Grid-Constrained AI Campus Allocation Director

The mandate

The developer has more prospective AI load than firm interconnection capacity and is negotiating tenants whose training, inference and reservation profiles behave differently under curtailment. Commercial teams value signed megawatts, energy specialists value controllable load, and campus design assumes utilisation shapes that no contract currently enforces. The narrow problem is to decide how scarce deliverable power should be allocated across customers, phases and flexibility products.

The named deliverable is a Grid-Constrained AI Campus Allocation Model and Investment Memorandum. It must join hourly deliverable capacity, network constraints, generation and storage options, cooling dependencies, tenant workload envelopes, curtailment rights, revenue quality, carbon obligations and phased capital. The artefact must permit the committee to compare contracts and infrastructure choices on a common risk-adjusted basis.

Four milestones structure five months: by day twenty, approve the data boundary and constraint map; by week eight, deliver the reconciled power-and-load baseline; by week fourteen, demonstrate the allocation engine against historical grid and representative tenant scenarios; and by week twenty, submit the investment memorandum, contracting guardrails and implementation specification. Invoices follow acceptance of each named output.

The allocation committee and chief investment officer will accept the final work only when grid engineers reproduce the capacity envelope, finance reproduces cash outcomes, customer teams can enter a proposed contract without consultant intervention, and scenarios expose rather than average away coincident peaks, connection delay, curtailment, cooling derate and battery limits. A single annual utilisation assumption will be rejected.

The client will provide interconnection studies, dispatch histories, generation and storage terms, campus designs, cooling curves, tenant proposals, price assumptions and controlled access to commercial and grid specialists. The consultant will not negotiate customer contracts, acquire land, submit grid applications, design electrical works or forecast wholesale prices beyond the agreed scenario set; missing inputs remain explicit sensitivities.

Why this is external work

Every internal function currently optimises a different scarce quantity, and several have incentives linked to contract signature or construction release. The company has not previously valued workload flexibility alongside connection firmness and tenant credit. External specialist modelling is required to create a neutral allocation rule before commercial commitments consume options the grid cannot replenish.

What you will own

• Reconcile firm, conditional and probabilistic grid capacity with campus auxiliary load, cooling derates, losses, maintenance and phased energisation.
• Translate training, fine-tuning, batch inference and latency-bound serving into contractible power envelopes, ramp rates and interruption tolerance.
• Build an allocation engine that values revenue, credit, workload flexibility, delivery date, congestion exposure, energy attributes and capital sequence.
• Stress scenarios for connection slippage, transmission outage, renewable drought, heat events, battery depletion, coincident tenant demand and cooling constraint.
• Define commercial guardrails for reserved power, take-or-pay, curtailment notice, rebound load, stranded capacity and performance evidence.
• Compare network upgrades, storage, flexible generation, workload shifting and staged customer activation without double-counting the same capacity benefit.
• Submit the validated model, user guide, decision memorandum, contracting term sheet and unresolved-assumption register at final acceptance.

Candidate qualifications

• Built capacity-allocation or investment models where grid constraints and digital-infrastructure demand interacted at subannual resolution.
• Translated AI or HPC workload behaviour into power, ramp, interruption and service commitments usable in commercial contracts.
• Reconciled engineering capacity, energy-market scenarios and project finance without hiding uncertainty inside one utilisation percentage.
• Evaluated storage, flexible generation, grid upgrades and workload movement under common deliverability and capital boundaries.
• Challenged a data-centre tenant or commercial forecast that exceeded physically supportable power and defended the conclusion to investors.
• Delivered a model that client operators could reproduce, govern and use after the consulting team left.

Non-negotiables

• Can complete the Santiago working sessions and three Chilean field visits within the five-month milestone calendar.
• Will disclose relationships with prospective tenants, utilities, generators, storage suppliers, developers and infrastructure investors.
• Brings direct grid-constrained data-centre or comparably flexible industrial-load work; annual energy procurement alone is insufficient.
• Accepts reproducibility, scenario transparency and client usability as fee-bearing acceptance conditions.

1. 49 words maximum. Which hourly inputs would you require before treating a megawatt of grid capacity as commercially allocable?
2. 49 words maximum. Describe a flexible-load assumption that looked credible in a contract but failed under actual operations.
3. 49 words maximum. How would you stop sales, engineering and energy teams from double-counting the same capacity option?