On a data center, the schedule is part of the estimate

On a data center, schedule risk and cost risk are the same risk, and precon teams that price them separately end up counting it twice or not at all. Consider one date: the release of a medium-voltage switchgear lineup. It decides the escalation on the gear. It decides whether the gear waits in a warehouse or the crews wait for the gear. It decides whether energization, integrated testing and a data hall's handover hit the dates that carry liquidated damages. On most estimates those four consequences live in four places, owned by four people. On a data center they should be tied to the one date that drives them.

This piece is for precon managers and directors, estimators and schedulers at GCs pursuing data center work, typically negotiated GMP or design-build programs for hyperscale and colocation owners. The underlying point applies to any job with long-lead equipment. Data centers make it impossible to ignore, because the long-lead equipment is most of the job.

The building is the short pole

On a commercial job the schedule risk sits inside the building: trade sequence, weather, inspections. On a data center the building is often the fastest thing on the project. The slow things are outside it.

• Power transformers: an average of 128 weeks in Wood Mackenzie's 2025 survey, as reported by POWER. Generator step-up units averaged 144 weeks.
• Substation transformers: more than 160 weeks in Wood Mackenzie's 2026 update, as reported by Data Center Knowledge. Switchgear there is "closer to one year."
• Grid connection: JLL reports waits of more than four years in primary markets. Dominion Energy warned in 2024 that large connections in Virginia could take up to seven years, as Bloomberg reported.

The industry's own confidence matches those numbers. In Turner & Townsend's 2025 data center cost index, 48 percent of respondents named power as the top obstacle to schedule, and only 19 percent were confident suppliers would meet delivery dates. A Financial Times analysis reported by Network World in April 2026 found that nearly 40 percent of data center projects due to open that year would be at least three months late.

The consequence for precon is the order of operations. Long-lead gear is released before the design is finished, often under a limited notice to proceed. Owners now reserve factory capacity the way they used to reserve land. Schneider Electric, for example, has announced supply capacity agreements with data center owners. By the time the estimate reaches a GMP, some of the most important dates on the job are already set, and the estimate has to price around them.

What moves the date, and where it lands in the estimate

The published delay analyses, from law firms such as Fenwick Elliott to consultants and trade press, converge on the same short list of drivers. Each one shows up in a specific part of the number.

Each row has a real counterpart on the record:

• Power. Meta's first Entergy gas plants for its Louisiana campus are scheduled for 2028 and 2029, years after construction began (Data Center Dynamics). A New Jersey data center developer running generators while it waited on power was fined more than $1 million for operating dozens of them without a permit, according to New Jersey Monitor.
• Permits. Prince William County's Digital Gateway rezoning was voided in court and the developer dropped the project in 2026. Data Center Watch counted more than $152 billion of projects blocked or delayed by local opposition in 2025.
• Design churn. Meta paused roughly a dozen buildings in late 2022 to redesign them for liquid-cooled AI hardware, and Microsoft paused part of its Mount Pleasant, Wisconsin site to reassess scope in light of changes in technology (Data Center Dynamics).

None of these are construction mistakes. They are dates the GC does not control, landing on costs the GC does carry.

One date, four costs

Go back to the switchgear lineup. Its release and delivery date drives four costs, and on most estimates each sits in a different place.

Escalation. The price the gear will cost depends on when it is bought, not when the job is bid. Wood Mackenzie's figures, reported by POWER, show power transformer prices up 77 percent and medium-voltage switchgear up 50 percent since 2019. On contractor-furnished gear, escalation belongs to the purchase date of each item. On owner-furnished gear, the same date still drives the installer's labor escalation and crew availability.

Storage. If the gear arrives before its room is ready, it has to be stored, climate-controlled, maintained and re-inspected. That is a real cost with its own line, and it exists only because of the gap between two dates.

Standby and extended general conditions. If the gear arrives late, the crews and supervision wait. On a large MEP package the time-related costs of a month of delay are substantial, and there is no published benchmark for them on data centers. Every team has to build its own from its own general conditions.

Damages exposure. If the gear is late enough to push energization, integrated testing or a hall's handover, it reaches the milestones that carry liquidated damages.

Now look at who usually prices each one. Escalation sits in the estimate's markup or an allowance. Storage is in general conditions, if anywhere. Standby is in the schedule contingency or nowhere. Damages exposure is in the contract review or the risk register. When the switchgear date moves, four people have to notice, and each has to adjust a number whose connection to the others is recorded nowhere. That is how the same month of risk gets carried twice in one place and missed in another.

How the contract turns the date into money

Data center contracts make the time-cost link explicit, and the terms are moving in a clear direction.

Liquidated damages follow the owner's rent. Bracewell describes data center liquidated damages as typically thousands to tens of thousands of dollars per day. The rate is built from the owner's real losses: extended costs, financing and lost revenue or replacement space. To see why owners care, run CBRE's published primary-market rent, about $196 per kW per month in its second-half 2025 report, across one megawatt. That is roughly $196,000 of rent per MW per month of delay. Treat it as an upper bound, since hyperscale rents for large blocks are lower and CBRE does not publish them, but the order of magnitude explains the contract terms.

They attach to milestones, not completion. Pillsbury and Bracewell both advise tying damages to energization, integrated systems testing and each data hall's turnover, measured against redundancy and uptime criteria. A job can carry several damages dates, and a slip early in the sequence can reach more than one of them. That is why commissioning is priced scope on these jobs.

The GC keeps a gap. Fenwick Elliott notes that an owner's exposure to its tenant is much higher than on a typical construction project and hard to pass down, so contracts are rarely back to back. Munich Re sells insurance aimed at the difference between the damages an owner pays its tenant and what it recovers from its contractor. Seyfarth, writing in Construction Dive, warns the other direction: damages rates set too high get priced into subcontractor and supplier bids.

Relief has to be written in. Late owner-furnished equipment, utility delay and government action do not relieve the contractor by default. Pillsbury advises spelling out schedule relief and compensation for late or defective owner equipment. Kilpatrick Townsend warns about force majeure clauses that grant time but no money, which leave the GC's time-related costs running. For the equipment side, see leveling bids when the owner buys the equipment.

One widely repeated figure is best left alone. A delay cost of "$2 million per day" circulates in articles about data center delays, attributed to unnamed industry estimates. We could not find a documented basis for it. If a delay figure matters to a pricing decision, build it from the owner's actual terms and your own costs.

What precon should do with it

None of this needs a new estimating method. It needs the estimate and the schedule to share their dates.

1. Make the procurement schedule part of the estimate. Every long-lead item gets a release date, a delivery window and the cost lines it drives, in one place.
2. Escalate to the purchase date of each item, not to the bid date or the start date.
3. Price the wait in both directions. Storage and re-inspection if gear is early, standby and extended general conditions if it is late, each tied to the date that creates it.
4. Price damages exposure per milestone. Read which milestones carry damages, price the exposure per hall where turnover is phased, and know what exposure the GC retains after flowing terms down.
5. Model cost and schedule risk together. AACE Recommended Practice 57R-09 describes integrated cost and schedule risk analysis driven by risk drivers on a CPM schedule. Even a simple range on the few critical long-lead items is better than a separate schedule contingency with no dates behind it.
6. Decide bridging power explicitly. If the contract makes the GC responsible for a date the utility controls, temporary generation, fuel, permits and removal are scope. Price them or exclude them by name.
7. Re-run it at every revision and work authorization. On a multi-phase program, each new building or phase is a chance to reprice. It is also a chance to discover that a date moved.

Design churn is schedule risk too

The date that moves is not always a delivery. Owners change their designs as computing hardware changes. A move from air to liquid cooling, a higher rack density or a colder supply-water requirement changes equipment selections, which changes procurement dates, which changes everything downstream of them. ASHRAE has already warned that future processors may need colder facility water than today's designs assume.

A revision on a data center therefore has a larger blast radius than on most buildings. It is not only a scope and cost change. It can reopen a procurement decision that set a critical-path date, as drawing revisions do to scope, bids and the estimate on any job, with the added weight of a 100-week lead time. This is one more reason the estimate is becoming a construction plan rather than a number with a schedule attached.

Where Piper fits

The failure described here is not a missing skill. It is a missing connection: a date in the procurement log that should move four numbers in the estimate, a damages clause that should change how commissioning is carried, a revision that should reopen a purchase decision. Each connection is obvious once someone sees it. The risk is that nobody holds all of them at once during a bid.

Piper is the AI operating system for preconstruction. It maintains one understanding of the project across the drawings, specifications, owner standards, equipment lists, proposals, contract terms and the estimate, and uses it to drive scope, leveling and review. When a delivery date, a milestone or a design revision changes, Piper traces where that change lands: which scopes, bidders, assumptions and cost lines depend on it, with the source behind each. The team still decides how to price the risk. It just stops relying on one person remembering every dependency.

Sources

• POWER, Transformers in 2026: shortage, scramble or self-inflicted crisis? (Wood Mackenzie 2025 lead times and price changes).
• Data Center Knowledge, AI data center boom rewires US power supply chain (Wood Mackenzie 2026 update).
• JLL, Data Center Outlook (grid connection timelines).
• Bloomberg, Data centers face seven-year wait for power hookups in Virginia (August 2024).
• Turner & Townsend, Data Centre Construction Cost Index 2025 (power and supplier delivery confidence).
• Network World, Almost 40% of data center projects will be late this year (reporting a Financial Times analysis, April 2026).
• Schneider Electric, supply capacity agreement with Switch (reserved manufacturing capacity).
• Data Center Dynamics, Entergy approval for gas plants serving Meta's Louisiana data center, Meta's AI data center redesign and Microsoft pauses construction at Mount Pleasant.
• New Jersey Monitor, Vineland data center fined (September 2026).
• Virginia Business, Prince William Digital Gateway project officially dies, and Data Center Watch, Q3 to Q4 2025 report.
• Fenwick Elliott, Delay claims in data centre construction, part 1 and Common issues in data centre construction.
• Bracewell, Managing contractual risk in data center construction, and Pillsbury, Data center construction contracting (liquidated damages and milestones).
• Construction Dive, Tips for owners and contractors on data center contracts (Seyfarth; master agreements and damages rates).
• Kilpatrick Townsend, The unique schedule risk of government action in data center construction (force majeure and change in law).
• Munich Re, Liquidated damage cover for data centers.
• CBRE, North America Data Center Trends H2 2025 (primary-market rental rates).
• AACE International, Recommended Practice 57R-09, Integrated Cost and Schedule Risk Analysis Using Risk Drivers and Monte Carlo Simulation of a CPM Model.
• ASHRAE, Emergence and expansion of liquid cooling in mainstream data centers (future facility water temperatures).