Abstract
A UL 2755 listing certifies a modular data center as a manufactured assembly. It does not certify the site the assembly is placed on, the occupancy it operates under, the water supply that protects it, the egress geometry that surrounds it, or the management systems that govern its operation. This paper examines the substitution argument that has taken hold in modular procurement, in which the listing is offered to an authority having jurisdiction and to a property insurer as though it discharged the operator's own compliance obligations, and sets out why the argument fails at every authority it meets.
The analysis develops a four-layer compliance stack. Layer one is the product listing, governed by UL 2755 and the component listings beneath it, and owned by the manufacturer. Layer two is site compliance, governed by the National Fire Protection Association standards, the International Code Council family, and the applicable energy standards, and owned by the operator. Layer three is the management-system layer, governed by the International Organization for Standardization standards and the applicable federal control frameworks, and owned by the operator. Layer four is insurance underwriting, governed by the property insurer's loss-prevention data sheets and owned jointly by the operator and the insurer. The operator is the accountable party on three of the four layers, and the listing addresses a portion of one.
The paper traces the consequences of the substitution argument through the three authorities that test it. The authority having jurisdiction tests it at submittal and returns a request for information that typically runs to twenty or more document categories and thirty to sixty days against the project critical path. The property insurer tests it at the underwriting walkthrough and returns a recommendation report, a surcharge, or a decline. The claims adjuster tests it after a loss, when the engineering basis of the asset determines what portion of the loss is recoverable. Each test is independent of the others, and none of them is gated by the listing event.
The recommendations are operational rather than exhortative. They center on a compliance crosswalk matrix issued at design freeze that names, for every applicable standard, the responsible party and the evidence artifact; on a dry-run walkthrough with the authority before the formal submittal; on an insurance engineering walkthrough at eighty percent design completion; and on an annual reconciliation of the three independent records the operator holds. The paper sets out a maturity model, a ninety, one hundred eighty, and three hundred sixty-five day implementation roadmap, and the capital argument for adopting the discipline across a portfolio of modular deployments rather than project by project.
Executive Summary
A modular data center arrives on site with a listing record that certifies what the manufacturer built. The operator's compliance obligation begins at the moment the unit is set on a foundation, connected to a service entrance, and placed inside a fire area with a defined occupancy. Those two facts are separated by a boundary the deployment itself creates, and the substitution argument attempts to erase it. This paper argues that the boundary is structural and that no product listing of any kind can bridge it.
Finding one. The substitution argument is a procurement response to real pressure rather than an engineering position. Modular delivery was adopted to compress schedule, and a single credentialed document that appears to discharge a distributed evidence set is exactly what a compressed schedule rewards. The listing is genuine, it is issued by a credentialed organization, and it carries a follow-up service obligation. None of that is in dispute. What is in dispute is the inference drawn from it.
Finding two. The scope boundary is explicit in the governing document. UL 2755 is an Outline of Investigation that scopes the listing to the enclosure, the internal wiring methods, the listed components, the internal cooling and detection, the internal grounding, the marking, and the component substitution rules. It is silent on the site, the occupancy, the campus water supply, the egress paths, the separation distances, and the utility interconnect. The silence is a scope statement rather than an omission.
Finding three. Three independent authorities test the argument and all three reject it. Authorities having jurisdiction reject listing-only submittals because their liability attaches to the installation rather than to the product. Property insurers evaluate the deployed configuration against their own loss-prevention data sheets, which routinely exceed the code minimums by fifteen to forty-five percent on representative engineering line items. Claims adjusters evaluate the engineering basis after a loss, and a listing-only record materially reduces recovery across property, business interruption, contingent business interruption, and third-party liability.
Finding four. The cost of the argument is schedule, and schedule on a modular program is the reason the program exists. A first-pass listing-only submittal typically produces five of seven recurring rejection categories and consumes thirty to sixty days of response time on a schedule that was compressed by design and therefore carries no float. That slip lands immediately before commissioning, where it cannot be recovered, and it is frequently absorbed by shortening commissioning, which converts a documentation problem into a reliability problem.
Recommendation one. Build the compliance crosswalk matrix and issue it at design freeze. The matrix names the applicable standard, the responsible party, and the evidence artifact for every line of the compliance demand. It is the load-bearing artifact of a defensible submittal, it is the section most often missing and most reliably requested, and it converts a contested inference into a set of assignments with owners and dates. The first build runs four to six weeks; every subsequent adaptation runs one to two.
Recommendation two. Run the regulatory and insurance tracks in parallel from design freeze rather than in sequence from permit submission. Dry-run the submittal with the authority before it lands, and bring the property insurer's risk engineer in at eighty percent design completion, which is the narrow window where findings are still actionable and not yet rework. Engaging risk management at design freeze, with a standing seat in design review, is the single organizational change that most reliably prevents late-stage insurance-driven rework.
Recommendation three. Operate the program rather than executing it. Annual reconciliation of the product listing record, the authority acceptance record, and the insurance underwriting record, routed through the management review the operator's certified systems already require, is what keeps the matrix current across code cycles, listing revisions, and renewal recommendations. The economics favor the portfolio: the matrix and the authority relationships are built once and applied repeatedly, while the rejection cycle is incurred project by project and never becomes cheaper. Across three or more modular deployments the discipline pays back several times over, and what it buys is schedule certainty, which is the outcome the capital plan actually needs.
Author's Preface
I have watched the substitution argument arrive in three different rooms over the past several years, and it arrives the same way each time. Someone produces a UL 2755 listing record, sets it on the table, and asks what else could reasonably be required. The question is sincere. The listing is a serious document produced by a serious organization, and the person asking has usually been told by someone they trust that it settles the matter.
It does not settle the matter, and the reason has nothing to do with the quality of the listing. A product certification answers a question about a manufactured assembly at a moment in time. An authority having jurisdiction asks a question about an installation on a specific parcel under a specific occupancy. A property insurer asks a question about a deployed configuration in its site context. A claims adjuster asks what evidence existed on the day of the loss. Those are four different questions, and the listing was written to answer none of them. Recognizing that is the whole of the argument in this paper; everything else is working out the consequences.
I wrote this paper for the people who have to carry those consequences. The operator whose submittal is about to be returned with a request for twenty additional document categories. The project manager whose schedule has no float to absorb it. The risk manager who will be asked at renewal why the asset was engineered to a code minimum rather than to the insurer's data sheet. The executive who will eventually be asked why a facility that held every certificate on the wall recovered sixty cents on the dollar. Each of those conversations is avoidable, and the thing that avoids all four is the same modest document assembled at the right moment.
The argument here is not against modular delivery. Modular deployment is the correct answer to the schedule and quality problems that gigawatt-scale artificial-intelligence infrastructure has created, and I expect its share of new capacity to keep growing. The argument is against one specific inference about what a listing buys. The compliance obligation travels with the site and with the operator, it always has, and a manufactured enclosure does not carry it across the fence line.
Full white paper below
