Cap Rates, Yield on Cost, and Your Actual Return

Three different numbers get called "the yield" in data center deals, and they differ by a factor of two. What each one means, where the exit cap rate hides, and what the largest data center REIT discloses about what private buyers are paying.

Three numbers in a data center deal all get called "the yield", and they are not the same number. Confusing them is the most expensive arithmetic error available to a retail investor in this category, and the confusion is easy because the offering material rarely labels which one it is showing you.

The three numbers

1. Cap rate (capitalisation rate). Stabilised net operating income divided by the value of the asset.

cap rate = NOI ÷ property value

This is a market-priced number. It describes what buyers are paying for income today, and it is the number used to value a stabilised, leased building. A lower cap rate means a higher price for the same income.

2. Yield on cost. Stabilised NOI divided by what it cost to build.

yield on cost = stabilised NOI ÷ total project cost

This is a development number. It is normally higher than the market cap rate, and the gap is the development profit — the reward for taking construction, leasing and power risk. If a project costs $100 million, stabilises at $8 million of NOI, and finished buildings in that market trade at a 6.5% cap rate, the yield on cost is 8%, the finished value is about $123 million, and the development spread is where the money is made.

3. Your return. What reaches you after fees, sponsor units, financing and timing.

your return ≈ (your share of NOI − fees) ÷ what you paid, adjusted for when the cash actually arrives

A private offering's advertised percentage is almost always number 2, sometimes dressed as number 1, and never number 3. See why the headline is not your return for the six deductions between them.

Why the spread, not the yield, is the deal

A development is worth doing when the yield on cost comfortably exceeds the cap rate at which the finished asset can be sold or refinanced. That difference — usually called the development spread — is the compensation for every risk the project has not yet retired.

So the honest way to read an 11%, 15% or 17.5% target is not "that is a high yield". It is: what does a finished, leased building in this market yield, and is the extra enough for construction risk, leasing risk, power risk, and the illiquidity of holding one building for years? If a finished asset yields 6% and a development yields 8%, two points is thin. If a development yields 17% against the same 6%, the gap is wide — and the obvious next question is which risks are so large that the market demands eleven points to take them.

The exit cap rate: the hidden assumption in every IRR

Any projection that shows a total return, an IRR or a multiple rather than only an annual yield has assumed a price at which the building is eventually sold. That assumption is an exit cap rate, and it is usually in a footnote or nowhere.

It matters enormously. At $8 million of NOI, a 6% exit cap rate gives a $133 million building; a 7.5% exit cap rate gives $107 million. A 1.5-point move in an assumption nobody discussed has removed $26 million — on a project of this size, plausibly most of the equity profit.

Two questions, both answerable:

1. What exit cap rate does the model assume, and why that one?
2. What does the return become at one point higher? A sponsor who has run that sensitivity will have the number. One who has not is showing you a single line and calling it a forecast.

What the market actually shows: the operator's own words

Transaction cap rates for data centers are not published in any free, citable source, and we are not going to quote a number we cannot show you. What is public is what the largest listed operator tells its shareholders about pricing, and it is directly relevant to anyone being offered a single-asset private deal.

Digital Realty, in its annual report for 2025, explaining why it expects to buy fewer individual buildings:

"the growing acceptance by private institutional investors of the data center asset class has generally pushed capitalization rates lower, as such private investors may often have lower return expectations than us. As a result, we anticipate near-term single asset acquisitions activity to comprise a smaller percentage of our growth while this market dynamic persists."

>

— Digital Realty 10-K, FY2025

Read that carefully, because it is a professional buyer's assessment of the market you are being invited into. Lower cap rates mean higher prices for the same income. The company that owns hundreds of these buildings is saying that private capital is paying more than it is willing to, and that it is therefore standing back from single-asset purchases.

That is not a reason to avoid the category. It is a reason to be precise about price, and to be suspicious of any argument that rests on data centers being cheap.

And prices can fall: three years of write-downs, in filings

The same company has recorded a provision for impairment on real estate investments in each of the last three years, each time on properties in secondary U.S. markets:

Figures from the consolidated income statement and the notes of the FY2025 10-K.

This is the most useful fact on this page. In the middle of the strongest demand cycle this asset class has seen, a best-in-class operator has written down data center property three years running — specifically in secondary markets. "AI demand is enormous" and "this particular building in this particular location will hold its value" are different propositions, and the filings show the gap between them. It bears directly on how to think about a project located outside a primary market: see data center markets.

How the professionals value an unleased building

Worth knowing, because it is the arithmetic that applies to a project before its tenant arrives. From the same filing's accounting policy: operating properties are valued "as if they are vacant" using "an income approach methodology using either a discounted cash flow analysis or by applying a capitalization rate to the estimated Net Operating Income ... of a property", and the as-if- vacant value considers "estimated carrying costs during expected lease-up periods and costs to execute similar leases (based on current market conditions)".

Three things an investor should take from that:

1. Value comes from NOI and a cap rate, not from cost. Spending $78 million does not make a building worth $78 million.
2. Lease-up costs money. Carrying costs during the period before a tenant pays — taxes, insurance, operations — are a real deduction from value, and they are exactly the costs an unleased development incurs.
3. "Current market conditions" means the conditions when the building is finished, not when the offering was written.

The three-line summary

• A cap rate prices leased income; a yield on cost prices a development; your return is neither, and is lower than both.
• The spread between yield on cost and market cap rate is the compensation for unretired risk. Judge the deal on the spread, not the headline.
• The exit cap rate is the assumption with the most leverage over any total return, and it is the one least often stated.

What to read next: returns and the six leaks for what happens between NOI and you, and what a hyperscaler lease contains for the terms that determine NOI in the first place.

Nothing here is investment advice or a recommendation of any security. See how this site is funded.