Appendix A4 · Method for §3 of the main paper · May 2026
How the numbers were tested
The corner tables in Section 3 are the decision-relevant output of the testing. This note records the method: the envelope tested, and what the model recomputed against the source pack. Nothing here changes the decision.
The envelope tested
Four inputs drive Vehicle A’s economics: capex per megawatt, the service-tier premium, the anchor pre-commitment, and the ramp. Vehicle B adds the operator pair: colocation price and stabilised utilisation. The May 2026 calculator exposed the first four as sliders; the underwriting model retired the sliders and replaced them with explicit recomputed grids in which every cell shows its full formula and can be stepped through cell by cell.
| Input | Central | Envelope tested | Where the corners sit |
|---|---|---|---|
| Capex per MW | AUD 45 m | AUD 35 to 60 m | Model Sensitivity grid; §3 corner 1 at 60 |
| Service-tier premium | +18% | +5 to +35% | Model Sensitivity grid; §3 corner 2 at +5 |
| Anchor pre-commitment | 60% | 0 to 100% | §3 corner 4 at 90%, with +25% premium |
| Ramp to full utilisation | 4 yrs | 2 to 6 yrs | §3 corner 3 at 6 |
| Colo price · Vehicle B | A$800/kW/mo | A$500 to 1,000 | Model operator grid, five price rows |
| Stabilised utilisation · Vehicle B | 78% | 60 to 90% | Model operator grid, five utilisation columns |
The corner outcomes themselves sit in the main paper’s Section 3 tables and are deliberately not repeated here; one set of published figures, one place to correct them.
What the model recomputed
The retired calculator displayed two headline outputs at the central case, both carried faithfully from the source pack, and neither survived the model rebuild unchanged.
The first was an unlevered yield of 8.7 per cent. The pack’s own figures give 9.2: NOI of AUD 116 m on AUD 1,260 m of deployed capital. Reaching 8.7 requires assuming roughly 94.5 per cent revenue realisation, a spot-pricing discount the pack describes in its method note but never reconciles to its headline. The model carries 9.2 and flags the difference rather than correcting it silently.
The second was a levered IRR of 12.7 per cent. On the pack’s own stated method the figure is (NOI less interest) over equity: a stabilised cash-on-cash, not a time-weighted IRR. The calculator’s own methodology note conceded the point, in its words: “levered IRR proxies steady-state cash-on-cash”. The model promotes that concession to the label. It reports the measure as cash-on-cash, 12.5 per cent on exact inputs, and computes the true 30-year levered IRR separately, about 8.8 per cent, on the full cash-flow row including ramp, refresh and exit.
Two further recomputations complete the set: the source specifies no exit cap rate (the model uses a flagged 6.5 per cent analyst assumption, affecting the 30-year IRR only), and the source’s +5 per cent corner cites a ~6.5 per cent yield where its own revenue figure implies 8.2 (the case still fails there, on premium economics). All four appear with arithmetic in the main paper’s §6.E and on the model’s Sources & Notes tab.
The calculator’s other live readouts, year-three NOI and contracted revenue mix at signing, were working states of its ramp arithmetic rather than underwriting anchors; they retire with it.
Main paper · The four variances, tabled: §6.E
The customer mix and the thesis map
The calculator’s final panel mapped whatever customer mix was selected onto the four global theses. The map itself is stable and worth keeping:
| Thesis | What activates it |
|---|---|
| Hyperscaler captive | Enterprise revenue persistence; Copilot-class seat economics |
| Neocloud merchant | Lab counterparty quality; GPU residual value; merchant pricing |
| REIT / infrastructure fund | Power and grid scarcity; tenant credit; long-tenor leases |
| Sovereign / strategic | A government anchor; concessional capital; political continuity |
At the central mix, the seven boutique segments with a 60 per cent anchor pre-commitment, the map lands on REIT / infrastructure with a sovereign adjacency on the government-mid segment: the same placement the screen in Appendix A3 reaches by its own route.