THE REVENUE IS REAL. THE NEXT PHASE ISN'T POWERED. THE MARKET IS ABOUT TO TELL THEM APART.

WHAT THIS IS

This is a forward call: the second of two in this series, and the harder of the two. Unlike the LL97 call in Note 05, where the obligation is fixed by statute, this one sits in the most hyped, fastest-moving corner of real assets, where consensus is loud and the facts move monthly. That is precisely why we want it on the record with a date and a falsifier. A call is only worth something if it can be wrong, and this one can. This note is issued Summer 2026. We will publish a scored update in Summer 2028. Here is the call.

THE CALL

As of Summer 2026, we project that data-center assets carrying strong current cash flow but whose next-phase or expansion value depends on power and interconnection capacity that is not yet secured will have that unbuilt upside discounted by lenders, buyers, and investors: as reduced growth credit, a wider gap between stabilized and development value, or weaker financing on the expansion component, in 2026–2029 capital events, materially before that upside fails to show up in current revenue.

In plain terms: the market will start paying full price for the megawatts that are real and deliverable, and stop paying for the megawatts that are only announced, and it will make that distinction at the financing table and the negotiating table, well before the unbuilt phase would ever have produced a dollar.

THE STRUCTURE: THE CONSTRAINT MOVED FROM THE RACK TO THE SUBSTATION

For most of the last cycle, the binding constraint on data-center value was demand and hardware: could you fill it, could you get the chips. That constraint has decisively moved. From 2025 onward, the bottleneck migrated from the server rack to the substation — the central concern is now systemic grid access, with interconnection queues delaying projects for years. As one search-and-advisory view put it: power is now the gating factor; if the megawatts are not real and deliverable on schedule, the rest of the strategy does not matter.

The numbers behind that shift are not subtle. In Northern Virginia, Phoenix, and Dallas, the markets hosting the largest share of announced 2026 capacity, and grid interconnection waits now run 4 to 7 years. Nearly 2,300 gigawatts of generation and storage sit stuck in U.S. interconnection queues, which is more than the country's entire installed power capacity, with projects averaging five years in queue before commercial operation. In PJM, which serves 65 million people across 13 eastern states, projects that became operational in 2025 had spent an average of eight years in the queue. The supply-demand gap is structural: the Department of Energy estimates an additional 100 GW of peak generating capacity will be needed by 2030, with roughly half driven by data centers.

And the gap between announced and energized is already visible in the build data. Against a 16 GW announced 2026 U.S. pipeline, only about 5 GW is actually under construction, and 30% to 50% of planned 2026 projects are now expected to be delayed or canceled. The capital is committed; the physical layer to absorb it is years behind. That divergence (fully funded ambition meeting an unbuildable timeline) is the raw material of the repricing we are calling.

THE MECHANISM: TWO ASSETS THAT LOOK IDENTICAL ON TODAY'S P&L

Here is the conversion: Note 03's Pathway 05, infrastructure constraint becoming stranded upside value.

Take two data-center assets with identical current operations: same tenants, same current load, same trailing cash flow. On this year's income statement, they are the same asset. The difference is invisible on the P&L: one has firm, contracted, deliverable power for its next phase of expansion; the other has a development plan, a land position, and a place in an interconnection queue.

For a while, the market prices them the same, because current revenue is the same. But the value of a data center is not only its current load: it is the credibly buildable expansion that justifies the multiple. The moment a counterparty underwrites the next phase, the two assets diverge. A lender sizing debt against expansion capacity asks whether the megawatts are firm. A buyer paying for growth optionality asks when energization actually happens. An investor underwriting the development upside asks what the queue position is really worth. Interconnection has shifted from a technical hurdle to a core underwriting variable; developers now review queue position and deliverability before finalizing land acquisitions, and model multiple energization scenarios before securing site control.

When that question gets asked, the asset with unsecured power doesn't lose its current revenue. Rather, it loses the credit for its future. The expansion that was carried at full value becomes a maybe: discounted, reserved against, financed on worse terms, or excluded from the buyer's number entirely. The current P&L never moved, but the value did. That is stranded upside: the growth was in the price, and the market took it back out, because the power to realize it isn't real yet.

The reason this is missed is structural, and it is the same reason Note 03 gives for Pathway 05: current revenue is strong, so the asset looks healthy today, while the market begins discounting the next phase. The team watching occupancy and the team watching the interconnection queue are not the same team, and the queue is where the value is quietly being decided.

For the owner holding the gap, this is where a number changes the decision. The cost of securing power (a behind-the-meter build, an on-site generation deal, a PPA) is usually knowable. What has been missing is the other side of the equation: what the unsecured upside is actually worth, and therefore what securing it is worth defending. A team that can set a quantified stranded-upside figure against the cost of firming the megawatts can make the investment on purpose, rather than freeze on the cost and discover the discount at the term sheet.

WHY THE PRICE MOVES BEFORE THE REVENUE DOES

The core of this call, like the LL97 call, is about timing: the discount lands before the shortfall does. The logic is the same and it is simple. A counterparty underwriting a multi-year hold or loan cannot wait for an unbuilt phase to fail to materialize; they price the probability and timing of energization now. A 5-to-8-year queue inside a hold period is not a distant abstraction to a lender; it is a deliverability risk they reserve against today. The capital market reprices on the forward view, not on the realized outcome. So the asset gets marked down for stranded upside years before the upside would have arrived: at the 2026–2029 refinance, sale, or capital raise, not at the eventual energization date.

There is also a hard cost signal making this concrete in real time. PJM's July 2025 capacity auction cleared at roughly $270 per MW-day, up from about $29 the prior delivery year, with two zones clearing near $440–$466. Power is not just slow to connect; it is repricing sharply where it is scarce. A counterparty underwriting expansion economics is now doing so against a power-cost curve that has moved an order of magnitude, which feeds directly into how much that unbuilt phase is worth.

WHAT WE ARE NOT CLAIMING

This is the call most exposed to consensus, so its limits matter most. We are not calling a data-center crash, or that the sector is overbuilt, or that AI demand is a bubble. The current cash flows are real and we are not betting against them. The call is narrow: it is about the gap between secured and unsecured expansion value, and the market's growing willingness to price that gap.

We are not claiming power scarcity is unknown. It is the opposite of unknown; it is the single most-discussed fact in the sector. Our call is not "power is a constraint" (everyone knows that). It is a specific, falsifiable claim about when and how that known constraint converts into a measurable valuation differential at a capital event, that the discount on unsecured-power upside shows up in 2026–2029 transactions, ahead of any revenue shortfall. The risk label is universal; the dated repricing pathway is the claim.

And we name the genuine ways we could be wrong, below, including the workarounds that could neutralize the constraint faster than we expect.

THE FALSIFIER: WHAT WOULD PROVE US WRONG

Observed by Summer 2028, the following would mean this call failed:

The call is wrong if, in 2026–2029 capital events, data-center assets with unsecured next-phase power are valued, financed, and transacted at full credit for that expansion — i.e., if lenders, buyers, and investors continue paying for announced-but-unpowered capacity on the same terms as firm, deliverable capacity, with no measurable discount, reserve, or financing penalty attributable to power deliverability. If the market keeps treating a queue ticket as equivalent to a megawatt, the conversion we are claiming did not happen.

The call is neutralized (right mechanism, wrong premise) if the power constraint itself is resolved far faster than the queue data implies, through behind-the-meter generation, on-site gas, SMRs, accelerated FERC/DOE large-load interconnection reform, or curtailment-tolerant "flexible" interconnection, such that unsecured-power assets quickly become secured-power assets and the value gap closes before it can be priced. This is the most realistic way we are wrong: the DOE has directed FERC to expedite large-load interconnection, and operators are pursuing self-generation and hybrid models to become operational while remaining in the queue. If those workarounds scale fast enough, the stranded-upside window closes on its own. We think they scale slower than the capital does; if we are wrong about that, we are wrong about this call.

The call is right if, by Summer 2028, there is identifiable evidence (financing structures, transaction comps, investor commentary, valuation practice) that data-center expansion value is being explicitly discounted for power-deliverability risk in capital events, ahead of any decline in current revenue.

HOW WE WILL SCORE IT

Three tests, gradable in Summer 2028:

Test 1 — Valuation differential. Is there observable evidence that data-center assets with unsecured next-phase power transacted, financed, or were marked in 2026–2029 at a discount on expansion/growth value relative to comparable assets with firm, deliverable power — with current cash flow held constant?

Test 2 — Ahead-of-revenue. Did that discount appear before the unbuilt capacity failed to generate revenue — i.e., did the market price the deliverability risk forward, rather than waiting for a realized shortfall?

Test 3 — Constraint-specific. Was the discount attributable specifically to power/interconnection deliverability, rather than to a general data-center repricing, an AI-demand reassessment, or a rate cycle?

A clean pass is all three. A partial is Test 1 without Test 2's timing. A miss is no measurable Test 1 effect, or a constraint that resolved before it could be priced (the neutralization case) — which we will report honestly as the call's premise failing, not as a win.

The market has already begun the conversion. Power has moved, in the industry's own words, from a due-diligence item to the primary investment constraint and a core underwriting variable. The question is no longer whether deliverability gets priced; it is whether it gets priced in the window we specify, on the assets we describe, ahead of the revenue. That is the bet.

If we are right, the owners holding unsecured expansion value who go to a capital event in 2026–2029 will meet the discount at the term sheet, having had since now to convert announced megawatts into firm ones, securing interconnection, documenting deliverability, building the evidence that turns a queue position into a credible, financeable phase. The owners who read the constraint early have a window to make their upside real before the market stops paying for it. The owners who wait will be told what their growth is worth by the counterparty who priced it first.

The risk label — "power is the constraint" — is the most familiar fact in the sector. The repricing pathway, and the date the market starts paying only for power it can verify, is what this call is about. The clock starts now.