Corporate server rooms are being nudged aside as AI racks, power bills and staffing shortages turn in-house IT into punishment.
The latest Uptime Institute Global Data Centre Survey found third-party sites now account for 46 per cent of enterprise IT workloads. That puts them ahead of enterprise-owned corporate server farms, which still handle 44 per cent.
The survey gathered responses from more than 800 data-centre owners and operators across multiple countries. More than 52 per cent were based in North America and Europe. Another 10 per cent still rely on smaller IT rooms and server cabinets.
Uptime Institute expects self-owned server halls to hold their current workload share through 2028. Third-party facilities are expected to grow further, reaching 48 per cent of workloads and pinching capacity from informal IT rooms.
Average rack power density has crossed 11kW for the first time. The rise is being driven by higher-powered hardware, which is corporate speak for hotter, thirstier kit. Strip out a small group of ultra-high-density sites, and the typical rack density sits closer to 7.8kW.
That is still up from 7.5kW in 2025, so the industry has managed to make even the boring racks need more juice. Some operators are shortening hardware refresh cycles to less than four years.
That reverses the trend among hyperscalers, where Microsoft, Google and Meta have stretched server lifecycles to six or seven years to trim depreciation costs. The rest of the market seems to be stuck between replacing kit quickly and explaining to finance why the server room now needs its own substation.
Uptime found nearly a quarter of respondents now run at least some racks rated at 30kW or above. That compares with 19 per cent last year. Much of the increase came in the 50kW-plus range.
Some operators are squeezing AI and GPU servers into racks above 100kW. Outages have declined for a sixth consecutive year. The share of respondents reporting one fell by three percentage points.
Uptime warned that unstable grids, extreme weather and supply-chain constraints still threaten future reliability. Roughly 71 per cent of operators said their worst outage in the past year cost at least $100,000.
That was up sharply from 57 per cent previously. Rising costs for power, staff and AI-related equipment are now the industry’s biggest financial worry. Capacity forecasting, power availability and ongoing supply-chain disruption are adding more grief for operators planning large data-centre expansions.
More than half of operators (53 per cent) reported difficulty finding qualified candidates for vacant roles. That was up from 46 per cent a year earlier.
Staff shortages make the power and density mess worse, with electrical and junior operations roles each showing skills gaps affecting 38 per cent of surveyed organisations.
This explains why third-party facilities have overtaken corporate server farms for the first time. Enterprises can dodge some capital spending, power headaches and staffing pain by shifting high-density workloads to specialist operators.
The catch is that those specialist operators are stuck fighting the same grid limits, supply-chain delays and talent shortages, only with bigger buildings and posher dashboards.
