Phasing costs more per megawatt and pays it back in deferred capex, avoided idle carry and an option not to build. Absorption speed decides which wins.
Build in phases when leasing is slower than one phase a year, and in one go when the campus is pre-let. On an illustrative 120 MW campus, three 40 MW phases cost $1,340 million against $1,260 million built at once, 6.3 per cent more, yet with 40 MW leased every two years they add $80 million of NPV and lift the unlevered IRR from 12.5 to 14.5 per cent. Pre-let, the single phase wins by $131 million.
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 single-phase build buys scale: one mobilisation, one set of procurement contracts for switchgear, generators and chillers, one commissioning programme. A phased build gives that up. It puts the land, the substation and the campus utilities in first, then adds data halls in blocks as tenants sign. Each block costs more per megawatt, but the money goes out later and nothing sits finished and empty. Which effect is larger depends on one variable the developer does not control: how fast the capacity leases.
| Input | Single phase | Three phases |
|---|---|---|
| Capacity (MW of IT load) | 120 | 3 × 40 |
| Shared works up front (land, substation, utilities) | included | 200 |
| Cost per phase | n/a | 380 ($9.5m per MW) |
| Total capex | 1,260 ($10.5m per MW) | 1,340 ($11.17m per MW) |
| Build time | 2 years | 2 years per phase |
Both plans earn $1.25 million of NOI per leased MW a year. Capacity that is built but not leased costs $0.08 million per MW a year to hold: staff, maintenance, insurance and the fixed part of the power bill. The campus is valued at the end of year 10 at a 6.5 per cent cap rate, and cash flows are discounted at 8 per cent. In the base case the market absorbs 40 MW in year 2, another 40 in year 4 and the last 40 in year 6.
The single-phase plan spends $630 million in each of years 0 and 1 and has 120 MW ready in year 2. Only 40 MW is leased, so year-2 NOI is 40 × 1.25 − 80 × 0.08 = $43.6 million: $50.0 million of rent less $6.4 million of carry on the empty halls.
The phased plan spends $390 million in year 0 (the $200 million of shared works plus half of phase 1) and $190 million in year 1. Each later phase is started two years before it is needed, so its $380 million is spent in the two years before the next 40 MW is due to lease.
| Year | 0 | 1 | 2 | 3 | 4 | 5 | 6 to 9 | 10 |
|---|---|---|---|---|---|---|---|---|
| Single phase | −630.0 | −630.0 | 43.6 | 43.6 | 96.8 | 96.8 | 150.0 | 2,457.7 |
| Three phases | −390.0 | −190.0 | −140.0 | −140.0 | −90.0 | −90.0 | 150.0 | 2,457.7 |
NPV = Σ CFt / (1 + 8%)t; Excel: =B2+NPV(8%,C2:L2) and =IRR(B2:L2)
Single phase: NPV $472 million, IRR 12.5%
Three phases: NPV $552 million, IRR 14.5%
From year 6 the two plans are identical: same 120 MW, same NOI of $150.0 million, same exit. The whole difference sits in years 0 to 5, and it has two sources.
Together that is $79.8 million, the $80 million gap. Turned round: in this market the phased plan stays ahead until each phase costs $412.0 million, $10.30 million per MW, a total of $1,436 million or 14.0 per cent above the single build.
| Absorption | Single IRR | Phased IRR | Single NPV | Phased NPV | Phasing gain |
|---|---|---|---|---|---|
| Pre-let, 120 MW in year 2 | 15.2% | 14.3% | 723.2 | 592.6 | −130.6 |
| 40 MW a year | 13.7% | 14.3% | 589.7 | 592.6 | 2.9 |
| 40 MW every two years (base) | 12.5% | 14.5% | 472.2 | 552.0 | 79.8 |
| Demand stops at 80 MW | 7.4% | 12.9% | −51.5 | 328.8 | 380.3 |
The table is the decision. If a hyperscaler has signed for the whole campus, phasing only delays rent on 80 MW and costs more, and it loses $131 million. At one phase a year the two plans are level, $2.9 million apart, so the premium is almost exactly paid for by the deferral and the avoided carry. Below that pace, phasing wins, and if demand stalls it wins by a mile: the single plan is left with 40 MW, $420 million of capex at $10.5 million per MW, that never earns, while the phased plan simply does not build its $380 million third phase and spends $960 million in total. Phasing is an option, and options are worth most when the outcome is uncertain.
The phased plan only works if the shared works are sized for the full campus from day one. A substation or grid connection sized for 40 MW turns each phase into a new interconnection request, and the queue, not the construction programme, sets the pace.
Comparing cost per MW and stopping there. On that test the single build is cheaper, $10.5 million against $11.17 million, and the phased plan looks like a 6.3 per cent penalty. The comparison that matters is the present value of capex plus the carry on empty halls, set against the leasing curve the developer actually believes. The second mistake is the opposite one: phasing a campus that is already pre-let because phasing sounds prudent, and giving up $131 million to hedge a risk that the lease has already removed.
The levered version of the same build, with a construction loan and an exit, is set out in how to calculate the levered IRR of a data centre development, and the business-model choice that sits beside this one is in powered shell vs turnkey. The book's own underwrite, three business models on one site, is rebuilt in the free workbook for this case.
Find the absorption rate first. If tenants will take more than one phase a year, or the campus is pre-let, build at once and bank the scale. If leasing is slower or uncertain, phase it: on this campus a 6.3 per cent build premium bought $80 million of NPV in the base case and $380 million of protection if demand stops.
It depends on how much shared infrastructure is built up front and what scale is lost per phase. In the illustrative case a single 120 MW build costs $10.5 million per MW, $1,260 million, while $200 million of shared works plus three $380 million phases total $1,340 million, $11.17 million per MW, a premium of 6.3 per cent. In the base case phasing still wins up to a 14.0 per cent premium.
When the capacity is leased before it is finished. If all 120 MW of the illustrative campus is pre-let to deliver in year 2, building at once gives a 15.2 per cent unlevered IRR against 14.3 for phases built back to back, an NPV gap of $131 million, because phasing delays rent on two thirds of the campus and costs more.
Demand that never arrives. If leasing stops at 80 MW, the single-phase campus carries 40 MW, $420 million of capex, that earns nothing and costs $0.08 million per MW a year to hold, and its IRR falls to 7.4 per cent. The phased plan never builds phase 3, spends $960 million and keeps a 12.9 per cent IRR.
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 1501 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.