Investment committee memorandum: Project Kestrel
This is the sample for the deal-memo product on the rate card, written in the format a credit committee already reads. Project Kestrel is fictional and every figure is a round illustrative number; no real sponsor, lender or counterparty is described.
ILLUSTRATIVE SAMPLE. Project Kestrel is a fictional asset and every figure below is a round number used to show the shape of the analysis, not a market statistic. No client, sponsor, lender or counterparty named or implied here is real. This is the format and the reasoning a credit committee reads.
| To | Investment Committee |
| From | Infrastructure Credit, Digital Infrastructure |
| Subject | Project Kestrel - 180 MW GPU-hosting campus, senior term loan |
| Request | Approval of a participation in a $1.70bn senior secured term loan facility |
| Recommendation | APPROVE, subject to the conditions in section 9 |
1. Recommendation
We recommend participating in the senior term loan for Project Kestrel, a 180 MW (critical IT) GPU-hosting campus in three phases, contracted to a single hyperscale tenant under a twelve-year take-or-pay agreement.
The case does not rest on the quality of the campus. It rests on four things, in order of importance:
- The tenant's contractual obligation runs for the full tenor of our loan, with no termination right that does not also require payment of the outstanding debt.
- The sponsor funds construction overrun from equity before our facility draws past 70 per cent of committed capital, so the completion risk sits with the party who controls it.
- The power position is owned, not queued: an executed interconnection agreement, a substation energised on a schedule with liquidated damages, and a fixed-price energy block covering the contracted load.
- The residual, at the end of the take-or-pay term, is a powered shell with liquid-cooling infrastructure in a constrained market - a defensible value even under a punitive re-contracting assumption.
The case fails on one thing: technology life. Our tenor is seven years; the tenant's GPU fleet will be replaced roughly twice inside it. If the tenant's obligation is written against a specific hardware generation rather than against a capacity specification, we are lending against an asset that becomes obsolete on a schedule the borrower does not control. Section 6 sets out why we think the contract as drafted avoids this, and section 9 makes it a condition.
2. The transaction
| Item | Amount (illustrative) |
| Total project cost, all three phases | $2.90bn |
| Senior secured term loan (this facility) | $1.70bn |
| Mezzanine | $250m |
| Sponsor equity and completion support | $950m |
| Senior LTV at financial close | 59% |
| Senior tenor | 7 years, fully amortising from year 5 |
| All-in coupon | SOFR + 325bp |
| Cash sweep | 75% of distributable cash above a 1.30x DSCR lock-up |
The structure is conventional for the asset class, which is the point: we are not being paid for novelty. We are being paid 325 basis points over the risk-free rate to underwrite a contracted cash flow with a technology risk attached to it.
3. What we are actually buying
We are not buying a building. A data-centre shell with power is worth roughly its replacement cost less depreciation, and in a supply-constrained market that is a floor rather than an exit. What we are buying is a twelve-year payment obligation from a counterparty whose credit we can underwrite, secured against an asset whose alternative use is credible.
That ordering matters when the deal is stressed. In a downside where the tenant defaults, our recovery is not the discounted value of the lease. It is the cost of finding a second tenant, less the time it takes, less the capital required to make the asset fit for that tenant's hardware. In a market where capacity is scarce and liquid-cooled halls are scarce absolutely, that gap is smaller than it looks. We have underwritten it at 55 per cent of cost, not the 80 per cent a refinancing model would show.
4. Technical position
Power. 180 MW of contracted capacity delivered in three energisation tranches. The interconnection agreement is executed and assigned to the project company, with the utility's milestone schedule carrying liquidated damages to the project. A fixed-price energy block covers the contracted load for the first eight years; beyond that the tenant bears power price risk under the offtake agreement, not the project. This is the single most important structural point in the deal, because it converts the largest uncontrollable input into a pass-through.
Cooling. The halls are built for direct-to-chip liquid cooling with air as a fallback, at a per-MW capital cost roughly 15 per cent above an air-cooled build (illustrative). We would normally resist paying for optionality. Here it is the reason the residual value argument survives: a liquid-ready hall can serve the next two hardware generations, and an air-only hall cannot.
Delivery. Three tranches, each with its own acceptance schedule and its own rent commencement. The acceptance criteria are technical, measurable and testable by a third party: sustained power density, coolant flow and temperature at the rack, network latency to named exchanges, and a defined period of continuous operation at load. This is the part of a data-centre contract that gets argued about two years after close, and it is the part we read most closely.
The trap we are watching. Tranche delivery slippage. A sponsor who is nine months late on tranche two still has a contracted asset; we have a facility drawing against a rent stream that has not begun. Mitigation is in the structure, not the contract: the tranche-two and tranche-three commitments are unavailable until the prior tranche has passed acceptance.
5. Commercial position (illustrative)
| Metric | Base case |
| Contracted revenue at full ramp | $310m per year |
| Contracted opex and power pass-through | $95m per year |
| EBITDA at ramp | $215m per year |
| Senior DSCR, minimum | 1.55x |
| Senior DSCR, average | 1.71x |
| Break-even contracted revenue for 1.0x DSCR | $150m per year |
Two observations. First, the DSCR cushion is real but not generous: a 12 per cent shortfall in rented capacity takes the minimum to the lock-up. Second, single-tenant concentration is total. There is no diversification argument available here, only a counterparty argument, and the credit work on the tenant matters more than the credit work on the asset. That work is in the appendix, not in this memo, and it is a condition of approval that it is re-run at each drawdown.
6. Technology life and tenor
This is the risk that decides the deal.
A GPU campus is not a twenty-year asset with a mid-life refurbishment. The tenant replaces compute on a four-to-five-year cycle, and each replacement changes power density, coolant temperature and rack weight. The building can absorb that if it was designed for it and cannot if it was not.
The contract as drafted obliges the tenant to pay for contracted capacity, not for a specified hardware generation, and to fund any tenant-driven fit-out. The tenant's obligation is not reduced by its own decision to refresh, retire or relocate hardware inside the contracted footprint. That is the right structure for us. The version of this contract that would be unacceptable is the one where the tenant's payment is tied to equipment it can remove.
We also price the residual inside the tenor. Even if the tenant renews on terms 20 per cent worse than today's, the loan amortises to a 40 per cent LTV position by maturity (illustrative). We are not depending on a renewal to be repaid.
7. Risk register
| Risk | Severity | Why | Mitigant |
| Tenant credit deterioration | High | Single tenant, no diversification | Contracted revenue, security package, drawdown re-underwrite, tenant rating trigger at 50% drawn |
| Tranche delivery slippage | High | Rent does not begin until acceptance | Tranche availability mechanics; sponsor completion guarantee; LDs from EPC and utility |
| Technology obsolescence | Medium | 7-year tenor against 4-year hardware cycles | Liquid-ready design; capacity-based obligation; residual underwritten at 55% of cost |
| Power price exposure post-year 8 | Medium | Fixed block ends inside the tenor | Pass-through to tenant under offtake; review at year 6 |
| Interconnection delay | Medium | Utility milestone risk outside sponsor's control | Executed IA; LDs; sequenced tranche funding |
| Construction cost overrun | Medium | Fixed-price EPC with limited contingency | Sponsor funds overrun before senior draws past 70% |
| Refinancing risk at maturity | Low | Amortising structure | 40% LTV at maturity (illustrative) |
| Counterparty concentration in the tenant's own customer base | Low | Not our direct exposure | Reviewed in tenant credit appendix |
8. Three-outcome view
| Scenario | What happens | Senior outcome (illustrative) |
| Base | All three tranches deliver within 60 days of schedule, tenant performs, refinanced at maturity | Full repayment, 1.71x average DSCR, IRR at coupon |
| Downside | Tranche two slips nine months; tenant renews at 80% of current rate at year 12 | Full repayment, DSCR troughs at 1.28x, lock-up triggers once, no loss of principal |
| Severe | Tenant defaults in year 5 after a demand contraction; asset re-let in 18 months at 65% of contracted rate | Recovery 78-85 cents on the dollar after cost to re-let and downtime |
The severe case is the one the committee should argue with. Our 78-85 cent range rests on two judgements: that liquid-ready capacity in a constrained market can be re-let inside eighteen months, and that the cost to refit for a new tenant is under $25m (illustrative). Both are assumptions, not evidence, and section 10 says so explicitly.
9. Conditions precedent
We recommend approval subject to five conditions:
- The capacity obligation is drafted as specified. Tenant payment tied to contracted MW, not to identified equipment, with no reduction on hardware refresh or relocation. Legal to confirm before signing.
- Tranche availability is sequential. Tranche two and three commitments unavailable until the prior tranche has passed acceptance, evidenced by the independent test report.
- Sponsor completion support is funded, not promised. Cash in escrow or a letter of credit sized to the full overrun case, not to a percentage of cost.
- Tenant credit is re-underwritten at every drawdown and at any time the tenant's rating falls below investment grade.
- A technology review at year 3 and year 5, commissioned by the agent and paid by the borrower, with the findings circulated to lenders.
10. Assumptions and evidence status
| Assumption | Status |
| Tenant obligated for full tenor with no market-out | Evidenced - contract reviewed, clause references in appendix B |
| Interconnection agreement executed and assigned | Evidenced - IA and assignment reviewed |
| Fixed-price energy block covers 8 years | Evidenced - PPA term sheet reviewed; final pricing subject to close |
| EPC is fixed-price with LDs | Partially evidenced - contract in negotiation; LDs provisional |
| Re-let achievable in 18 months at 65% of rate | Assumption - no comparable transaction available to us; based on market enquiry only |
| Refit cost under $25m | Assumption - two contractor estimates, neither binding |
| Hardware refresh does not reduce tenant payment | Assumption pending legal - drafting intent confirmed verbally, clause not final |
| Utility energisation on schedule | Partially evidenced - milestone schedule exists; utility's historical performance not assessed |
The committee should read the last four rows as the deal. Everything above them is diligence; the four rows below are where we could be wrong.
11. What would change the recommendation
We would withdraw the recommendation if any of the following were true: the capacity obligation ends up tied to identified equipment; the sponsor's completion support is a guarantee rather than funded; the tenant's credit is below investment grade at close; or the energy block is shorter than six years. Each of those converts a contracted cash flow into a development risk, and development risk is not what this facility is priced for.
Prepared as a sample. Project Kestrel is fictional; all figures are illustrative. Chris Romei writes investment-committee memos, sector primers and lender packs for infrastructure funds, data-centre developers and neoclouds. Portfolio: chris-romei-writing.pages.dev