centers is linked to growth in the need to store and process data and to interconnect. The sector also blurs the line between real estate funds and infrastructure funds, because digital infrastructure, including data centers and cell towers, can fit either mandate.
The blurring matters for pricing. A buyer from a real estate mandate and a buyer from an infrastructure mandate can look at the same facility against different return targets and different holding assumptions.
Capital has followed. The global data center market has attracted around $25 billion per year of new investment in recent years. Specialist funds exist that both buy and manage existing data centers as well as construct new facilities for tenants.
Allocation data points the same direction. In one sample of sovereign and pension real estate portfolios, data centers rose from virtually nothing to about 14% of the portfolio over the period studied. That is one sample over one period, and it does not describe all investors.
A rising weight shows where capital is flowing. It says nothing about whether a given facility earns its price.
“Specialist funds exist that both buy and manage existing data centers as well as construct new facilities for tenants.”
The operating platform is part of the investment
Data centers and a handful of other alternative sectors often require more sophisticated management platforms, focused on specialized development and operations and on the human capital behind them. In a conventional lease the landlord supplies space and the tenant runs the business. Here the owner or its operating partner may supply a technical service as well.
Deal stage changes the analysis. An existing facility has operating capacity and observable income. Development for a tenant has a construction budget, a delivery date and a bridge from spending to income at completion. Specialist funds do both, yet one deal is observed and the other is modeled, so the diligence differs.
Consolidation adds a market structure point. In colocation, larger operators have been buying smaller data center operators to expand their geographic reach. A seller of a smaller operator may therefore face buyers that value footprint as well as cash flow.
Cost follows the same boundary. Income attributed to specialized operations needs the expense of supplying those operations, measured over the same period. Counting the benefit and leaving out the cost overstates the residual income in the model.
Worked example: a hypothetical 10 megawatt leased facility with 4 megawatts rolling off
This hypothetical example compares a base case and a stressed case on the same capital. A stabilized 10 megawatt facility is bought for $200 million with $2 million of closing costs. A loan of 60 percent of price, or $120 million, carries 6.5 percent interest only. Equity is price plus closing costs minus the loan: $200 million plus $2 million minus $120 million, or $82 million.
Rent is $125 per kilowatt per month, and power and operating costs pass through to the tenant. The landlord keeps $1 million a year of nonrecoverable expense in both cases. The purchase and the loan are identical in both cases, and only the leased megawatts change.
Base case. Ten megawatts is 10 multiplied by 1,000, or 10,000 kilowatts. Annual rent is 10,000 multiplied by $125 multiplied by 12 months, or $15 million. Net operating income is $15 million minus $1 million, or $14 million. The cap rate on price is $14 million divided by $200 million multiplied by 100, or 7 percent.
Annual interest is $120 million multiplied by 6.5 divided by 100, or $7.8 million. Debt service coverage is $14 million divided by $7.8 million, or 1.79 times. Cash after interest is $14 million minus $7.8 million, or $6.2 million, and the cash yield on equity is $6.2 million divided by $82 million multiplied by 100, or 7.56 percent.
Stressed case. A tenant vacates 4 megawatts at its lease end, leaving 10 minus 4, or 6 megawatts leased. Six megawatts is 6,000 kilowatts, so annual rent is 6,000 multiplied by $125 multiplied by 12, or $9 million. Net operating income is $9 million minus $1 million, or $8 million, and the cap rate on the same price is $8 million divided by $200 million multiplied by 100, or 4 percent.
Interest stays at $7.8 million. Coverage is $8 million divided by $7.8 million, or 1.03 times. Cash after interest is $8 million minus $7.8 million, or $200,000, and the cash yield on equity is $200,000 divided by $82 million multiplied by 100, or 0.24 percent.
The sensitivity comes from fixed cost and fixed debt. Rent falls by $15 million minus $9 million, or $6 million, which is 40 percent of base rent. Net operating income falls by $14 million minus $8 million, or $6 million, which is 42.86 percent of the base figure, because the $1 million of expense stays the same. Cash after interest falls by $6.2 million minus $200,000, or $6 million, which is 96.77 percent of the base figure, because the interest bill stays the same too.
Lenders set coverage and debt yield thresholds differently, so the loan terms in hand govern whether either case passes. The comparison shows that a loss of 40 percent of rent removes nearly all equity cash flow.
Pitfalls and mistakes in data center underwriting
Capitalizing an operator’s service revenue at a property cap rate is an ownership boundary error. Colocation revenue belongs to the provider that leases physical space to customers and supplies power and related services. A property buyer is entitled only to the rent that provider pays, and correct analysis models that rent alone unless the operating business is in the deal.
Adding backlog to operating capacity in a single total is a status error. The arithmetic of 500 plus 775 is right, and the combined figure still misleads when income is attached to all of it. Correct analysis keeps operating megawatts and megawatts under construction on separate lines with separate timing and funding assumptions.
Pricing by floor area measures the wrong variable. A data center combines real estate with power and technology risk, and tenant needs can change quickly. Two buildings with the same square footage can support very different usable power, so comparables need to be normalized to capacity and power availability.
A cap rate is one year of net operating income divided by price. It does not measure a yield, a return or an IRR over a hold, so quoting it as any of those is an arithmetic error. In the example the cap rate falls from 7 percent to 4 percent when 4 megawatts leave, and that single year figure says nothing about re-leasing time or exit value.
Treating a coverage number as a universal pass mark misreads how lenders work. The example shows 1.79 times in the base case and 1.03 times in the stressed case on the same loan. Lender thresholds differ, so the useful test is how far net operating income can fall before the loan terms in hand are breached.
Sector momentum is a weak substitute for fundamentals. Some institutional authors argue that data centers show how investors can misread fundamental value drivers. A rising allocation weight explains why buyers compete, and it leaves open whether the price rests on contracted income or delivered power.
Questions for a data center investment checklist
Which interests and associated income streams does the purchase include across the real estate, the operating business and any development rights?
Is the facility an enterprise facility or a colocation facility, and who pays rent or earns service revenue under that structure?
How many megawatts are operating with income in hand, and how many are still to be built with income dependent on construction?
Do site control and a confirmed supply connection support the stated capacity and allow the building to deliver what the income model assumes?
Which modeled receipts are rent and which are operating revenue, and do their contractual recipients match the interests being valued and acquired?
What happens to coverage and cash yield if part of the capacity goes vacant, given that fixed expense and fixed interest turn the example’s 40 percent drop in rent into a much larger drop in cash flow?
Does income credited to specialized operations account for the matching expense and the cost of the human capital behind the operating platform?
Which return target applies, real estate or infrastructure, and how might a fund that can invest in either category price the asset differently from a buyer with a single mandate?