Data center deal red flags: the checklist

The teaser deck is a selection exercise. Every sponsor puts their best foot forward — and the risk lives in what is absent, vague, or structurally impossible to verify at headline level. Most of the failures visible in the public record were legible in the deck. The questions just weren't asked systematically enough, early enough.

What follows is a working checklist of the five flags that surface most consistently across distressed data-center situations. They are not exhaustive — the full diligence universe runs to 130+ items — but they are the patterns that, when present, predict the highest downstream pain.

1. Power capacity that exists only on paper

The most common headline in a data-center teaser is megawatts: "200 MW campus, phase one energized Q3 2026." What the deck rarely shows is the status of the grid offer — whether it is an allocation letter, a formal offer, or a signed connection agreement — and, critically, where the site sits in the interconnection queue.

Queue position matters because most Western grids are now severely backlogged. A project that received an allocation letter in 2022 may be sitting behind several GW of solar and storage that have priority rights. The energization date in the teaser may be arithmetically impossible given the utility's stated connection programme, a fact that only emerges when you ask for the queue confirmation number and check it against the utility's published study schedule.

2. Planning consent described as "expected" rather than "granted"

Data-center planning risk is structurally different from conventional real estate. The opposition vectors are multiple — local amenity groups, environmental regulators, heritage bodies, grid infrastructure agencies — and the interaction effects between them are hard to model. A project that has cleared environmental screening may still face a judicial review triggered by a local resident group; the consent clock resets.

Teasers routinely use language like "planning anticipated Q4 2026" or "consent in progress." These phrases carry no legal weight and mask significant variance in outcome. A project in a jurisdiction with a history of contested data-center applications — parts of Ireland, the Netherlands, Northern Virginia — faces a structurally different risk profile than one in a permissive industrial zone.

3. Anchor tenant economics that do not survive a stress test

Wholesale colo deals frequently underwrite to a single anchor tenant — a hyperscaler or large enterprise — taking 40–60% of campus capacity on a long-term lease. The headline lease rate and term look attractive; the structural risk is in the break provisions, the customer concentration, and whether the rent-per-kW holds if the tenant exercises a scale-back right embedded in the schedule.

The table below shows how the same 100 MW campus underwrites across three illustrative anchor-tenant scenarios. The base case (full occupancy, no break) looks fundable; the stress cases reveal the real equity sensitivity.

The anchor-exits scenario is not a tail risk in today's market. Several large enterprise tenants have exercised early-termination rights as they consolidate into cloud providers' own campuses.

4. Construction cost that assumes a 2022 supply chain

Data-center construction costs have moved materially since 2022. Generator lead times that were 12 weeks are now 72–110 weeks for data-center-class OEM orders in most Western markets. Switchgear has stretched beyond 60 weeks. Substation transformer lead times now exceed 160 weeks in many cases, with high-capacity units quoted at close to four years. A teaser that still underwrites at sub-$10M/MW for a fully-fitted new-build campus in a core market is almost certainly using a pre-inflation cost assumption, often from a feasibility study that predates the current procurement environment.

This matters disproportionately for equity returns. Unlike cost overruns in conventional real estate — where the asset still produces rent — a data-center delay caused by equipment lead times produces a simultaneous cost overrun, debt accrual, and potential ESA breach. The triple penalty is severe.

5. The jurisdiction risk that the teaser treats as boilerplate

Political and regulatory risk in data-center deals is frequently buried in the boilerplate section of a CIM — "the project is subject to applicable local laws and regulations" — rather than surfaced as a structured risk item with mitigants. This is not an accident. It reflects a sponsor preference to minimize the time investors spend on risk factors that are hard to quantify.

The risks that matter most are jurisdiction-specific. Ireland replaced its de facto Dublin connection moratorium in December 2025 with a stringent new CRU regime: new connections require matching on-site dispatchable generation or storage and grid export obligations. Dublin remains heavily capacity-constrained, with data centers accounting for roughly half of regional electricity demand. Any CIM drafted before December 2025 is describing a regulatory environment that no longer exists — the applicable rules depend entirely on when the connection application was submitted. The Netherlands has Amsterdam prohibiting all new data center construction or expansion within the municipality, with Utrecht province and several other municipalities following with their own restrictions, and new hyperscale projects nationally confined to three designated zones. Northern Virginia has seen Loudoun County eliminate by-right development as of March 2025, requiring Special Exception approval for every project, and Fairfax County impose enclosure, setback, and buffer requirements. Each of these is a known, documented constraint that should be in the risk section of any credible CIM — its absence is itself a signal.