Data Center Construction Market (2026-2036)

This market covers the design, engineering, construction, installation and commissioning through which a data centre facility is delivered, together with the building systems installed in it, and it excludes the information technology equipment the facility is built to house. The scope convention follows the treatment applied to comparable facility construction markets, in which the production equipment the building exists to accommodate is excluded while its installation is included. Readers should note that a total project value basis including servers, storage and network equipment would produce a considerably larger figure.

The growth of this market is driven by committed capacity programmes for artificial intelligence workloads, by the retrofit of existing halls for liquid cooling and higher rack densities, and by the expansion of the construction scope itself. That last point is the one most often missed. Connection conditions in constrained jurisdictions are beginning to require developers to deliver generation and storage as part of the project rather than to rely on the utility, which moves work that previously sat outside the construction boundary inside it. The market is therefore growing both because more facilities are being built and because each facility now involves more scope. The second effect is the less visible of the two and the one that changes which contractors can bid credibly.

Data centre construction is the delivery of a building whose purpose is to supply power and remove heat reliably. Everything about the project follows from those two functions, which is why the electrical and mechanical scopes dominate cost and schedule in a way they do not in commercial construction, and why the civil and architectural work, substantial though it is, is the less difficult part. The industry that performs this work consists of design and engineering firms, general contractors with sector-specific experience, specialist electrical and mechanical subcontractors, and commissioning agents, and the scarce capability in it is the ability to deliver to an energisation date that has already been committed to a customer.

The binding constraint on this market is not capital, construction labour or equipment supply, serious though the last two are. It is the availability of power at the point where the facility is to be built, and the evidence on that constraint is unambiguous. Analysis published by Lawrence Berkeley National Laboratory records 2,061 GW of generation and storage actively seeking transmission interconnection, a fall of about 10 percent from 2024 to 2025 against a peak near 2,600 GW at the end of 2023, with more than 750 GW of requests withdrawn during 2025. Natural gas capacity in the queues rose 86 percent to 253 GW, while solar fell 19 percent to 773 GW, storage fell 16 percent to 749 GW and wind fell 19 percent to 220 GW. The non-ISO West holds the most active capacity at 567 GW, followed by ERCOT at 408 GW, and only ERCOT and SPP recorded increases in 2025. Of the capacity from requests submitted between 2000 and 2020, 75 percent had been withdrawn and only 13 percent had come online by the end of 2025, and withdrawal rates still exceed 40 percent even after an interconnection agreement has been signed. In 2025 the median time from request to a signed agreement was well over three years, and the time from request to commercial operation exceeds five years in the regions for which data is available. Some 536 GW of active capacity already holds a draft or executed interconnection agreement, with CAISO and ERCOT each executing more than 30 GW during 2025.

Those figures should be read against the build cycle of the facilities seeking the power. A data centre shell can be delivered in eighteen to thirty months, while the generation intended to supply it takes more than five years from interconnection request to commercial operation. Generation therefore cannot follow demand on anything resembling the same timetable, and the mismatch is structural rather than a transient queue backlog. The composition of the queues reflects how the market is responding, with natural gas capacity rising 86 percent in a single year to 253 GW while solar, storage and wind all contracted, which is the signature of buyers seeking dispatchable capacity they can rely on at a known date rather than the lowest-cost energy.

The data also carries a warning about pipeline arithmetic that applies directly to this market. Of requests submitted between 2000 and 2020, three quarters were withdrawn and only 13 percent came online, and withdrawal still exceeds 40 percent even after an interconnection agreement has been signed. A queue position, and by extension an announced project, is therefore a weak predictor of delivery. Anyone sizing data centre construction from announced capacity pipelines rather than from contracted and financed projects will overstate it, and the same discipline that applies to generation queues should be applied to data centre announcements.

Where power cannot be supplied on the required timetable, regulators have begun to transfer the obligation to the developer, and the clearest example sets a template others can copy. The Commission for Regulation of Utilities in Ireland published its decision on a new electricity connection policy for data centres on 12 December 2025. Under the policy, new data centres must provide generation or storage capacity, onsite or nearby, matching their requested maximum import demand, and that generation must participate in the wholesale electricity market, which can also provide the developer with a revenue stream. At least 80 percent of annual demand must be met by additional renewable electricity projects generating in the Republic of Ireland, and renewables already contracted under support schemes such as REFIT, RESS or ORESS cannot count towards that requirement. Developers are given a six-year glide path for those renewable projects to be developed and begin generating, and must present a credible plan to the System Operators. Connection requests are assessed as being in a constrained or unconstrained location based on the specific application and its grid location rather than on a regional basis, the System Operators must regularly publish information on available grid capacity and current and future constraints, and EirGrid and ESB Networks were required to publish an engagement and connection process for applicants by 31 March 2026.

The commercial significance of that decision for this market is larger than its energy policy significance. A connection condition of this kind folds a power project into a data centre project. The developer must now deliver generation or storage matched to its maximum import demand, participate in the wholesale market with it, and contract additional renewable capacity that did not previously exist, all within the same programme as the building. Scope that sat with the utility moves to the developer and therefore into the construction market, and the skills required change accordingly, since delivering a generation asset and a grid connection is not the same discipline as delivering a hall. For contractors with energy infrastructure capability this is an expansion of addressable scope rather than a constraint.

The second force reshaping the work is thermal. The transition from air cooling to direct liquid cooling changes the mechanical scope fundamentally, introducing pressurised coolant circuits, coolant distribution equipment, secondary loops and far larger heat rejection plant, and it does so in retrofit as well as new build. Retrofitting a standing hall is the harder case, because the work proceeds in occupied or partly energised space with short possession windows, which constrains method and favours prefabricated and mechanically jointed solutions over site fabrication. This is why mechanical and cooling installation is the fastest-growing service scope even though electrical remains the largest.

Resilience expectations are moving in a direction that surprises people outside the sector. Facilities built for artificial intelligence training are frequently specified to lower facility redundancy than conventional enterprise or colocation halls, because a training run interrupted by a power event is restarted from a checkpoint rather than lost, and because redundancy is increasingly provided at the compute and scheduling layer rather than in the building. A lower tier specification reduces electrical and mechanical scope per megawatt, and the growth of this category therefore moderates construction value per unit of capacity even as capacity grows quickly. It also shortens programmes, which is often the actual objective.

Prefabrication is advancing for schedule reasons rather than cost reasons. Factory-built power and cooling modules, skidded plant and preassembled electrical rooms remove work from a congested site, compress the critical path and reduce dependence on local trade availability, at a higher cost per unit of capacity that buyers accept because the binding constraint is time. The approach also transfers quality control into a controlled environment, which matters where commissioning failures would delay energisation. Traditional stick-built construction nonetheless remains the larger share, because it accommodates site-specific conditions that a standard module cannot.

A caution belongs in any assessment of this market. Construction demand is a derivative of capacity commitment, and capacity commitment reverses faster than it builds. A construction business sized against a committed pipeline carries fixed overhead, trained labour and subcontractor relationships that cannot be shed on the timescale over which a pipeline can be deferred, and the sector has limited experience of a downturn at current scale. The exposure is greatest for contractors concentrated on a small number of hyperscale customers, and least for those with retrofit and multi-sector capability.