A data centre sells IT kilowatts but buys facility kilowatts, and the hottest hour of the year decides how many it can sell.
Leasable IT capacity is the grid connection divided by the peak PUE, not the annual one. On an illustrative 120 MW connection, a facility that markets a 1.25 annual PUE but draws at 1.40 on the hottest hour can lease 85.7 MW, not 96.0. The missing 10.3 MW is $12.96 million a year of rent at $105 per kW-month and about $45 million of development value.
Worked in full in The Data Center Development Handbook by Julian R. Sterling, with every figure reproduced in a free workbook.See the book on Amazon →
A developer has secured 120 MW of utility capacity at the meter for a turnkey campus. The design team quotes a PUE of 1.25, which is the annualised figure: total facility energy over IT energy across a year, flattered by cool nights and mild months. On a design-day afternoon, with chillers and fans at full duty, the same building runs at 1.40. Rent and cost are the turnkey figures the book uses for its Chapter 13 campus. Everything else is illustrative.
| Input | Value |
|---|---|
| Utility capacity at the meter, MW | 120 |
| Annualised PUE (marketing figure) | 1.25 |
| Peak design PUE (hottest hour) | 1.40 |
| Rent, $ per kW-month of IT, net of power | 105 |
| NOI as a share of gross rent | 94.8% |
| Exit capitalisation rate | 7.75% |
| Capital cost, $ million per MW of IT | 11 |
A tenant leases IT kilowatts, but the utility delivers facility kilowatts. Every kilowatt of IT load brings cooling, power conversion losses and lighting with it, and PUE is the ratio between the two. The capacity you can sell is therefore the capacity you can deliver divided by the overhead ratio at the moment the overhead is highest. A lease commits you to that load on every hour of the year, including the worst one.
Leasable IT MW = grid capacity / peak PUE = 120 / 1.40 = 85.7 MW
Annual rent per MW = 1,000 kW × $105 × 12 = $1.26 million
Value created per MW = (rent × NOI share / exit cap) − cost = (1.26 × 94.8% / 7.75%) − 11 = 15.41 − 11 = $4.41 million
In Excel, with grid in B1 and peak PUE in B2: =B1/B2 for leasable MW, then =B3*1000*105*12 for rent.
Run the same arithmetic on the marketing figure and the campus looks larger than it is.
| Line | Sized on 1.25 | Sized on 1.40 | Difference |
|---|---|---|---|
| Leasable IT, MW | 96.0 | 85.7 | 10.3 |
| Overhead load assumed, MW | 24.0 | 34.3 | |
| Annual gross rent | 120.96 | 108.00 | 12.96 |
| Capital cost | 1,056.0 | 942.9 | 113.1 |
| Stabilised value | 1,479.6 | 1,321.0 | 158.5 |
| Value created | 423.6 | 378.2 | 45.4 |
Sized honestly, the connection supports 85.7 MW of IT, 10.7 per cent less than the 96.0 MW the annual figure suggests. Because rent, cost and value all scale with leasable megawatts, value created falls by the same 10.7 per cent, from $423.6 million to $378.2 million. The capital cost falls too, by $113.1 million, which is why the value loss is $45.4 million rather than the full $158.5 million of stabilised value: you do not build what you cannot power.
The worse outcome is the one where nobody runs this check. A developer who signs leases for 96.0 MW and then sees every tenant ramp to full load on a hot afternoon needs 134.4 MW at the meter, 14.4 MW more than the utility has committed. That is not a return haircut. It is a service level breach, or a load-shedding conversation with the utility, or a scramble for on-site generation, each of which costs more than the rent ever earned.
PUE is quoted two ways and they answer two different questions. The annualised figure sets the energy bill. The peak figure sets how many megawatts you can sell. A deck that uses the first to answer the second overstates capacity on every page that follows.
Each step of peak PUE moves leasable capacity, rent and value created together. On the same 120 MW connection:
| Peak PUE | Leasable IT, MW | Annual rent | Value created |
|---|---|---|---|
| 1.15 | 104.3 | 131.5 | 460.4 |
| 1.20 | 100.0 | 126.0 | 441.2 |
| 1.25 | 96.0 | 121.0 | 423.6 |
| 1.30 | 92.3 | 116.3 | 407.3 |
| 1.40 | 85.7 | 108.0 | 378.2 |
| 1.50 | 80.0 | 100.8 | 353.0 |
The relationship is not linear: capacity is grid divided by PUE, so each tenth of a point costs a little less in megawatts as PUE rises. On a connection this size, a tenth of a point costs 7.7 MW between 1.20 and 1.30 and 5.7 MW between 1.40 and 1.50. A cooling design that buys a lower peak, such as liquid cooling for dense halls, should be valued in leasable megawatts first and in energy savings second.
The energy saving is real but smaller, and it mostly belongs to the tenant, since turnkey rent here is net of power. At 70 per cent average utilisation the 85.7 MW campus runs 60.0 MW of IT on average, and every 0.10 of annualised PUE adds $3.68 million a year of overhead energy at an illustrative $70 per MWh. That matters in the tenant's bid. The capacity effect matters in the developer's value.
Most early underwrites take the PUE from the design brief, which is usually the annualised target, and divide the grid capacity by it. The error does not show in the yield on cost, because rent and cost are both per megawatt and fall together. It shows only in the megawatt count, which is the one number the leasing team, the lender and the utility all rely on. Ask the mechanical engineer for the PUE at the design ambient temperature and at full IT load, and size the campus on that figure. Keep the annualised PUE for the tenant's energy model and the sustainability report.
Two refinements push the same way. Utility capacity is often stated before transformer and distribution losses, so the facility sees slightly less than the headline figure. And phased campuses frequently energise a first block before the most efficient cooling plant is installed, so early-phase peak PUE can run higher than the steady-state design.
Divide the grid connection by the peak PUE to find what you can lease. On this case the difference between the two PUE figures is 10.3 MW, $12.96 million of annual rent and $45.4 million of value created. The same discipline, power first and then everything else, runs through the book's three business models on its 180 MW campus, compared in powered shell against turnkey returns. The turnkey rent and cost used here come from the free workbook for the book's campus, where you can rebuild the underwrite on your own connection.
Annual PUE is total facility energy over IT energy across a year and drives the energy bill. Peak PUE is the ratio on the hottest hour at full IT load and drives how much IT capacity a fixed grid connection supports. In the illustrative case the same building runs at 1.25 annually and 1.40 at peak, a gap worth 10.3 MW of leasable capacity on 120 MW.
Divide by the peak PUE. At 1.20 it supports 100.0 MW of IT, at 1.30 about 92.3 MW, at 1.40 about 85.7 MW and at 1.50 exactly 80.0 MW. At $105 per kW-month, each megawatt of IT is $1.26 million of annual rent, so the PUE assumption moves revenue directly.
Not in a turnkey underwrite where rent and capital cost are both per megawatt of IT: they fall together, so the yield on cost barely moves. What changes is scale. In the case, sizing on 1.40 instead of 1.25 cuts value created from $423.6 million to $378.2 million, the same 10.7 per cent as the megawatts.
This article is one calculation from The Data Center Development Handbook. The book takes the same case from first principles to the decision, chapter by chapter, and every figure it prints is a live formula in the free companion workbooks.
Get the book on Amazon →Free companion files
Also on Amazon UK · Amazon Germany · Amazon France · Amazon Canada
Reading guide: infrastructure, data centres and energy → · All 324 articles →
If this book helped, or didn’t, a few lines on Amazon are worth more than they look: they are what the next reader goes on. Write a review. The workbook stays free either way.