DPR Construction — which describes itself as “a self-performing general contractor focused on highly complex and technical projects” — has published a white paper on what artificial intelligence does to healthcare facilities, and its central finding is a sequencing claim: operations change first, buildings change later, and the gap between the two is where owners lose money.
The paper is Planning with Intelligence: How AI Is Reshaping the Healthcare Built Environment — the release spells out “Artificial Intelligence” in the title — posted on DPR’s Healthcare Insights hub — the page is dated 9 October, updated 12 October — and promoted in a press release on 22 October datelined Santa Clara, California.
The white paper page describes it as a qualitative study with 29 individuals, drawn from health systems, technology, finance, advisory services, healthcare planning, architecture and consulting. That is a well-chosen panel rather than a survey sample, and the paper presents itself as the former. The release separately reports that 91% of respondents said AI is breaking down traditional project-delivery silos — a figure that appears in the announcement rather than on the white paper page, which does not say how many people it is drawn from; the release describes the research as resting on “dozens of professionals.” The two documents do not tie the percentage to the 29, and we are not going to tie it for them.
The Mismatch Is Arithmetic, Not Attitude
The structural argument the paper makes does not depend on any forecast about AI at all.
Major healthcare capital projects, it says, “may take 5 to 10 years to plan and deliver,” while “the facilities themselves may remain in service for 30 to 50 years or longer.” Clinical software, staffing models and care pathways move on a cycle measured in quarters. So a programming decision made in 2026 — how many exam rooms, how much administrative floor area, where the imaging suite sits — gets locked in against an operating model that will be unrecognisable before the building opens, and then has to survive several more rewrites of that model across its service life.
This is the most useful thing in the document, and it is the part that requires no view on whether AI delivers what its vendors promise. The planning horizon and the service life are facts about construction. The volatility is a fact about software. Put them together and “plan for multiple futures” stops being consultant-speak and becomes the only defensible position.
Space Gets Redistributed, Not Deleted
The paper’s most commercially consequential finding is a refusal of the obvious conclusion.
If AI absorbs administrative and support work, the tempting inference is that hospitals need less space. DPR’s finding is that space is more likely to be redistributed than uniformly reduced: AI may reduce demand for some documentation, administrative and support space, while higher-acuity care, virtual operations, automation, equipment support and technical infrastructure may need capacity elsewhere. The recommendation that follows is specific — measure utilisation and throughput before cutting programme area.
That is a sentence written for a particular reader: the owner who has been told by a consultant that an AI-enabled operating model justifies a smaller building, and who is about to take square footage out of a project the same paper expects to be in service for thirty to fifty years. Floor area removed in design is the hardest thing to add back.
Two related recommendations follow the same logic:
- Judge existing assets on capability, not condition alone. The paper’s criteria include power, cooling, pathways, connectivity, data readiness, conversion potential and upgrade access, and it notes that a building’s future value may depend on more than its age, physical condition and current utilisation — a different test, on our reading, from the usual facility-condition index. A sound building that cannot feed and cool the equipment a service line now needs is the constraint, however good its roof is.
- Treat infrastructure headroom as a capital-planning question. Power, cooling, connectivity and data readiness may set the pace of adoption, so the capacity to add load is itself an option worth paying for.
Anyone who has read this newsroom’s data-centre coverage will recognise the second point: it is the same power-and-cooling constraint that runs through the grid-interconnection queue now gating data-centre construction in Texas, arriving inside healthcare capital planning. The difference is that a hospital cannot be sited for electricity.
Governance Is the Part Contractors Will Feel
The governance finding, as the announcement puts it, is that oversight has to become a lifecycle capability spanning clinical operations, facilities, IT, cybersecurity and capital planning, built around shared systems and data. The white paper’s own summary of the point reaches further down the project, naming design, construction, commissioning and operations as well. Either way it is a standing function rather than, on our reading, a policy owned by one department.
Hamilton Espinosa, DPR’s healthcare leader, framed the practical consequence as a scheduling problem: “AI is pushing technology decisions earlier in facility and capital planning.” Bringing clinical, technology, facilities and capital leaders together earlier, he said, helps owners decide what they need now and where to keep flexibility. His other quoted line is the paper’s thesis in one sentence: “Health systems can make smart capital decisions today without knowing exactly what AI will look like years from now.”
For design and construction teams the implication is a different conversation at a different time. Deciding in schematic design what a floor’s power density and pathway capacity must support — against equipment nobody has specified yet — is a requirements problem that currently has no good home in the project organisation. That is also, for a contractor whose business is complex technical delivery, a commercially convenient conclusion. Worth noting plainly: this is research published by a company that sells the services its conclusions call for, which does not make the sequencing argument wrong, but is context a reader should have.
Where It Sits Against the Other Evidence
Most AI-in-construction research measures adoption or productivity. This one measures a timing mismatch, and that puts it alongside a small set of findings this newsroom has covered which point the same way: the constraint is organisational, not technical.
McKinsey’s estimate that 39% of non-physical construction work is technically automatable against single-digit real adoption described the same gap from the capability side. ServiceTitan’s commercial-trades research on the ROI gap found it in the economics. Buildots’ global benchmarks found that planned weekly work goes undelivered at rates that have nothing to do with software availability. DPR’s version locates it in the calendar: the decisions are being made too early, by the wrong group of people, against assumptions with a shorter half-life than the concrete.
DPR has taken a position on construction’s data problems before: its venture arm WND Ventures invested in the hiring platform Skillit alongside Suffolk Technologies, and the contractor is a customer of it too.
What To Watch
Does any owner publish an actual flexibility premium? “Preserve optionality” has a price — oversized electrical rooms, spare pathways, convertible shell space. Until a health system discloses what it paid for headroom and what it got, this remains a well-argued principle rather than a decision rule.
Does the 91% survive a larger sample? The release does not give a base for it. A qualitative panel of people already working on AI implementation is the right place to generate a hypothesis about silos collapsing and the wrong place to settle it.
Do the specifications change? The test of this paper is whether healthcare programming documents start carrying capability requirements — kilowatts per floor, pathway counts, conversion allowances — rather than room counts alone. That shift would show up in bid documents long before it showed up in a building.
