---
title: 'Before you carry a data center number: a precon checklist'
url: 'https://piper-ai.com/resources/data-center-precon-checklist'
description: 'Ten questions a GC precon team should answer before carrying a data center number: owner gear, redundancy, commissioning, liquid cooling, schedule and subs.'
contentTypes: [guide]
publishedAt: '2026-09-29T12:00:00.000Z'
updatedAt: '2026-09-29T14:20:33.651Z'
author: 'Ido Gedanken, CEO'
readingTime: 7
---

# Before you carry a data center number: a precon checklist

1. On a data center, the number is only as good as the handoffs behind it: who furnishes, sets, connects, tests and carries each piece
2. Count electrical equipment from the redundancy topology, not from the megawatts
3. Commissioning, liquid cooling boundaries and bridging power are scope, and each needs a named owner
4. Tie escalation, storage, standby and damages exposure to the procurement dates that drive them
5. Coverage means crews that can staff the peak in your window, not proposals received

**Most data center numbers do not go wrong on quantities. They go wrong at a handoff nobody priced:**

- the owner-furnished UPS with no one assigned to terminate it;
- the load banks no bidder carried;
- the hoses between the facility piping and the cooling units;
- the switchgear date that moved and took four costs with it.

This checklist is ten questions to answer before a GC carries a number on a data center. Each one points at a handoff where money commonly goes missing.

It is for estimators, precon managers and directors on data center electrical, mechanical and general contract packages. It fits best on negotiated GMP and design-build programs for hyperscale and colocation owners, and still applies to enterprise work. Use it at the final review before the number goes out, alongside your normal [final bid review](https://piper-ai.com/resources/final-bid-review-qa-qc-general-contractors). It does not replace reading the documents. It tells you where to read hardest. Each question links to a longer guide.

## The ten questions at a glance

| Question | What goes wrong when it is skipped |
| --- | --- |
| 1. Is the job measured by MW and topology, not square feet? | Benchmarks and quantities are off before pricing starts |
| 2. Who does each step for every owner-furnished item? | Bids look comparable and are pricing different chains |
| 3. Does the owner's purchase include what installation assumes? | A missing bus, part or accessory becomes a change after award |
| 4. Is commissioning priced by level, with named owners? | Load banks, fuel and re-test days sit in no one's number |
| 5. Who owns each liquid cooling boundary? | Hoses, valves, flushing and fluid fall between three parties |
| 6. Is escalation carried to each item's purchase date? | Long-lead gear is priced at a date it will not be bought |
| 7. Is the wait priced in both directions? | Early gear needs storage, late gear leaves crews standing |
| 8. Which milestones carry damages, per hall? | One slip triggers exposure the estimate never carried |
| 9. Can each bidder staff the peak in your window? | A complete price from a sub that cannot perform |
| 10. What changed since the last building? | A revision reopens a purchase date nobody traced |

## 1. Is the job measured by MW and topology, not square feet?

Data centers are sized, sold and leased in megawatts of IT load. Cushman & Wakefield's 2026 cost guide puts the US greenfield average at $17.6 million per MW, excluding servers, with power infrastructure as the largest line. A square-foot benchmark from office work misprices both the enclosure and what goes inside it.

Then check the redundancy topology. N+1, block redundant ("4 to make 3") and 2N designs of the same megawatt rating carry very different amounts of switchgear, UPS, feeders and terminations. Count lineups from the one-line diagram, and confirm every bidder counted the same ones.

**Good looks like:** a scope basis built from the one-line, the topology and the owner's standard. For the full comparison, see [data center precon is not commercial precon with more power](https://piper-ai.com/resources/data-center-preconstruction-vs-commercial).

## 2. Who does each step for every owner-furnished item?

When the owner buys generators, switchgear, UPS and cooling units directly, the subcontract bid becomes a price for a chain of steps. Receive, store, rig, set, assemble, terminate, start up, support functional testing, support the integrated systems test, handle warranty.

**Good looks like:** each owner-furnished item as a row, each step as a column, and each bidder marked included, excluded, qualified or silent for every cell. Price the missing steps before comparing totals. The method is in [leveling bids when the owner buys the equipment](https://piper-ai.com/resources/data-center-owner-furnished-equipment-bid-leveling).

## 3. Does the owner's purchase include what installation assumes?

The owner's procurement team buys against its own specifications and schedule. Those do not always match the construction documents. Uptime Institute has described an owner that bought a UPS without the bus needed to connect it. Loose parts, shipping-split connections, accessories and spares are the usual gaps.

**Good looks like:** every owner-furnished item checked against the installation drawings for connection points, accessories and parts. Every gap assigned before the leveling sheet is closed.

## 4. Is commissioning priced by level, with named owners?

On a data center, commissioning proves the redundancy works, and contract milestones are increasingly tied to it. The final level, the integrated systems test, runs the whole building at design load on load banks with the utility dropped.

**Good looks like:** a commissioning build-up by level and by system, with named owners and crew-days. It should cover:

- load banks and their technicians;
- test fuel and refill;
- temporary power;
- manufacturer field days beyond startup;
- controls integration between the BMS and power monitoring;
- a stated number of re-test days.

See [commissioning is scope](https://piper-ai.com/resources/data-center-commissioning-scope-sub-bids).

## 5. Who owns each liquid cooling boundary?

Direct-to-chip cooling splits the water into a condenser loop, a facility loop and a clean technology cooling loop to the racks. The mechanical contractor, the cooling unit vendor and the owner's IT integrator each own a part. The commonly unassigned items are:

- the connection between facility piping and the cooling unit;
- the hoses and quick disconnects at the racks;
- flushing, first fill and fluid quality;
- water treatment while the loop sits idle before IT load.

**Good looks like:** a segment-by-segment responsibility matrix resolved in words in the solicitation. See [who owns the liquid cooling connection](https://piper-ai.com/resources/data-center-liquid-cooling-scope-boundaries).

## 6. Is escalation carried to each item's purchase date?

Long-lead gear is often released before the design is finished. Wood Mackenzie's surveys, reported by POWER and Data Center Knowledge, put average power transformer lead times at about 128 weeks in 2025 and substation transformers above 160 weeks in 2026. The same surveys put power transformer prices up 77 percent since 2019. Escalation belongs to the date each item is actually bought, not to the bid date.

**Good looks like:** a procurement schedule inside the estimate, with each long-lead item carrying its release date and the cost lines it drives.

## 7. Is the wait priced in both directions?

Gear that arrives before its room is ready has to be stored, protected, maintained and re-inspected. Gear that arrives late leaves crews and supervision waiting. Both are real costs, and both exist only because of the gap between two dates.

**Good looks like:** storage and re-inspection carried against the early side of each delivery window, standby and extended general conditions against the late side. Include bridging power if the contract makes the GC responsible for a date the utility controls.

## 8. Which milestones carry damages, per hall?

Data center contracts increasingly tie liquidated damages to energization, integrated testing and each data hall's handover rather than to a single completion date. Law firms advising on these contracts, including Pillsbury and Bracewell, recommend exactly that. Damages are built from the owner's real losses, including rent. Because the owner's exposure to its tenant is hard to pass down in full, the GC often keeps a gap.

**Good looks like:** damages exposure priced per milestone and per hall, with the retained exposure after flow-down known. See [on a data center, the schedule is part of the estimate](https://piper-ai.com/resources/data-center-schedule-risk-estimate) for how questions 6 to 8 connect.

## 9. Can each bidder staff the peak in your window?

The qualified electrical and mechanical market for large data centers is small and selective. Public MEP contractors describe planning labor before they bid and taking only the work they can staff. A proposal can be complete and priced from a firm that cannot put the crews on site when the schedule needs them.

**Good looks like:** each bidder's manpower loading, supervision depth and open commitments reviewed like a line item. When only one or two bids arrive, level against the project scope basis and your own history from similar buildings. See [bid coverage when only a few subs can do the work](https://piper-ai.com/resources/data-center-subcontractor-bid-coverage).

## 10. What changed since the last building?

Owners repeat a design standard across a campus, then revise it as computing hardware changes. A move from air to liquid cooling, a higher rack density or a colder supply-water requirement changes equipment selections. That changes procurement dates, and everything downstream of them moves too.

**Good looks like:** each new building or work authorization priced as a revision of the last one, with the owner's changes traced to the scopes, bidders, dates and cost lines they touch. The general version of this problem is in [how drawing revisions cascade through scope, bids and the estimate](https://piper-ai.com/resources/drawing-revisions-cascade-scope-bids-estimate).

## How to use the checklist

Run it twice. Run it once at the start of a pursuit, when the answers shape the solicitation and the scope basis. Run it again at final review, when every "no one" and "not sure" becomes either a priced carry, a clarification or a named qualification in the proposal to the owner.

A question the team cannot answer is not a failure. A question nobody asked is how a number goes out with a hole in it.

## Where Piper fits

Every question on this list is about a connection between documents:

- the owner's equipment list and the one-line;
- the commissioning plan and the schedule;
- the procurement dates and the estimate;
- each proposal and the scope basis.

The answers usually exist somewhere on the project. The risk is that no one sees them together during bid week.

Piper is the AI operating system for preconstruction. It builds one understanding of the project from the drawings, specifications, owner standards, equipment lists, proposals, contract terms and the estimate. It uses that understanding across scope, leveling and final review. For a checklist like this one, that means showing where each question is answered in the documents, where it is not, and which bids, carries and dates depend on the answer, with the source behind each finding. The team still decides what to carry.

### What should a GC check before submitting a data center bid?

Check the handoffs: who performs each step for owner-furnished equipment, whether redundancy was counted from the topology, who pays for commissioning support and load banks, who owns liquid cooling boundaries, how escalation and delays are tied to procurement dates, and whether each subcontractor can staff the peak.

### How is data center estimating different from commercial estimating?

Data centers are measured per MW of IT load, most of the cost sits in power and cooling systems, major equipment is often owner-furnished, commissioning is priced scope, and the schedule is driven by utility power and equipment lead times rather than by the building.

### When should a precon team run a data center checklist?

Twice. Early, to shape the solicitation and scope basis, and at final review, to turn every open question into a priced carry, a clarification or a named qualification before the number goes out.

## Sources

- Cushman & Wakefield, [2026 Data Center Development Cost Guide](https://ir.cushmanwakefield.com/news/press-release-details/2026/Cushman--Wakefield-Releases-2026-Data-Center-Development-Cost-Guide-Citing-21-Rise-in-Per-MW-Construction-Costs/default.aspx).
- Schneider Electric, [White Paper 75, Comparing UPS System Design Configurations](https://it-resource.schneider-electric.com/white-papers/wp-75-comparing-ups-system-design-configurations).
- Uptime Institute, [Avoiding data center construction problems](https://journal.uptimeinstitute.com/avoiding-data-center-construction-problems/).
- POWER, [Transformers in 2026](https://www.powermag.com/transformers-in-2026-shortage-scramble-or-self-inflicted-crisis/), and Data Center Knowledge, [AI data center boom rewires US power supply chain](https://www.datacenterknowledge.com/build-design/ai-data-center-boom-rewires-us-power-supply-chain) (Wood Mackenzie lead times and prices).
- Pillsbury, [Data center construction contracting](https://www.gravel2gavel.com/data-center-construction/), and Bracewell, [Managing contractual risk in data center construction](https://www.bracewell.com/resources/managing-contractual-risk-in-data-center-construction/).
- Fenwick Elliott, [Common issues in data centre construction](https://www.fenwickelliott.com/knowledge-hub/annual-review/ar-2025/common-issues-in-data-centre-construction-and-how-to-avoid-them/).
- Comfort Systems USA, [Q1 2026 earnings call](https://www.investing.com/news/transcripts/earnings-call-transcript-comfort-systems-usa-q1-2026-earnings-outshine-expectations-93CH-4636354) (labor planning before bidding).