About this paper
You have read the case for building. This paper asks the opposite question: if the things that can go wrong financially do go wrong, together and at once, does the investment survive?
This is a companion to the main paper, built on the same underwriting model. Nothing in that model was edited. Every stressed figure here lives in a new workbook, Gowings_Datacentre_Underwriting_Model__STRESS_v1.xlsx, which recomputes each scenario from the model's own central inputs. Where a source figure looks wrong it is flagged with the arithmetic, never changed. No new facts, counterparties or market data were introduced; the only new inputs are stressed values of inputs the model already carries, plus three flagged analyst assumptions noted in Section 7.
The method is a hard-constraints pass: each assumption the case rests on is pushed to a bad but plausible value, one at a time and then in combination, and the equity return, interest cover and cash drawdown are read off at each point. The verdicts use the model's own thresholds, defined in Section 1.
RulingThe prospect survives as a staged commitment. It does not survive as an unconditional build.
The ruling is carried by three findings below, each traceable to the stress workbook. The decision remains yours.
First, no single financial failure sinks either vehicle. The building keeps covering its senior interest down to 39% occupancy (Vehicle A) or 46% utilisation (Vehicle B). Single misses wound returns; none of them, alone, reaches the equity's solvency.
Second, combinations kill, and they kill the recommended vehicle first. A world in which each key input misses moderately at the same time takes Vehicle B from 10.6% to 2.1% cash-on-cash and breaches a 1.5x interest-cover test. The same world leaves Vehicle A at 8.0%. Under a worst-case lens the vehicle ranking reverses: the operator vehicle owns the higher ceiling, the landlord vehicle owns the floor.
Third, every killing combination is measurable before the money is spent. Premium, pre-commitments, contracted power terms, tendered capex, debt terms: each is observable at the gates. The strongest mitigation in this paper is not an operating fix; it is sequencing. Capital commits only behind the six gates and the conditions in Section 8.
The verdicts below use the model's own pass marks, not new ones invented for this paper.
| Measure | Test | Source |
|---|---|---|
| Levered cash-on-cash | Clears at 8% or above; marginal from 4%, proceed only with structural protection; fails below 4% | Model, Assumptions B.8 |
| Interest cover (DSCR) | Stress benchmark 1.5x; below it a typical senior covenant is breached and the lender takes control of cash | Analyst assumption, flagged; no covenant exists in the model |
| Equity at risk | Deepest cumulative equity outflow before the project turns cash-positive | Stress workbook, IRR Scenarios tab |
Thresholds: Assumptions!C118 (8%) and C119 (4%), defined for Vehicle B; applied to both vehicles here for comparability and flagged as such (Section 7). Central case for reference: Vehicle A 12.5% cash-on-cash on A$1,260M deployed; Vehicle B 10.6%.
Recommendation first: Vehicle B remains investable only if its two load-bearing inputs, utilisation and colo price, are contractually protected before FID. Unprotected, it is the faster vehicle to fail.
| Stress | EBITDA to Gowings | Levered c-o-c | Verdict |
|---|---|---|---|
| Utilisation 60% (central 78%) | A$73.1M | 4.6% | Marginal |
| Colo price A$650/kW/mo (central A$800) | A$80.2M | 5.7% | Marginal |
| Colo price A$500/kW/mo | A$49.6M | 0.9% | Fails |
| Power A$200/MWh, no pass-through (central A$130) | A$92.1M | 7.6% | Marginal |
| Senior coupon 9.0% (central 7.0%) | A$110.7M | 8.6% | Holds |
| JV split worsens to 70/30 (central 80/20) | A$96.8M | 8.4% | Holds |
Source: stress workbook, VehB Stress tab (live formulas from Inputs). Bands per Section 1. The power stress retires entirely if metered power pass-through is written into every colo contract; that condition appears in Sections 6 and 8.
| Stress | Levered IRR | Deepest equity drawdown |
|---|---|---|
| Exit at 10x EV/EBITDA (central 16x); exit equity A$477M vs A$1,141M | 7.3% | A$673M |
| Energisation two years late (transformer lead) | 6.5% | A$761M |
| Exit written to zero: asset worthless at year 30, debt repaid | 5.1% | A$673M |
Source: stress workbook, IRR Scenarios tab. The central row ties to the main model's Operator (JV colo)!C44. Delay years carry interest with no revenue: about A$44M per slip year.
The reading is plain. Vehicle B holds against financing stresses and against any one commercial miss, but its two load-bearing inputs sit close together: a price miss and a utilisation miss are the same customer decision arriving twice. Section 4 prices what happens when they arrive together.
Recommendation first: Vehicle A is the downside-resilient structure. Its worst single failures dent the return; almost nothing it faces alone pushes it below the marginal line.
| Stress | Unlevered yield | Levered c-o-c | Verdict |
|---|---|---|---|
| Premium +5% (central +18%), v3 Corner 2 | 8.2% | 10.3% | Thesis fails |
| Premium 0%: sovereign demand never shows | 7.8% | 9.4% | Thesis retired |
| Capex A$60M/MW (central A$45M), v3 Corner 1 | 6.9% | 7.4% | Marginal |
| Senior coupon 8.5% (central 6.5%) | 9.2% | 10.1% | Holds |
Source: stress workbook, VehA Stress tab. The premium rows carry v3's own corner verdicts, and they deserve a sentence: at +5% the arithmetic still shows 10.3% cash-on-cash, yet v3 rules the case failed, because a boutique that cannot earn its premium is a subscale commodity landlord competing against operators with cheaper capital. The number holds; the reason to build does not. That is a thesis failure, not a solvency failure, and it is fully observable at the premium gate before FID.
| Stress | Result | Reading |
|---|---|---|
| Anchor defaults year 5; 18 months dark; re-let at baseline rent | A$63.0M/yr burn; then 9.4% c-o-c | Wounded, survivable: roughly A$95M of equity burnt while dark (fixed opex plus interest), and the re-let case is the Premium 0% row |
| Exit cap rate 8.0% (central 6.5%): terminal value A$1,450M vs A$1,784M | IRR 8.4% | Wounded; exit timing flexes on a 30-year hold |
| Refresh capex doubles to A$360M | IRR 8.2% | Holds |
| Energisation two years late | IRR 7.0%; drawdown A$690M | Holds; carry is A$45.0M per slip year |
| Exit written to zero: asset worthless at year 30, debt repaid | IRR 5.8% | The harshest exit assumption available; payback (~11 yrs) never depended on the exit |
Source: stress workbook, VehA Stress and IRR Scenarios tabs. Central IRR row ties to the main model's Cash Flow & Returns!C25. Anchor-default mitigations (guarantee, parent covenant, step-in) are priced in Section 6.
One world, both vehicles. Each leg is a moderate, plausible miss, not a catastrophe: demand softer, price softer, build dearer, debt wider, power dearer. No single leg would sink either vehicle on its own.
| Line | Vehicle A | Vehicle B | B + power pass-through | B + pass-through + 40% LTV |
|---|---|---|---|---|
| Deployed capital | A$1,400M | A$1,400M | A$1,400M | A$1,400M |
| Operating result to Gowings | A$108.1M | A$70.4M | A$75.7M | A$75.7M |
| Senior interest | A$57.8M | A$56.0M | A$56.0M | A$44.8M |
| Levered cash-on-cash | 8.0% | 2.1% | 2.8% | 3.7% |
| Interest cover | 1.87x | 1.26x | 1.35x | 1.69x |
| Verdict | Marginal | Fails | Fails | Fails |
Source: stress workbook, Adverse World tab. Legs: Vehicle A premium +10%, capex A$50M/MW, coupon 7.5%; Vehicle B utilisation 70%, colo price A$700/kW/mo, power A$150/MWh, capex A$50M/MW, coupon 8.0%. Under NNN the power leg belongs to the tenant, which is itself part of the finding.
This table is the paper. Vehicle A walks through the adverse world at 8.0%, on the clears line, with 1.87x interest cover. Vehicle B falls to 2.1% and a 1.26x cover: below a 1.5x covenant, the lender controls distributions long before insolvency, and the equity is trapped rather than lost. The two contractual mitigations recover solvency, not returns: power pass-through lifts B to 2.8%, and sizing debt at 40% LTV holds cover at 1.69x, but the return stays below the 4% line in every mitigated form.
The worst case therefore reverses the recommendation's emphasis. The main paper recommends Vehicle B for the margin it owns and the ceiling it keeps. This paper adds the condition: B's floor must be bought contractually before FID, or the downside belongs to Vehicle A's structure.
Break-even algebra at central pricing, from the model's own cost structure. These are the numbers to watch during ramp, not after it.
| Floor | Utilisation | Meaning |
|---|---|---|
| Clears line (8% c-o-c) | 70% | Below this the case no longer clears its own hurdle |
| Marginal line (4% c-o-c) | 58% | Below this the case fails outright |
| Covenant line (DSCR 1.5x) | 57% | Below this a typical lender takes the cash keys |
| Cash break-even after interest | 46% | Below this the equity subsidises the building |
| EBITDA break-even | 25% | Below this the building loses money before debt |
| Floor | Level | Meaning |
|---|---|---|
| Occupancy where NOI just covers interest | 10.9 MW of 28 (39%) | Interest covered at little more than a third full |
| Rent where NOI just covers interest, at 28 MW | A$2.06M/MW/yr | 61% below the anchor rate |
| Central interest cover | 2.57x | Vehicle B central: 2.51x |
Source: stress workbook, Floors tab (live formulas). The Vehicle B clears and marginal floors are this paper's arithmetic for the retire and fail lines; note the open variance on those figures flagged in Section 7.
Every mitigation lands in a contract, a gate or the capital structure. None is an operating hope.
| Risk | Mitigation | Where it lands | What it costs | What remains |
|---|---|---|---|---|
| Power price (B) | Metered power pass-through in every colo contract | Power contract gate | Slightly lower headline colo rate | PUE overrun stays with the owner; ~A$1.3M/yr at the NABERS floor |
| Utilisation shortfall (B) | Anchor pre-commitments before FID, mirroring Vehicle A's 60% test; minimum-take clauses | Anchors and JV terms gates | Rate concession to anchors | Churn drag, currently unmodelled: ~1.7pp of c-o-c (Section 7) |
| Colo price compression (B) | Term contracts with CPI escalators signed pre-FID; never underwrite spot | Anchors gate | Caps repricing upside | Renewal repricing at years 5 to 10 |
| Capex overrun | Fixed-price EPC with liquidated damages; contingency already A$105M in the stack | Site gate; FID re-price on tender | EPC risk premium | Scope change; FX on imported plant (~A$28.5M per 10% AUD move, flagged) |
| Interest rates | Hedge the full drawn balance at FID, tenor through ramp plus five years | Funding workstream | Swap or cap premium | Refinance beyond hedge tenor |
| Construction delay | Transformer and grid works pre-ordered before FID; completion guarantee; LDs | Site gate | Deposits ahead of FID | ~A$44M to A$45M carry per slip year |
| Anchor default (A) | Parent covenant, 12 to 24 month bank guarantee, step-in rights | Anchors gate | Nothing material | Re-let at baseline: c-o-c 9.4% |
| Exit compression | Underwrite the hold, not the sale: payback ~11 years without terminal value | Exit gate | Foregone exit-timing upside | Zero-exit IRR 5.8% (A) / 5.1% (B) |
| Equity and funding | Equity fully committed, with completion support, before FID; staged break points | Source-of-funds workstream (open) | Commitment fees | Partner concentration; the one exposure with no operating fix |
| Covenant breach (B) | Size senior debt to the adverse world: 40% LTV holds cover at 1.69x there | Funding workstream | 0.6pp of central c-o-c (10.6% to 10.0%) | Adverse-world return still below 4% |
Sources: stress workbook tabs as cited in Sections 2 to 5. The equity line is the largest unpriced exposure in this paper: a funding partner withdrawing mid-build leaves a part-built asset, and no line in the model prices a distressed part-built sale. It is flagged, not modelled, and it is why the committed-equity condition sits first in Section 8.
A stress paper is honest about the stresses it cannot compute. Seven flags, each with its arithmetic, none silently fixed.
F-S1 · No construction period. The model books capex at year 0 and first revenue at year 1. Transformer lead times run 2 to 3 years. Each slip year costs about A$45M (A) or A$44M (B) in interest before the first dollar of revenue, so the central IRRs are flattered. The delay rows in Sections 2 and 3 quantify.
F-S2 · Churn is an input to nothing. The model carries a 10% annual churn assumption that no formula reads. A 10% churn with a six-month re-let is roughly a five-point utilisation drag: about A$10.4M of EBITDA to Gowings, ~1.7pp of cash-on-cash. Correcting this is a model-first edit for a future round; flagged here, not changed.
F-S3 · An open variance on Vehicle B's own floor. The model's Summary tab says the operator case retires below about 65% utilisation and fails below about 55%; the sensitivity grid and the main paper say about 70% and 60%. This paper's algebra at central price puts the lines at 70% and 58%, siding with the grid. The variance is open for adjudication; no cell was changed.
F-S4 · PUE sits below the regulatory floor. The model assumes PUE 1.35; the NABERS 5-star floor is 1.4. At 1.4 the penalty is about A$1.3M a year of EBITDA to Gowings. Small, and carried.
F-S5 · The two exits are on different bases. Vehicle A exits on a property cap rate, Vehicle B on an EV/EBITDA multiple. The exit stresses in this paper are each on the vehicle's own basis; they are not directly comparable, and neither exit is load-bearing given the hold-forever rows.
F-S6 · No currency line. The import-heavy capex lines total A$570M at central. At an assumed 50% import share, a 10% AUD depreciation adds about A$28.5M to deployed capital. The import share is an analyst assumption, flagged.
F-S7 · One tenant stands for all of them. The anchor-default stress treats a single tenant as 100% of Vehicle A revenue. Real tenancy mix is a diligence output; the stress is conservative on purpose.
All seven flags, with arithmetic, live on the stress workbook's Sources & Flags tab. Flagging without fixing follows the engagement's variance law.
The prospect survives the worst case the model can price, on one reading and six conditions.
The reading: this is a staged commitment, not a build decision. Nothing in this paper changes the six gates, and nothing here requires a new one. What the worst case adds is precision about what must be true inside gates that already exist, before any capital commits.
| # | Condition | Sits inside |
|---|---|---|
| 1 | Equity fully committed, with completion support, before FID | Source-of-funds workstream (open) |
| 2 | Metered power pass-through in every colo contract | Power contract gate |
| 3 | Anchor pre-commitments for the operator vehicle mirroring Vehicle A's 60% test | Anchors gate |
| 4 | Debt hedged at FID; operator-vehicle LTV sized to the adverse world (40%) | Funding workstream |
| 5 | Fixed-price EPC and long-lead pre-orders before FID | Site gate |
| 6 | Walk-away re-priced at every gate; the option to stop is the cheapest insurance in this paper | All six gates |
On vehicles: if the Board weighs the floor more heavily than the ceiling, the stress case favours Vehicle A, or Vehicle B only with conditions 2 to 4 in hand. One structural option is named without being recommended: an anchor floor on NNN terms with a colo overlay above it, a hybrid of the two vehicles, would carry Vehicle A's floor and part of Vehicle B's ceiling. Pricing it belongs at the JV-terms gate, not in this paper.
The answer to the question this paper opened with: yes, the prospect survives, because the structure of the commitment lets you find out whether the worst case is arriving before you have paid for it. Strip the staging away and the honest answer becomes no.