Summary
- Dell’Oro forecasts data centre physical-infrastructure manufacturer revenue reaching $120bn by 2030, representing 22% annual growth from 2025.
- Nearly 200GW of new capacity is expected through 2030, with thermal management and liquid cooling leading equipment growth.
- EMEA is the only region revised down from Dell’Oro’s January outlook because power availability, permitting, and community opposition are constraining delivery.
Worldwide manufacturer revenue from data centre physical infrastructure is forecast to reach $120bn by 2030 as nearly 200GW of new capacity drives a much larger market for cooling, electrical distribution, UPS systems, racks, and supporting facility equipment.
Dell’Oro Group expects the market to grow at a compound annual rate of 22% between 2025 and 2030. The forecast is based primarily on additions to installed IT capacity rather than an assumption that operators will simply spend more on every megawatt they build.
That distinction is useful because higher-density AI infrastructure is changing where capital is allocated inside the facility. More power per rack does not automatically increase every equipment category at the same rate. It moves spending towards the parts of the electrical and thermal chain that have to manage concentrated loads.
Dell’Oro expects annual net capacity additions to peak in year-on-year growth terms during 2026 before moderating, although additions are forecast to remain in double digits through the end of the decade. North America is expected to contribute the most new capacity, followed by China.
Thermal management is forecast to be the fastest-growing physical-infrastructure segment. Liquid cooling is expected to lead at technology level as rack densities move beyond the range where conventional air cooling can economically carry the full load.
The research also expects water-cooled chillers to grow faster than air-cooled systems, attributing the difference primarily to scalability rather than a simple efficiency claim. Even facilities designed around warm-water liquid cooling can still require chillers because the amount of heat that can be rejected through free cooling falls during the hottest parts of the year.
That connects the equipment forecast directly to the operating conditions exposed by Europe’s current heatwave. High-density infrastructure creates large continuous thermal loads, while extreme ambient temperatures can reduce the usefulness of economisation just when the wider electricity system is also under greater pressure.
Electrical architecture is moving at the same time. Dell’Oro expects UPS growth to concentrate in larger three-phase systems as AI clusters require bigger power blocks. Medium-voltage designs are gaining ground, moving conversion and protection equipment closer to the grid connection and changing how operators divide the electrical chain between utility supply and the rack.
Solid-state transformers could begin to weigh on conventional UPS demand from 2029, according to the forecast, initially in the largest AI factories. That remains a forward-looking technology shift rather than an established replacement cycle, but it illustrates how quickly electrical design is being reconsidered when individual campuses are being planned at hundreds of megawatts.
The customer mix is changing too. Hyperscalers are expected to remain the largest single contributor to physical-infrastructure revenue through 2030, but Dell’Oro expects them to rely more heavily on colocation providers outside the United States.
That changes who buys the equipment without removing the underlying requirement. A powered-shell development can transfer procurement of racks, cooling, and some electrical systems from the developer to the tenant, while the physical building still has to provide enough upstream capacity to support the eventual load.
Dell’Oro has also separated specialist AI cloud providers and model builders into their own customer category. That segment is expected to become one of the fastest-growing parts of the market as neocloud operators secure increasingly large amounts of data centre capacity.
The European outlook is more constrained. EMEA is the only region Dell’Oro revised down from its January forecast, citing slower power availability and a more difficult permitting environment. The research also identifies community opposition as a material constraint capable of delaying or blocking announced projects.
That is consistent with the development pattern already visible across Europe. Operators continue to report strong demand, but new hyperscale projects are increasingly being pushed towards locations where grid capacity, land, and planning can be secured on a workable timetable rather than simply towards the largest established metropolitan markets.
Dell’Oro’s separate data centre capital-expenditure forecast, published a day earlier, now puts cumulative global spending above $3tn by 2030. That number includes large amounts of IT hardware and therefore should not be confused with the $120bn physical-infrastructure market. The two forecasts nevertheless describe the same transmission mechanism: more accelerators create higher electrical loads, which create larger cooling and power requirements at facility level.
The physical-infrastructure forecast gives the stronger measure of what that means outside the server chassis. UPS systems, chillers, liquid-cooling equipment, busway, rack power distribution, and associated plant all have manufacturing, installation, commissioning, and maintenance requirements that cannot be delivered at software speed.
Dell’Oro consequently identifies delivery rather than demand as the central risk. Equipment lead times, construction labour, grid interconnection, permitting, and local consent can all interrupt the route from an announced AI investment budget to an operating data hall.
That constraint is particularly relevant in Europe, where available power is already redistributing development between markets. The $120bn forecast describes a rapidly expanding equipment opportunity, but it also leaves a practical test hanging over the sector: whether enough substations, cooling plant, electrical systems, skilled labour, and consented sites can be delivered to absorb nearly 200GW of planned capacity.
