A capacity approval needs to account for the contract dates, the recovery capacity required on those dates, and the period over which costs are counted. Consider the synthetic planning model's twelve-month horizon beginning in November 2026: a nine-month contract on older racks, followed by a three-month contract on replacement racks. The service needs two primary racks and two standby racks in every month.
In an illustrative review, three people each check one of three slides. The schedule slide shows standby-site power ready in month 2. The hardware slide shows every rack the plan needs already qualified. The finance slide shows twelve months of cash. Each slide passes its own check, so the review approves the commitment. But the separate checks miss how the dates constrain the reservation. Recovery capacity must be ready in the months the service needs it, and the financial result depends on the horizon chosen for assets that may outlive the contract.
The IEA reported in April 2026 that equipment supply chains were tightening and that a growing project pipeline was straining planning and regulatory systems and delaying grid connections. IEA, 16 April 2026 A commitment review therefore needs to check which months can be reserved, whether recovery capacity is ready for them, and which cost horizon the financial result uses. CL-02
Baseline reservations
The baseline covers twelve UTC calendar months from November 2026 at two synthetic sites, Dublin and Frankfurt. Its nine older-profile demand-months and three replacement-profile demand-months total twelve committed demand-months. Each requires two primary racks and two standby racks. A separate, uncommitted flexible training request seeks one older rack for nine months.
Contractual standby is capacity reserved as part of the service promise. Physical warm spares are inventory that may be available, but their presence alone does not establish a qualified recovery path for every committed month. Treating the two as interchangeable understates what the commitment requires.
With the baseline inputs, all twelve committed months are reserved. Net modeled revenue is \$11,671,743.60. CL-01
Whole-window reservations
A contract is accepted only if its entire window can be reserved. The planner uses deterministic first-fit placement with fixed sites and fixed hardware. If any month cannot satisfy the required primary and standby reservation, the planner reserves none of the contract's months. There is no partial credit under this rule.
Three cases each withhold nine of the twelve committed demand-months. In the power case, standby-site power is ready in month 2; in the fabric case, standby-site fabric is ready in month 2; in the maintenance case, the older primary cohort is maintained in month 5. Months use zero-based indexes, so months 0 and 1 precede the standby site's power or fabric readiness date in the first two cases. Net modeled revenue falls to \$3.888 million in each case.
These nine months count the full refused contract window. At least one month of the nine-month contract could not be reserved, which caused the whole window to be refused. They are not nine months of outage and cannot be expressed as an availability percentage.
A US Department of Energy report from July 2024 said producers reported large power-transformer lead times reaching and exceeding 36 months. US DOE, July 2024 The report is not a model input. It is a reason to check a power-ready date as closely as the commercial start date.
First-fit has no proof of optimality. A global rearrangement might place a contract that this rule refuses, but the approval must use the months that the rule in force will reserve. CL-03
Recovery capacity requirements
Standby-site readiness causes two of the three cases that withhold nine committed demand-months. Recovery capacity must use the same qualified hardware as the primary cohort, sit on a different failure domain and feeder, and be reservable throughout the committed window. A standby-site delay can therefore cause the planner to refuse the whole contract. It does not reschedule it.
The model charges standby costs to the primary service bundle. The committed service bills delivered reserved primary rack-hours; that unit does not measure fungible compute performance. A warm spare elsewhere may have operational value, but it satisfies a contractual standby requirement only if it is qualified and ready on the required dates.
Checking only the primary site's dates can leave an approved commitment with no qualified recovery path for the same months. Checking both sites' calendars exposes that mismatch before the commitment is counted as reserved. CL-07
Costs and the NPV horizon
The baseline reserves every month and still produces a project NPV of -\$9,315,112.17 over the twelve-month window, with flows discounted at month end. Much of that result comes from charging rack purchases in full within the twelve-month horizon while later service and financing obligations remain outstanding at the end of the window. The calculation applies no terminal enterprise valuation. The negative NPV does not establish that the racks are unprofitable over their lifetime.
A reviewer needs the horizon stated explicitly to judge whether the calculation evaluates a contract or an asset lifecycle. That boundary determines which costs and later obligations the result includes. CL-01
The baseline's \$1.6 million of eligible resale depends on a supplied sanitization reference string. Checking that string does not verify erasure or establish legal compliance.
Energy tariffs can change costs while reservations stay fixed. Doubling both tariffs adds exactly the baseline energy cost to gross cost and leaves reservations and net revenue unchanged. Modeled energy is held fixed in this tariff intervention, and net revenue means revenue minus credits, before costs.
Scope and further work
The results come from a synthetic model applying one fixed planning rule to frozen inputs. They make no claim about real sites, service levels, compliance, market prices, hardware throughput, lifetime profitability or operational qualification. The site names are labels only; they do not show that the two sites fail independently. The paper's tests are implementation QA, not independent acceptance testing or operational qualification.
Under one rule and one set of inputs, the baseline reserves every committed month and reports a negative NPV. Single-change interventions on that baseline cause whole contracts to be refused. Both results belong in the same review. Further work would compare whole-window contracts with explicitly permitted partial contracts in a controlled experiment and test sensitivity to contract length and discount rate. CL-09
Approval requirements
Before the commitment is signed, the approval should name the months the whole-window rule will reserve, the recovery capacity those months require and the horizon used to judge the NPV.
Source: Decision Application frozen source 5c4c8434dd0f92556aed46b4ebcda0c11ae468de, executable manifest f5e10f78728faab62c848c7adf442e5b2c82dba8a24921454c0fac1ed4ce5dfb. Read the technical paper and its source, simulation results and tests.