Annual consumption is capacity times three factors: how much of it runs, how much the building adds on top, and how many hours there are in a year.
Multiply the critical IT capacity by its average utilisation, by the annual average PUE and by 8,760 hours. An illustrative 100 MW of IT running at 70 per cent utilisation with an annual PUE of 1.30 uses 797,160 MWh a year, about 0.80 TWh, which costs roughly $55.8m at $70 per MWh. The same campus at full load and its peak PUE would use 1,226,400 MWh; the gap is why the inputs matter more than the headline megawatts.
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 →
The question gets asked by utilities sizing a connection, by investors pricing a power purchase agreement, by tenants comparing pass-through costs, and by planners worried about what a campus does to the local grid. All of them need the same four numbers, and most published estimates hide at least one.
| Input | Value |
|---|---|
| Critical IT capacity | 100 MW |
| Average IT utilisation over the year | 70% |
| Annual average PUE | 1.30 |
| Peak PUE (hottest hours) | 1.40 |
| Hours in a year | 8,760 |
| All-in energy price (illustrative) | $70 per MWh |
Two of these are routinely confused. The annual average PUE drives energy; the peak PUE drives the size of the grid connection. And utilisation is the average share of contracted IT capacity that tenants actually draw, which for a cloud campus is well below 100 per cent and for a dense AI training hall can sit close to it.
IT energy (MWh) = IT capacity (MW) × utilisation × 8,760
Facility energy (MWh) = IT energy × annual average PUE
Annual cost = facility energy × price per MWh
In Excel: =IT_MW*Util*8760*PUE, then =B6*Price. If you model monthly, use the hours in each month and a monthly PUE, because cooling overhead is seasonal.
Data centre rents are quoted per kW of IT capacity per month, so put the power bill on the same basis: $55,801,200 ÷ 100,000 kW ÷ 12 = $46.50 per kW-month. Against an illustrative turnkey rent of $105 per kW-month, the power bill adds another 44 per cent on top of the rent. That is why leases pass power through at cost or with a fixed PUE, and why who carries a PUE miss is a negotiated term rather than an engineering detail.
Utilisation matters for the grid too. The facility averages 91 MW against a design maximum of 140 MW at full load and a 1.40 PUE, a load factor of 65.0 per cent on that maximum. Demand charges are set by the metered peak, which sits somewhere between the two. How that ratio changes the effective price per MWh is worked in the demand charge per MWh by load factor.
| Utilisation | PUE 1.20 | PUE 1.30 | PUE 1.50 |
|---|---|---|---|
| 50% | 526 | 569 | 657 |
| 70% | 736 | 797 | 920 |
| 90% | 946 | 1,025 | 1,183 |
| Price per MWh | Annual cost | Per kW-month |
|---|---|---|
| $50 | $39.9m | $33.21 |
| $70 | $55.8m | $46.50 |
| $90 | $71.7m | $59.79 |
Utilisation moves consumption more than PUE does across realistic ranges: from 50 to 90 per cent utilisation the answer rises by 80 per cent, while from a 1.20 to a 1.50 PUE it rises by 25 per cent. A campus leased to AI training tenants that run near full load will use far more energy per MW than the same campus leased to enterprise cloud.
Is 100 MW the IT load or the connection? If the 100 MW is the utility connection, sized for a 1.40 peak PUE, only 71.4 MW of IT fits, and at the same utilisation and average PUE the campus uses 569,400 MWh, not 797,160. The trade-off between PUE and leasable capacity on a fixed connection is in how a higher PUE limits leasable capacity.
The common mistake is nameplate times 8,760: 100 MW × 8,760 = 876,000 MWh. Here it lands only 9.9 per cent above the answer, but by accident, because it leaves out utilisation, which lowers the figure, and PUE, which raises it, and the two roughly cancel at these inputs. At 90 per cent utilisation and a 1.50 PUE the same shortcut would be badly short; at 50 per cent and 1.20 badly long. The second mistake is using the design or peak PUE for annual energy, which overstates the bill in every month but the hottest.
Energy is IT capacity times utilisation times average PUE times 8,760: 797,160 MWh and about $55.8m a year for this 100 MW campus, or $46.50 per kW-month beside the rent. State all four inputs, and say whether the megawatts are IT or grid. The free workbook for this book carries the campus underwrite, and the data center financial model template lays out the power line monthly.
At full load with no overhead, 8,760 MWh, one MW for every hour of the year. In practice multiply by average utilisation and annual PUE: at an illustrative 70 per cent and 1.30, each MW of IT capacity uses 7,972 MWh a year. At 90 per cent utilisation and a 1.50 PUE it would be about 11,830 MWh.
Ask, because the answer moves consumption by almost 30 per cent. If 100 MW is the utility connection sized for a 1.40 peak PUE, only 71.4 MW of IT can be installed, and at 70 per cent utilisation and a 1.30 average PUE the campus uses 569,400 MWh a year, not 797,160.
Each 0.10 of annual PUE is 10 per cent of the IT energy. At 613,200 MWh of IT energy a year, cutting PUE from 1.30 to 1.20 saves 61,320 MWh, about $4.3m a year at an illustrative $70 per MWh. Whether that saving belongs to the landlord or the tenant depends on how the lease passes power through.
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 453 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.