CMP-02Compare · Delivery method
Same scope of work, different location. What the delivery method actually changes is schedule overlap, inspection regime, customization freedom, and transport risk — not the physics of power and heat.
PUBLISHED LAST VERIFIED BY JOSEF ELIMELECHREVIEWED PODOS AI ENGINEERING
The honest answer
Framing
Every megawatt of AI capacity needs the same components either way: switchgear, distribution, cooling plant, a thermal loop to the chips, structure, fire protection, and a commissioning campaign that proves it all works together. Factory construction does not delete that scope. It moves the standardizable part indoors and leaves the rest — earthwork, foundations, utility service, final tie-ins — on the site.
That matters more than it used to because of what goes inside. Uptime Institute's 2025 survey puts typical rack densities in the 10–30 kW band with AI clusters above it[4], and ASHRAE documents liquid cooling moving into mainstream facilities as densities climb[5]. Dense racks demand tight integration between power distribution and direct-to-chip liquid cooling — repeated, tolerance-sensitive assembly that a production line does well and a rotating field crew rebuilds from scratch on every job. This page compares the two methods of building; for the product-level comparison, see modular vs traditional AI data centers.
The schedule argument is narrower than it is usually presented. A site-built project is largely sequential because the work shares one physical space: the slab cures before the structure rises, the structure before mechanical rough-in, rough-in before fit-out, fit-out before commissioning. Each trade inherits the previous trade's delay.
Factory construction breaks that dependency by putting the work in two places at once: while units are produced and tested indoors, the site crew is clearing, grading, pouring, and running conduit. The compression comes entirely from that overlap, so it is bounded by it. Three things do not overlap:
The test is simple: draw the critical path, then ask what fraction of it is construction labour. That fraction is the ceiling on what factory production can win.
Table 1 · Quality control
Neither method is inherently higher quality. They fail differently, and the difference is worth understanding before choosing. Row FB-05 is the one buyers underweight: site-built quality varies unit to unit, while factory-built quality correlates, so a repeated mistake ships to everyone in the batch. Traceability, documented test records, and a corrective-action path matter more than any single unit's test report.
| Ref | Control point | Site-built | Factory-built |
|---|---|---|---|
| FB-01 | Who performs the work | A rotating crew from the local trade market, different on every project and often between phases of one project. | A fixed crew repeating the same build, where the tenth unit is assembled by the people who assembled the first nine. |
| FB-02 | When defects surface | During on-site commissioning, at the end of the schedule, when rework is most expensive and most likely to move the date. | During in-plant inspection and pre-shipment testing — but a process defect repeats across every unit already built. |
| FB-03 | Inspection regime | Progressive inspections by the local authority having jurisdiction as each phase closes; the record is per-project. | In-plant inspection by a third-party agency or a state industrialized-building program, plus on-site final inspection after assembly. |
| FB-04 | Test conditions | Cooling and power are proven under whatever weather exists on the commissioning date. | Integration is tested indoors on instrumented benches, with a failed test re-run the same day. |
| FB-05 | Failure mode of the QC system | Variance — quality tracks the crew and the market; no two rooms are identical. | Correlation — a line defect is a fleet defect. Concentration is both the benefit and the exposure. |
A schedule advantage quoted without naming it — by any manufacturer, ours included — is a best case, not a forecast.
1205
ICC/MBI standard for off-site inspection
Approvals
Factory-built capacity does not avoid permitting. It splits approval across two authorities: an off-site path that reviews and inspects the unit where it is made, and the ordinary local path that permits the ground it lands on. The International Code Council and the Modular Building Institute published ICC/MBI 1200 and 1205 to standardize that handoff — 1200 covering planning, design, fabrication, and assembly in off-site construction; 1205 covering permitting, in-plant and on-site final inspection, third-party inspection, industrialized-building departments, state modular programs, and the authority having jurisdiction.[2][3]
| Ref | Approval interface | Authority | What changes with delivery method |
|---|---|---|---|
| FB-P1 | Land use and zoning | Local jurisdiction | Identical for both. Use, setbacks, noise, screening, and height are decided by the parcel, not by where the equipment was assembled. |
| FB-P2 | Utility interconnection | Utility / grid operator | Identical for both, and frequently the governing item. No delivery method shortens a queue position. |
| FB-P3 | Module plan review and in-plant inspection | State modular program or approved third-party agency | Unique to factory-built. ICC/MBI 1205 sets out permitting, in-plant and final inspection, third-party inspectors, and industrialized-building departments.[2] |
| FB-P4 | Site permit for foundations and connections | Local authority having jurisdiction | Still required. The AHJ permits the pad, utilities, and tie-ins, then inspects the assembled result.[3] |
| FB-P5 | Fire protection and life safety | Fire code official / AHJ | Detection and suppression for IT areas are reviewed locally under NFPA 75 wherever the enclosure was built.[6] |
| FB-P6 | Transport and route approval | State DOT / permit offices | Unique to factory-built. Oversize or overweight moves need routing, escorts, and bridge analysis before a date can be promised.[1] |
The advantage is repetition, not exemption: one plan set reviewed once and inspected in-plant serves many identical units, and local review narrows to the site scope. Where the jurisdiction has no off-site program, that advantage disappears.
The envelope
Site-built construction stays customizable late; factory-built construction is customizable early and expensive to change late. A field crew can absorb a redesigned electrical room in week thirty because the room is still a drawing and a stack of conduit. A production line cannot: the change propagates to tooling, test fixtures, approved plan sets, and units already in progress.
So the question is not “how customizable is it?” but “when does customization close?” Factory-built products configure inside a fixed envelope — rack layout, power topology, coolant interface, ambient rating. Requirements that break the envelope push the project back toward a custom building; a manufacturer that agrees to break its own envelope is selling site-built work at factory prices.
The hardest limit on factory-built infrastructure is not manufacturing capability. It is the public road. Federal Highway Administration standards on the Interstate System set 80,000 lb gross vehicle weight, 20,000 lb on a single axle, and 34,000 lb on a tandem axle; on the National Network no state may impose a width limitation of more or less than 102 inches, while height limits are left to the states.[1] Those figures are the real design constraint behind every standardized unit on the market.
Two strategies follow. The first stays inside the legal envelope — often at or near standard intermodal freight-container proportions — so delivery is ordinary freight on ordinary trailers. The second exceeds it and accepts oversize or overweight permitting: state-by-state routing, escorts, bridge analysis, curfews, and a date set by a permit office. Neither is wrong, but the second reintroduces on the highway the schedule variance factory production was meant to remove.
Placement is the second half. Units land by crane, which needs an approach, a laydown area, a load-rated pad, and overhead clearance. Tight infill sites, weight-restricted bridges, and sites with no crane standing room can rule out factory delivery on geometry alone — a constraint that never appears in a spreadsheet comparison.
When each wins
Run these in order. Any one of them can settle the decision on its own.
If construction dominates, factory production removes it from the critical path. If utility or environmental review dominates, the method barely moves the date.
Configurable-within-a-product favors factory-built. Bespoke floor plans, security zoning, or unusual redundancy favor site-built.
Factory-built turns late changes into change orders. Moving scope is safer built in place.
Inside the legal road envelope, transport is routine freight. Outside it, permits and escorts become schedule risk of their own.
A state modular program or accepted third-party agency preserves the approval advantage. Jurisdictions that review units as ordinary buildings erase it.
If the building and the service are already there, fitting out usually beats shipping new enclosures.
Repetition pays for tooling and process. One-off capacity rarely recovers that setup cost.
Factory delivery consolidates responsibility with one manufacturer; site-built spreads it across trades and a general contractor. Choose which counterparty you would rather hold.
Honest limits
Manufacturers rarely publish this list. It is short, and it is decisive when it applies.
Disclosure
PODOS builds on the factory side of this line, and the constraints above are the ones the product is designed around. Each PODOS Pod is designed as a standardized 1 MW building block and designed for 128 GPUs, with power and liquid cooling integrated and tested before the unit leaves the plant. Because the integration work happens on a line rather than on a pad, PODOS targets a 90-day window from order to commissioning for a standard unit — a target that assumes the site work and utility connection proceed in parallel, which is exactly the overlap this page describes and exactly the assumption a buyer should test against their own critical path.
The rest follows from that choice: the deployment sequence treats integration as cargo instead of construction, the platform composes capacity from repeated units, and the use cases that suit it best are the ones where units land inside the envelope above. Where a project fails the checklist — bespoke scope, an unreachable site, a campus large enough to amortize its own design — site-built is the better engineering answer, and we would rather say so here than at a kickoff meeting. Terms used on this page are defined in the AI infrastructure glossary.
QUESTIONS
The scope of work is nearly identical; its location moves. A site-built data center is assembled in place, trade by trade, under a local building permit. A factory-built one is produced on a production line, inspected in the plant, then transported and connected on a prepared pad. The difference is schedule overlap, inspection regime, and transport limits — not physics.
Yes. Factory-built units add an approval interface rather than removing one: plans reviewed once and inspected in-plant by a third-party agency or a state industrialized-building program, while the local authority having jurisdiction still permits foundations, utilities, and final connection. ICC/MBI 1205 describes that handoff.
Only the part of the critical path that is construction work. Unit production overlaps sitework and permitting, so integration and testing leave it. Utility interconnection, environmental review, and long-lead equipment do not move — if a grid connection governs the date, the delivery method changes little.
The road, not the factory. Federal Highway Administration standards set 80,000 lb gross vehicle weight, 20,000 lb on a single axle and 34,000 lb on a tandem axle on the Interstate System, plus a 102-inch width no state may set above or below on the National Network. Beyond those figures delivery becomes an oversize/overweight permit exercise.
When the requirement is bespoke, when a powered shell already exists, when the campus is large enough that one design amortizes across hundreds of megawatts, or when the route cannot accept a unit. Then conventional construction is the honest answer.
Bring the site, the schedule, and the route to the pad. If the checklist points to site-built, we would rather say so before a kickoff meeting.