Program & Project Management: The Integrated Master Schedule & Critical Path

Part 1 decided what to build and whether the economics work. This chapter is where the abstraction ends and the calendar begins. An AI data center is a program with a deadline set by physics and supply chains, not by the owner’s ambition: the day the cluster can take its first synchronous training step, or serve its first revenue token, under the workload’s acceptance conditions. Everything upstream of that day is a race, and every decision about how to run it has consequences denominated in time. In the guide’s contested bear case the GPUs earn for as little as 2–3 years, so time converts directly into money. → the depreciation clock that prices every lost month is in Chapter 1.8.

This chapter applies that frame to schedule. It covers the phase-gate lifecycle and the build as a time-to-first-train race; the Integrated Master Schedule (IMS) and the critical path across three tracks that move at different speeds; schedule risk quantified with Monte Carlo and the P50/P90 dates the long poles force on you; the owner’s project controls (earned value, milestone deposits, change orders and claims); the facility and cluster schedules bound by integration milestones; and the stage-gate governance and assumptions/decisions register that record what the program is betting. In a power-bound, allocation-constrained market the schedule is the project, and the long poles are not the ones a traditional general contractor watches.

Keep the cash-flow milestones distinct. Utility energization can start utility billing, accepted facility service can start a landlord’s rent, and accepted compute service can precede billable use. Bind each to its own contract and signatories in the IMS.

A data-center program moves through a recognizable sequence: scope and design basis → site control and entitlement → interconnection and power → procurement → construction → commissioning → go-live → operations. The mature way to govern it is a phase-gate (stage-gate) model. Each phase ends in a gate where capital is released, assumptions are tested, and the program advances, holds or is killed. The gate exists to make the irreversible commitments explicit and to put a named owner and a dated decision on each one before the money leaves. → the reversible-vs-irreversible discipline this inherits is set in Chapter 1.1.

An AI build differs from a 2018 enterprise data center because the gates are no longer evenly spaced. In a power-bound market the early gates, interconnection and long-lead procurement, release the commitments that set the finish date. The late gates (fit-out, commissioning) govern execution against the dependency path the current IMS identifies as governing. The construction industry’s instinct is to gate on design maturity. An AI program has to gate time-critical packages on power-service evidence and allocation evidence before the whole design is mature, once each released package’s interfaces are stable. Otherwise you arrive at a finished building with no megawatts and no GPUs.

So the phase-gate model has to be re-weighted. Front-load the gates that release time-critical deposits, and accept that you are committing capital against assumptions you have not yet fully retired. On a permit-controlled path, an earlier transformer deposit cannot recover the permit date. Before releasing a package, require its approved interfaces and the remaining risk owner alongside the schedule benefit.

The Integrated Master Schedule is the single time-logic network that ties every deliverable, dependency and milestone into one critical-path-method (CPM) model. The mistake that defines failed AI programs is running it as one undifferentiated Gantt. An AI data center is really three schedules braided together, each governed by different physics and a different supplier ecosystem, each with its own critical path:

• The power track: interconnection studies and agreements, the utility’s grid upgrades, the substation, load-serving HV/substation and medium-voltage transformers (plus GSUs only where generation/export is in scope), switchgear, and increasingly on-site or behind-the-meter generation as a bridge. The owner cannot compress most of these lead times: load-serving substation transformers run above 160 weeks in Wood Mackenzie’s May 2026 US assessment (its Q2 2025 survey averaged 128 weeks for large power transformers; generator step-ups ~143 weeks). Power-track readiness also depends on the host utility’s project-specific service studies, upgrades, agreements and energization milestones. This track often governs, but the configured dependency network, not the track label, identifies the critical path to a common end milestone.
• The building track: entitlement and permits (the air permit is a recurring long pole where on-site gas is involved), earthworks, shell, mechanical/electrical/plumbing, and the cooling plant. Mastt’s April 2026 practitioner estimate puts greenfield hyperscale construction at 18–24 months. Define the track’s start and completion events in the IMS, and compare it with the power track at the same required milestone before assigning float.
• The IT / cluster track: the GPU allocation (a slot, not a purchase, negotiated quarters ahead), CoWoS/HBM-gated accelerator delivery, network fabric, storage, structured cabling, then rack-and-stack, fabric validation, burn-in and the reference run. Allocation gates this track, not the owner’s cash. → the allocation game lives in Chapter 2.3; the HBM constraint behind it in Chapter 7.6.

Factory-built modules can shorten the building track. SemiAnalysis (July 2026) estimates that a fully modular, liquid-cooled AI hall reaches IT-ready in about 24–30 months including permitting, where a stick build takes 30–35 or more. The modular hall also comes in about 8% cheaper per MW. Meta needed two to three years for the first five permanent buildings at its Ohio campus. Its fabric-shelled halls there went up in a fraction of that time, though the tents shortened the shell, not the whole facility schedule. In exchange the owner accepts a hall that is harder to re-lay for future equipment and, on some fast schedules, a squeezed commissioning phase that pushes risk into operations. Operators and MEP contractors also report reliability problems with the modules themselves, bad enough to give back the time they saved. Modules and fast shells speed the building, not the interconnection, so they pay only where the building path drives the date.

The IMS exists to expose the float between these tracks and the integration milestones where they must meet. Float is the schedule’s shock absorber. Each track’s float comes from the project dependency network and the selected supply and permitting path, and the discipline is to spend it deliberately, sequencing fit-out to land just-in-time against energization, rather than letting it evaporate into early-but-idle completion. The program slips when the activities driving productive service slip, even while the team celebrates a floating building task finishing early on a slab with no power.

Read the table as a sequencing problem. Calculate which dependency path drives the common first-productive-run milestone, then measure float on the others. Check the lowest-float path as well as the longest one: an imposed contract date can leave a shorter path with less total float than the longest path. Power governs on many 2026 sites, but if a permit or building path governs, an earlier transformer delivery cannot recover that date. The cluster track cannot start meaningful integration until the powered-shell handoff, and then carries a bring-up tail that inexperienced owners forget to schedule. Direct acceleration spending to the activities that move productive service.

A deterministic CPM schedule produces one finish date from point-estimate durations, and that date carries no confidence level. The mature program runs a quantitative schedule risk analysis (QSRA), which calculates the date’s probability and the contingency needed for the confidence you select. Assign each activity a duration distribution (typically three-point: optimistic, most likely, pessimistic), model the correlations (a transformer delay and a switchgear delay are not independent, because they share a strained supply chain), and run a Monte Carlo over the network a few thousand times. The output is a distribution, not a date. The two numbers that matter are the P50 (the date you have a coin-flip chance of beating) and the P90 (the date you are 90% confident of beating).

The gap between P50 and P90 is dominated by a handful of long poles with long right-tails: the load-serving HV/substation transformer (and a GSU only where generation/export is in scope), the grid interconnection energization date, the air permit where on-site generation is in scope, and the GPU/HBM allocation. These are not normally distributed. Their failure modes move the date a lot, not a little: a transformer factory slot slips a quarter, a required utility study or upgrade milestone slips, an air-permit challenge adds eighteen months. A gate you have cleared can also reopen: Wisconsin regulators revoked the completeness determination for ATC’s Ozaukee County interconnection application serving the Lighthouse campus, and ATC refiled on 2026-09-15 with a new preferred route, dropped facilities and an added E-STATCOM, still subject to Wisconsin DNR review (ATC). Keep cleared permitting and interconnection gates on the risk register until the approvals are final. A schedule whose P50–P90 spread is six months is telling you that one of these poles can eat two quarters of revenue, and the deposit on that pole goes out the door before the design is frozen.

Cedar’s critical path is a dependency calculation. In weeks, dock = release + manufacture + common delay + shipping; building = baseline readiness + permit delay; installation finish = max(dock, building) + install duration. Facility acceptance = max(utility readiness + common delay, installation finish) + acceptance duration. Racking = max(IT delivery, building) + rack duration; productive date = max(facility acceptance, racking) + validation duration. Multiply independent event weights, then accumulate the completion distribution. The first date that reaches the selected cumulative probability is its percentile.

Release the package by the date in the result and keep the permit branch in the promise. The no-disruption path runs through manufacture, shipping, installation, facility acceptance and cluster validation. Utility float is measured to installation finish, so adding another contingency for the same supply event double-counts it. Releasing later also moves the crossover: with the dock two weeks later, the building path needs a longer permit delay before it takes control. Once building readiness falls later than dock delivery, accelerating the transformer cannot recover the building path.

The GAO Schedule Assessment Guide supplies the dependency and risk method. Placing the package order is covered in 2.3, custody and remedies in the contract in 2.4, and cluster acceptance in 13.10.

A schedule you cannot measure against is a wish. Project controls is the owner-side discipline that turns the IMS into a steering instrument: a cost-and-schedule baseline, periodic measurement of progress against it, and a forecast that updates honestly. The backbone is earned value management (EVM). It compares the budgeted cost of work performed (BCWP/EV) against the budgeted cost of work scheduled (BCWS/PV) and the actual cost (ACWP/AC) to derive a schedule performance index (SPI) and cost performance index (CPI). On an AI build, EVM forces physical-percent-complete discipline and produces an estimate-at-completion early enough to act on, instead of a surprise at the end.

But EVM was built for labor-and-materials projects, and an AI data center’s cost is dominated by a few enormous milestone-deposit equipment orders: the transformer, the switchgear, the turbines, the GPU allocation. They are paid against vendor manufacturing milestones, not installed progress, and that breaks naive EVM. Booking the full PO value as “earned” on deposit overstates progress; booking nothing until delivery understates it for two years. The owner’s controls function has to track a commitment/cash curve alongside the EVM curve: when each deposit is contractually due, what it secures (a factory slot, a queue position), and what its forfeiture costs if the program pivots. On AI builds the deposit schedule, not the construction draw, is the dominant near-term cash event. → deposit and slot-reservation instruments in Chapter 2.3; the contract that governs them in Chapter 2.4.

Change-order and claims management is the other half. A GPU-generation change during design (NVL72 to a denser successor) can change cooling duty, floor loading and busway ratings; an interconnection re-study can move the energization date and change the fit-out sequence. Each change is a fork with a schedule and cost consequence. The owner who has not stood up a disciplined change-control board on day one ends up litigating those consequences as claims at the end. The cheap move is a tight baseline plus a fast, well-documented change process; the expensive move is a loose baseline that turns every density surprise into a dispute.

The chronically under-managed seam in an AI build is the boundary between the facility (the powered, cooled shell, delivered by the construction and MEP world) and the cluster (the GPUs, fabric and software, delivered by the IT and platform world). These are two organizations, two cultures, two schedules and two definitions of “done”, and the project lives or dies on how cleanly they are bound. The right structure is an explicit two-track schedule with a small set of named integration milestones where the tracks hand off, each with an unambiguous entry/exit gate and an owner. → the powered-shell delivery model that creates this seam is in Chapter 2.2.

The integration milestones that bind the two tracks, in order:

• Powered-shell handoff. The facility delivers a hall with conditioned space, structural floor capacity, and the power and cooling distribution stubbed to the white space, but not yet energized to the rack. This is the contractual seam between base-building and IT fit-out, and the cleanest place to split scope and risk.
• Energization (power-on). Permanent power is live from the medium-voltage switchgear through the unit-substation transformers to the low-voltage busway and rack PDUs, and UPS and any on-site generation are commissioned (L3/L4). Until this gate the cluster track cannot draw load, and it is the most common place for the power track’s slip to surface as a cluster-track delay. → electrical acceptance in Chapter 13.3.
• Water-on / cooling-ready. The facility cooling loop and CDUs are flushed, leak-checked, balanced and proven to spec. This is non-negotiable before energizing liquid-cooled racks, because coolant temperature and flow must stay inside each installed rack’s OEM operating range. → CDU commissioning in Chapter 13.5.
• Integrated systems test (L5 IST). The facility proves it carries simulated full IT load and transfers cleanly to backup when utility power is lost. Commissioning should also show that the load rides through a normally cleared voltage sag without disconnecting, something a transfer test cannot demonstrate. Where the grid operator’s rules require ride-through, this test is mandatory. For a liquid-cooled AI hall, set the Level-5 duration from the approved test scripts, phasing, witness plan, defect-correction allowance and retest scope. → IST in Chapter 13.6.
• Cluster burn-in and the reference run. Now the IT track owns the clock: node diagnostics, fabric BER validation, a burn-in campaign selected by the signed test plan and measured defect-discovery rule, and a reference training/inference run at its target goodput. This is first-train. → burn-in in Chapter 13.8; cluster-scale validation in Chapter 13.9.

Making these milestones explicit matters because the seam is where finger-pointing lives. When the building is “done” but the cluster is not earning, the question is always whose milestone slipped. A program with named integration gates and per-gate owners answers it in a stand-up; a program with one Gantt answers it in a claim.

The phase-gate model only protects the program if the gates actually gate something irreversible. The governance question is concrete: at which board approval is each one-way-door commitment released? The interconnection-study deposit is committed before any building exists. Under PJM’s Expedited Interconnection Track, accepted by FERC on June 9, 2026 and effective July 31, 2026, that means a $500k study deposit that turns non-refundable once the generation request is valid, plus a $15k/MW readiness deposit; for the campus load itself, it means the host utility’s own study fees and the security behind its service agreement. The HV transformer PO reserves a delivery slot against US substation-transformer lead times reported above 160 weeks (Wood Mackenzie, May 2026). The GPU allocation reservation commits a slot quarters ahead of silicon that is itself CoWoS/HBM-gated. Each of these is capital released against assumptions that have not been fully retired, which is why the gate exists: to force the board to look at the assumption, name its owner, and accept the bet on the record.

The artifact that makes this auditable is the assumptions-and-decisions register, the schedule-and-commercial analogue of the design-basis document from scoping. For every load-bearing assumption (the energization date, the transformer delivery date, the GPU generation the cooling plant is sized for, the contracted-vs-merchant power split the financing assumes), it records what was assumed, who owns it, when it must be confirmed or becomes a risk, and which downstream commitments depend on it. When a long pole’s tail fires, say a transformer slips a quarter, the register tells you in minutes which downstream dates and deposits move and who has to be told. Without it, the same event becomes a forensic reconstruction conducted under deposition.