Most UK comms rooms were built around 3–8 kW racks. In 2026, average rack density in AFCOM's 10th anniversary State of the Data Center report has hit 27 kW per rack — up from 16 kW the year before and 6.1 kW in the earliest edition. If your last refresh is due, this piece is a readiness audit: what to measure, where UK regulation now draws the line, and how to decide between upgrading power and cooling in place or moving density elsewhere, using understand rack power density as your baseline reference.
View the data behind this chart
| kW per rack | Earliest edition | Prior year | 10th anniversary… |
|---|---|---|---|
| Average rack density | 6.1 | 16 | 27 |
Why rack density planning is now a board-level issue
For years, UK server room planning was mostly a space problem: did the comms room have enough U's of rack for the next three years of growth? That question is now secondary. Modern CPU refreshes, denser storage arrays and any GPU capacity draw far more power and reject far more heat per rack than the kit they replace, and many rooms were never designed with that load in mind.
The practical risk isn't running out of floor space — it's tripping a breaker, overwhelming a CRAC unit, or discovering the raised floor or slab was never rated for the weight of the new chassis. A rack density audit forces those three questions — power, cooling, structure — to the front of any 2026 hardware refresh, before procurement decisions are locked in.

Understanding rack density: what the current data actually shows
Two separate industry data points illustrate how fast density has moved, and it's important not to blend them. AFCOM's 10th anniversary State of the Data Center report, published in January 2026, puts the average rack density in its survey at 27 kW per rack, up from 16 kW in the prior year's edition and 6.1 kW in the earliest edition of the report. That is a single, consistent survey tracked over ten editions.
A separate figure, reported by Data Center Knowledge from the 2024 State of the Data Center report, cites an average rack density of 12 kW in 2024. This is a different report edition with its own methodology and sample, and should be read as a distinct 2024 data point rather than reconciled against the 27 kW 2026 figure — the direction of travel across both is the same, but the numbers are not directly comparable.
What both agree on is the trajectory: average density has climbed sharply within a few reporting cycles. For a UK buyer, the useful takeaway is simpler than the exact figure — if your comms room's design load sits anywhere near the legacy 3–8 kW range, a straightforward like-for-like hardware refresh in 2026 can realistically push individual racks well beyond what the room's power and cooling infrastructure was ever engineered to deliver.
Assessing your current infrastructure: the audit checklist
Before specifying new hardware, audit what the room can actually support today. This is the step most competing guides skip in favour of generic advice — it needs hard numbers from your own site.
To gather these figures without guesswork, start with the building's electrical schematics and as-built drawings — these normally show three-phase supply routing and the structural loading ratings for raised floors and slabs. Where drawings aren't readily to hand, your facilities manager or building surveyor will usually hold copies or know where they're archived. For spare breaker capacity at the distribution board, and for any assessment of whether the incoming supply itself needs augmenting before new load is added, it's worth looping in an electrical contractor early rather than estimating from nameplate ratings alone.
- •Three-phase supply capacity and spare breaker capacity at the distribution board serving the room
- •UPS runtime and headroom against the new expected IT load — use determine the right UPS for your power requirements as a starting sizing check
- •Existing cooling unit rated capacity versus the new anticipated heat rejection load, not the load the room was designed for originally
- •Raised floor or slab loading rating against the physical weight of denser, liquid-ready chassis
- •Containment (hot/cold aisle) status — open rooms lose efficiency fastest as density rises
- •Idle-state power draw of the current server fleet against EU ecodesign baseline allowances, to understand how much headroom newer, more efficient hardware could free up
A step-by-step methodology for rack density planning
Whether you're upgrading an existing comms room or specifying a new build, the same sequence applies, and it should always start with load, not layout.
First, baseline your current average kW per rack and total IT load, measured at the distribution board rather than estimated from nameplate ratings. Second, map the wattage of the planned hardware refresh — including any AI or GPU nodes, which typically draw disproportionately more per rack than general-purpose compute; estimate GPU requirements for AI workloads before finalising a spec. Third, if you're planning new or extended white space, apply the same net-to-gross ratios used at data centre scale: net data hall areas typically carry 2,500–4,000 W/m² of power and cooling load, and net hall area is generally only 40–50% of the gross internal area of the building, with external plant space typically accounting for 30% or more of gross internal area on top of that.
A worked example from RICS shows how these ratios scale: a 30 MW facility running at 3,500 W/m² needs roughly 8,570 m² of net white space, translating to 17,140–21,425 m² of gross internal area once the 40–50% ratio is applied. The same ratios apply proportionally at any scale — the maths doesn't change because your room is smaller, only the absolute numbers do.
Fourth, decide your cooling strategy against the density band you're planning for. Fifth, check whether your planned or consolidated IT load approaches UK regulatory thresholds discussed below. Finally, take your baseline and target figures to vendors and colocation providers and insist on quotes expressed in £ per kW, not just £ per rack unit, so that density-adjusted options are genuinely comparable.
Power and cooling: matching infrastructure to density
Air cooling has comfortably served the legacy 3–8 kW rack for two decades using standard perimeter CRAC units and basic containment. The density band now reported as average — 12 kW in the 2024 edition of the State of the Data Center report, and 27 kW in the 2026 10th anniversary edition — sits well beyond what conventional air cooling was designed around, which is why liquid cooling has moved from a niche AI-cluster technology to a mainstream planning consideration for any refresh landing in that range.
The right answer depends on your specific chassis and airflow design rather than a single rule of thumb, which is why it's worth working through air vs. liquid cooling for AI servers in detail before committing capital either way. In parallel, use the RICS power-density figures (2,500–4,000 W/m² of net hall space) as your planning metric for translating a target rack density into a physical space and heat-rejection requirement, and calculate your server room cooling needs against your actual planned load rather than the room's original design spec.
View the data behind this chart
| Net white space | GIA (lower estimate) | GIA (upper estimate) | |
|---|---|---|---|
| 30 MW at 3,500 W/m² | m²8570 | m²17140 | m²21425 |
The AI factor: planning for genuinely high-density workloads
GPU-dense hardware is the single biggest reason average rack density has climbed so quickly across both reporting cycles cited above. A single AI-focused rack can carry a fraction of the compute density of a general-purpose refresh while drawing several times the power, which is why blanket "one room, one density" planning breaks down once any AI capacity enters the mix.
If AI workloads are even a possibility in your 2026–2028 roadmap, model them separately from your general enterprise refresh. Use assess if you need a 130kW AI rack to right-size rather than over-provisioning, and treat power delivery — not GPU count — as the constraint that usually bites first.
Cost, ROI and the UK regulatory backdrop
No single reliable GBP figure for a "typical" rack density upgrade currently exists across UK suppliers — costs vary too much by site condition, existing electrical infrastructure and cooling choice to quote responsibly without a site-specific survey. The more useful discipline is comparing options on a consistent £ per kW basis: in-place power and cooling upgrades, liquid cooling retrofit, or migrating denser workloads to colocation, all priced against the same target density rather than against rack count alone. In practice, it's worth approaching several types of specialist in parallel for comparable quotes: colocation providers with established high-density or liquid-cooled halls, liquid cooling integrators who can retrofit CDU-based systems into existing racks, and data centre design or engineering consultants who can model the full power-cooling-structural envelope before capital is committed.
UK regulation is also shifting the calculus. Government planning documents and a 2026 legal briefing on the Cyber Security and Resilience Bill both describe colocation data centres with a rated IT load of more than 1 MW as potentially designated Operators of Essential Services, while single-organisation enterprise data centres fall into scope at 10 MW or more. Most individual comms rooms sit well under these thresholds today, but consolidating several dense racks onto one supply, or expanding into a dedicated hall, can bring a site closer to the 1 MW line faster than expected — worth checking before you commit to an on-premise expansion over colocation.
On the standards side, CIBSE notes that the UK Net Zero Carbon Buildings Standard exempts new-build data centres from its general energy-demand limits, but requires them to meet all space-heating demand through heat reuse within the facility — a material design consideration if you're building new high-density white space rather than retrofitting. On procurement, EU ecodesign rules set base idle-state power allowances by server class — 25 W for 1-socket servers, 38 W for 2-socket servers, and 40 W for blade or multi-node servers — a useful benchmark when comparing how much idle headroom newer hardware frees up against your existing fleet.
Future-proofing your UK server room or data centre
The room that copes with 2026 hardware is the one where power, cooling and structural capacity were sized to the target density before procurement started, not retrofitted after the new chassis arrived. Start every refresh with a baseline audit, size your cooling strategy to the density band you're actually planning for, and revisit the 1 MW and 10 MW regulatory thresholds whenever you consolidate load. Configure hardware against a confirmed density envelope, not a wish list, using server configuration tools to validate power draw before the order is placed.
Sources
Every figure in this article traces to the sources below.
- •AFCOM — average rack density trend across report editions
- •Data Center Knowledge — 2024 average rack density figure
- •UK Government — colocation and enterprise data centre scope thresholds
- •Gowling WLG — Cyber Security and Resilience Bill scope thresholds
- •EUR-Lex — EU ecodesign idle-state power allowances by server class
- •RICS — net data-hall power density and space ratios
- •CIBSE — UK Net Zero Carbon Buildings Standard and data centre heat reuse
View the data behind this chart
| Server class | Base idle allowa… | Rule reference | |
|---|---|---|---|
| 1-socket server | 1-socket server | 25 W | EU ecodesign 2020 |
| 2-socket server | 2-socket server | 38 W | EU ecodesign 2020 |
| Blade/multi-node… | Blade/multi-node server | 40 W | EU ecodesign 2020 |
