What each model actually does
EPCM (engineering, procurement, construction management). The contractor acts as the owner's agent. It designs or manages the designer, runs procurement, and manages construction, but the owner signs the trade contracts and carries the price risk. Fee is usually a percentage or fixed amount for the management service. Overruns land on the owner.
Progressive design-build. One entity holds design and construction under a single contract. Phase one is a fee-based design and preconstruction agreement. Once design is developed enough, often somewhere around 30 to 60 percent, the parties agree a guaranteed maximum price or lump sum for phase two. If they cannot agree, the owner can walk away with the design.
Integrated project delivery (IPD). A multi-party agreement among owner, designer, and builder, with profit at risk against shared project outcomes and mutual waivers of claims. It is the most collaborative model and the hardest to execute. Many contracts called IPD in practice are looser than the textbook version.
Where the risk lands
| Risk | EPCM | Progressive design-build | IPD |
|---|---|---|---|
| Design errors | Owner and designer | Design-builder | Shared, with limited claims |
| Cost overrun after pricing | Owner | Design-builder, above GMP | Shared through risk pool |
| Trade interface and coordination | Owner carries most | Design-builder | Shared |
| Early price certainty | Low, firms up with each package | Low until GMP | Target cost, firms up over time |
| Schedule performance | Owner, with CM influence | Design-builder, with liquidated damages | Shared incentives |
| Owner control over trade selection | High | Moderate, through open-book buyout | High |
Nothing in that table is free. EPCM gives you control and price transparency, and you pay in risk and in the staff you need to hold it. Progressive design-build moves more risk to the builder, and the builder prices it, particularly on a campus where power delivery and equipment allocation sit partly outside its control.
Cost predictability: when does the number become real?
Every model has a moment when the price hardens. The sooner you reach it, the earlier you can lock financing and customer commitments. The sooner you reach it with incomplete design, the more contingency sits in the number.
Under traditional lump-sum EPC, the number is fixed at award, but only after a long design and bidding period, and change orders erode it. Under EPCM, the number hardens package by package as trades are bought out. Under progressive design-build, it hardens at the GMP, and the risk is a rushed GMP agreed on thin design. Set a GMP deadline in the contract, tied to a defined design maturity, and attach consequences if the parties miss it.
On a multi-gigawatt program, price discipline comes from owner-furnished long-lead equipment, indexed escalation on copper, steel, and electrical gear, and tiered contingency held separately by design, construction, and owner. Those mechanisms work under any model. Keep them out of the builder's lump sum where you can, and the builder stops pricing risks it cannot control.
Schedule speed
Fast-track EPCM often starts earliest, because early works packages can be released before design finishes. Progressive design-build is close behind, since the builder is engaged during design and can start procurement and early site work under phase one. Traditional EPC with a competitive bid is typically slowest to mobilize, because design has to be complete enough to bid.
Speed only counts if the long-lead equipment is already moving. A contract that mobilizes in month three while the transformer queue starts in month nine has bought nothing. See OEM Capacity Reservations and Supply Chain Capital Protection.
Multi-site rollout changes the answer
On a single campus, you can choose based on that site. Across five or ten sites, repeatability becomes the main driver.
A workable structure is a master agreement with a design-builder or EPCM partner for a reference design, followed by site-specific work orders. The first campus pays the learning cost. Later campuses should show falling unit costs and tighter schedules, and the contract should say so with agreed benchmarks, shared savings, and the right to bring in a second partner if performance lapses.
Guard the competitive tension. Repeat awards without a price check drift upward. Require open-book accounting, competitive subcontract buyout (three bids per major package is a common standard), and periodic market testing against an independent estimate.
Clauses that carry the weight
- Definition of cost of the work. List what is reimbursable and what sits in fee. Ambiguity here is where open-book contracts turn into arguments.
- Fixed fee and general conditions caps. Set them at award. Percentage fees on a growing cost base reward the wrong behavior.
- GMP convergence deadline. Date, design maturity standard, and the owner's right to terminate for convenience and keep the design.
- Owner-furnished equipment interface. Delivery dates, site receiving, storage responsibility, and who pays when late OFE delays the builder.
- Allowances and contingency. Cap allowances, define who controls contingency, and require shared reporting.
- Liquidated damages. Tie them to the owner's real daily cost of delay and apply them to energization milestones, not just substantial completion.
- Commissioning authority. Independent of the builder, with the power to fail a level.
- Key personnel. Named people for named roles with removal rights. Builders staff mega-projects thin when competing sites call.
Choosing between them
Run each question honestly.
- Does your team have the procurement and construction management depth to run a trade-contract program? If not, EPCM puts you in a seat you cannot fill.
- How mature is your reference design? Mature design favors EPCM or competitive GMP. Immature design favors progressive design-build, where the builder helps develop it.
- How much price certainty do you need, and when? Financing deadlines may force a GMP early.
- Is this a one-off or a program? Programs reward long-term relationships and IPD-style incentives. One-offs rarely sustain them.
- How much do you trust your counterparty's data? Open-book models fail when either side hides cost.
Most mature owners end up with a hybrid: owner-procured long-lead equipment, progressive design-build or CM at risk for the core and shell and site works, and IPD-style incentives with the two or three partners they plan to keep across the program. Whichever model you choose, write the governance down. For structure of the owner's program office and stage gates, see Hyperscale Program Governance and Capital Risk Mitigation.
Related briefs: Labor Capacity Risk and Craft Mobilization Guarantees | OEM Capacity Reservations and Supply Chain Capital Protection