For the first time, more corporate IT workloads run in third-party facilities than in businesses' own server rooms, according to Uptime Institute's Global Data Center Survey 2026. Alongside that shift, average rack power density has crested above 11kW, forcing UK buyers to rethink refresh timing, colocation contracts and the on-prem versus cloud calculus — understanding cloud vs on-prem TCO has never mattered more.
View the data behind this chart
| Layer | Detail |
|---|---|
| Third-party facilities | 46% of workloads — largest share for the first time |
| Enterprise-owned data centres | 44% of workloads, share holding steady |
| IT rooms and server cabinets | 10% of workloads, forecast to keep shrinking |
The tipping point: off-prem now the majority
Uptime Institute's annual poll of more than 800 data centre owners and operators, over half based in North America and Europe, found third-party sites now host 46 percent of enterprise IT workloads, edging past the 44 percent still running in company-owned server halls. The remaining 10 percent sits in IT rooms and server cabinets rather than dedicated facilities.
This isn't a sudden lurch. Uptime's own modelling projects the self-owned share holding steady at 44 percent through 2028, while third-party capacity grows to 48 percent — mostly by absorbing what used to sit in ad-hoc IT rooms and cabinets. For UK buyers still running comms rooms or small server cupboards, that's the segment most likely to be squeezed out first.
Why rack power density is now the real cost driver
The headline number — average rack density above 11kW for the first time — is real, but it's skewed. Strip out a small number of new ultra-high-density facilities running racks above 30kW, and the more representative average is 7.8kW, barely up from 7.5kW in 2025. That gap matters: a mixed UK estate is really two pricing problems, not one.
Uptime found 24 percent of respondents now operate at least some racks at 30kW or above, up from 19 percent a year ago, with the increase concentrated in the 50kW-plus range — including AI and GPU deployments configured for over 100kW per rack. Wider industry commentary suggests traditional air cooling starts to struggle around 20kW per rack, and that AI-ready racks are increasingly being designed for 40-100kW, with liquid-cooled deployments already reaching roughly 120kW and vendor roadmaps discussing 600kW racks by the end of 2027. Buyers should read rack power density explained before assuming a like-for-like refresh will fit existing power and cooling budgets.
What this means for server refresh planning
Uptime's report notes that some operators are pursuing more aggressive refresh cycles of under four years — a notable reversal from the 6-7 year lifecycles some hyperscalers have adopted to manage depreciation. For UK buyers, this split matters: newer hardware delivers better performance-per-watt, but each generation typically draws more power per rack, not less, so a faster refresh can quietly inflate your electricity and cooling bill even as compute density improves.
Before committing capital to a full-fleet swap, it's worth stress-testing whether a shorter, targeted refresh — replacing only the racks nearing end-of-support or capacity limits — delivers better returns than a blanket cycle. Buyers with ageing but serviceable estates should also weigh third-party maintenance services and explore refurbished servers as ways to extend usable life on lower-density workloads while reserving capital for the genuinely power-hungry tiers.
UK colocation cost impact: pricing is splitting by density class
Because most enterprise racks in the UK still sit well under 10kW, while a growing minority need 30kW-plus or 50kW-plus accommodation, colocation pricing is bifurcating. Commodity rack space with standard power distribution remains relatively cheap and widely available; specialised high-density suites with reinforced power feeds and liquid-cooling infrastructure command a premium and are in shorter supply.
Buyers evaluating colocation should ask providers explicitly what density tier they're quoting for, not just a blended £-per-kW figure, since a facility optimised for 7-8kW racks may not be able to accommodate a 30kW-plus deployment without a costly retrofit or a move to a different hall entirely. Understanding what is colocation covers today versus what AI-ready capacity actually requires is now a prerequisite for accurate budgeting.
On-prem vs cloud-hybrid: a decision framework for 2026
The off-prem majority doesn't mean every workload should move. Uptime's data suggests the migration is being driven disproportionately by the expensive, dense workloads that are hardest to justify keeping in-house — not a uniform exodus. If your organisation already has spare power and cooling headroom for moderate-density racks, keeping those workloads on-prem can still be the cheaper option.
Where AI or GPU-heavy workloads are pushing into the 30kW-plus range, the calculus flips quickly: the capital cost of upgrading in-house power distribution and cooling to hyperscale-adjacent standards rarely competes with colocation or cloud on a like-for-like basis. Buyers should calculate your cloud vs on-prem TCO per workload tier rather than at the estate level, and use calculate server room cooling needs to check whether existing plant room capacity can even support a planned density increase before signing off any refresh.
View the data behind this chart
| 2025 | 2026 | |
|---|---|---|
| Share of operators… | %19 | %24 |
Outages, staffing and the hidden risk premium
Two other findings from the survey should feed into any build-versus-buy decision. First, outage frequency has improved for a sixth consecutive year, but the financial impact of the outages that do happen keeps rising — 71 percent of respondents said their worst outage cost at least $100,000, up from 57 percent a year earlier. Second, skills shortages are worsening, with 53 percent of operators struggling to fill vacancies, up from 46 percent, and the sharpest gaps in electrical, junior operations and mechanical roles.
Together these point to a widening gap between what an under-resourced in-house team can reliably operate at higher densities and what a specialist colocation or cloud provider can staff and maintain. For UK buyers weighing capital outlay against operational risk, IT finance options can help spread the cost of a targeted, density-appropriate refresh rather than forcing an all-or-nothing capital decision this financial year.
- 01The Register — The majority of corporate IT is now off-premises for the first time · 30 July 2026
- 02Data Center Dynamics — Density Dilemmas · 30 July 2026
- 03Data Center Dynamics — The Path to Power · 30 July 2026
- 04Data Center Dynamics — The Future is Liquid · 30 July 2026
- 05Data Center Dynamics — Nvidia GTC: Jensen Huang on data center rack density · 30 July 2026