The Executable-Capacity Test

The Edge · Portfolio · Monday, 31 August 2026

Researched by an agentic pipeline · reviewed and gated by the author

The portfolio should treat its weakest verified link as the maturity of the whole investment.

The difference between funded and executable

A portfolio committee approves an AI data-centre campus. The capital is available. The land is under option. A utility has discussed service. The financial model assumes energisation in thirty months, and the programme appears in the portfolio dashboard as funded capacity.

Yet none of those facts means the facility can operate. The interconnection study may still change. Firm power may arrive later than the construction programme. Transformers may sit on a different critical path. Permits may be conditional. The intended tenant may not have signed an obligation strong enough to support financing. Curtailment and non-use costs may have no agreed owner.

This is the execution gap now opening inside AI-infrastructure portfolios. Bank of America’s August 2026 infrastructure initiative illustrates the scale of capital seeking deployment: its $250 billion target covers lending, investment, capital-markets and advisory activity across digital, energy and core infrastructure through July 2027. The methodology measures eligible financial activity, not completed physical assets. [S1] The distinction matters because a financed project and an executable project are different portfolio objects.

The practical response is not to replace capital allocation with a new obsession about power rights. Capital, customer credit and expected returns remain essential. The stronger move is to govern an executable-capacity chain: a sequence of rights, obligations and physical conditions that must become credible before irreversible capital is released.

Capacity has four states

Portfolio reporting often compresses unlike claims into one number: megawatts. That hides the maturity of the capacity behind the figure. A useful first correction is to separate four states.

State	What it actually means	Portfolio treatment
Announced	A sponsor has stated an ambition or included capacity in a pipeline	Strategic interest, not supply
Reserved	A queue position, option or preliminary allocation exists	Optionality with expiry and carrying cost
Contracted	Enforceable agreements allocate material rights and obligations	Conditional commitment requiring dependency tests
Deliverable	Power, interconnection, permits, equipment and demand align to an executable date	Capacity eligible for irreversible deployment

These states are not a linear promise of success. A reservation can expire. A contract can contain curtailment rights that make the nominal megawatts unsuitable for the workload. Deliverable power without a creditworthy customer can still leave an unfinanceable asset.

The wider evidence makes that caution necessary. Lawrence Berkeley National Laboratory estimated US data-centre electricity use at 176 TWh in 2023 and projected a wide 325–580 TWh range for 2028. [S4] The US Department of Energy’s July 2026 draft transmission study likewise uses a broad 13–55 GW range for projected data-centre demand growth by 2030. [S5] These are planning ranges, not promises. They describe uncertainty large enough to change which projects should proceed.

The executable-capacity chain

The central framework is a six-link chain. The portfolio should treat its weakest verified link as the maturity of the whole investment.

1. Interconnection. Is there more than a queue position or an indicative will-serve discussion? The committee needs the studies, upgrade responsibilities, milestone dates and termination conditions that make connection credible.

1. Delivered power. Is supply firm enough for the workload, at the required date and load profile? A right to connect is not the same as dependable service. Curtailed or emergency-transfer arrangements may be valuable, but they change operating economics.

1. Permits and consent. Can the site, generation, transmission, cooling and water arrangements be built and operated? Behind-the-meter generation may reduce grid delay while introducing fuel, emissions and community constraints.

1. Equipment access. Are transformers, switchgear, cooling systems and compute hardware available on the same schedule? The IEA describes simultaneous bottlenecks across electricity, grid connections, manufacturing capacity, chips, planning and social acceptance. [S2]

1. Creditworthy demand. Is there enforceable offtake rather than an optimistic pipeline? Demand depends on adoption, workload mix, model efficiency and willingness to pay. The IEA notes that energy use per AI task has been falling rapidly even as more intensive applications grow. [S2]

1. Risk allocation. Who pays when energisation slips, power is curtailed, demand arrives late or reserved capacity goes unused? The answer sits in cancellation rights, take-or-pay terms, liquidated damages, repricing provisions and exit options.

The weakest verified link in the delivery chain, not the largest approved budget, determines executable capacity.

This model changes the portfolio conversation. Instead of asking, “How many megawatts have we secured?”, the committee asks, “How many megawatts are deliverable by date, under what operating conditions, for which committed demand, and with whose balance sheet carrying the failure?”

A gate for irreversible capital

The chain becomes useful when it controls release of capital. Consider a composite portfolio with three proposed campuses.

Campus North has strong tenant credit and a signed construction facility, but only a preliminary utility position. Campus Central has an advanced interconnection agreement and equipment slots, but its tenant commitment is cancellable. Campus South can build behind the meter, yet its fuel supply and air permit remain unresolved.

A conventional ranking may place North first because its financing is clearest. A capacity-rights ranking may place Central first because its power position is strongest. Both answers are incomplete. The executable-capacity gate instead identifies the next irreversible decision and the evidence required before it is made.

• Development gate: fund options, studies and diligence while limiting termination exposure.

• Conditional commitment gate: release long-lead deposits only when interconnection, equipment and permitting milestones are mutually credible.

• Construction gate: commit major capital when firm delivery and customer obligations support the same operating date.

• Expansion gate: add capacity only when utilisation, demand quality and infrastructure performance justify exercising the option.

This is real-options discipline applied to infrastructure. It does not eliminate risk. It makes the price of waiting, committing and exiting visible. An option premium can be rational when uncertainty is high; a speculative reservation with no demand match is simply a concealed portfolio exposure.

JLL reports connection waits exceeding four years in leading data-centre markets and describes speed to power as a primary site-selection criterion. [S7] Ropes & Gray similarly reports that market participants distinguish non-binding indications from arrangements credible enough to support financeability. [S3] Both are commercially interested sources, so their market observations should inform due diligence rather than substitute for it.

The strongest objection

The strongest objection is that today’s power scarcity is regional, temporary and exaggerated by duplicated requests. Developers may be reserving more capacity than they will use. Efficiency gains, distributed inference, geographic relocation or slower adoption could reduce demand before long-lived generation and grid assets earn their cost. On this view, an emphasis on capacity rights could deepen the problem by rewarding early queue occupation and shifting stranded costs to utilities, tenants or communities.

That objection changes the framework in two ways.

First, rights cannot be valued without obligations. A queue position with a low-cost exit may create useful optionality for the developer while creating poor information for the system. A take-or-pay commitment may improve project finance while transferring excess-capacity risk to the tenant. Portfolio governance must record both the right and the liability.

Second, the chain needs a demand-matching test. The Belfer Center highlights the regional concentration of data-centre load and the risk that infrastructure costs outlive or exceed realised demand. [S6] Therefore each capacity tranche should be tested against low, base and high demand, including delayed adoption, improved efficiency, tenant default and workload portability.

The conclusion is narrower than “power is the new capital”. In constrained markets, power and interconnection can become the critical path. Elsewhere, capital, chips, customer demand or permits may bind first. The framework is designed to find that constraint, not assume it.

What the dashboard must stop hiding

Boards should refuse a single pipeline number that adds announced, reserved, contracted and deliverable capacity. The minimum useful dashboard carries maturity and exposure together.

Decision field	Required evidence
Delivery	Earliest credible energisation date and confidence range
Firmness	Firm, interruptible or conditional service; curtailment terms
Dependencies	Interconnection, permits, equipment, fuel, water and construction
Demand	Named counterparty, commitment strength, ramp and cancellation rights
Exposure	Option premiums, deposits, take-or-pay, upgrade and non-use costs
Exit	Transferability, expiry, termination cost and recoverable value

The point is not more reporting. It is to prevent a category error. Financial approval measures willingness to fund. A queue position measures a place in a process. A contract measures enforceable allocation. Deliverable capacity measures whether the physical and commercial chain can operate. Treating them as equivalents creates false confidence precisely when the portfolio appears largest.

Three tests for the next committee

Before releasing the next capital tranche, a committee should make three judgements explicit.

1. The critical-path test. Which unresolved dependency can move the operating date furthest, and what evidence would retire that uncertainty?

1. The symmetry test. Does the organisation holding the capacity right also bear a meaningful share of cancellation, curtailment and non-use cost? If not, the option may be privately valuable and systemically wasteful.

1. The falsification test. What observable change would make the project unattractive: shorter connection waits elsewhere, lower demand, improved model efficiency, a tenant-credit event, or a permit delay? At what point will the portfolio act on that signal?

These tests should be applied to each tranche, not once at initial approval. The IEA’s 2026 outlook captures the underlying tension: data-centre electricity demand is growing quickly, while bottlenecks reduce the likelihood of the most aggressive near-term scenarios and financial conditions still shape build-out. [S2] There is no honest basis for a single deterministic forecast.

The evidence also has a firm boundary. Public data do not yet show whether portfolios with better contracted-capacity positions achieve superior risk-adjusted returns after controlling for tenant credit, sponsor capability, technology choice and regional demand. That uncertainty should remain visible. The framework is a governance response to an observable dependency problem, not proof of an investment factor.

Beyond capital approval

AI infrastructure has not escaped the economics of capital. It has exposed how little a capital decision says about execution when physical systems, contracts and community consent move at different speeds.

The portfolio advantage is therefore neither the biggest financing announcement nor the largest collection of nominal megawatts. It is the discipline to distinguish an ambition from an option, an option from an obligation, and an obligation from capacity that can actually serve demand. Once that distinction is made, the final investment question changes: not “Can we fund it?”, but “Which links are executable, who carries the unresolved risk, and what is the last responsible moment to commit?”

Sources

1. Bank of America — Bank of America Launches $250 Billion, 18-month Critical Infrastructure Finance Initiative — 12 August 2026 — https://newsroom.bankofamerica.com/content/newsroom/press-releases/2026/08/bank-of-america-launches–250-billion-critical-infrastructure-fi.html

1. International Energy Agency — Key Questions on Energy and AI: Executive Summary — 2026 — https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary

1. Ropes & Gray — Data Center Investment in 2026: AI Demand, Power Constraints, and Private Equity Trends — 21 May 2026 — https://www.ropesgray.com/en/insights/viewpoints/102mvfl/data-center-investment-in-2026-ai-demand-power-constraints-and-private-equity

1. Lawrence Berkeley National Laboratory — 2024 United States Data Center Energy Usage Report — 19 December 2024 — https://eta.lbl.gov/publications/2024-lbnl-data-center-energy-usage-report

1. US Department of Energy — National Transmission Needs Study Draft for Consultation and Public Comment — July 2026 — https://www.energy.gov/documents/national-transmission-needs-study-draft-july-2026

1. Harvard Kennedy School Belfer Center — AI, Data Centers, and the U.S. Electric Grid: A Watershed Moment — 10 February 2026 — https://www.belfercenter.org/research-analysis/ai-data-centers-us-electric-grid

1. JLL Research — 2026 Global Data Center Market Outlook — 5 January 2026 — https://www.jll.com/en-us/insights/market-outlook/data-center-outlook