Demystifying high-density computing

High-density computing concentrates more processing capacity into fewer racks, increasing the demands placed on the infrastructure that powers, cools and supports them.

Deerns brings multidisciplinary engineering and international project experience into that discussion, working with clients as partners. We connect cooling decisions with power provision, physical space, environmental performance and operational needs, making the consequences of each choice clear. With realistic capacity assumptions and a practical route to adaptation, clients can invest confidently in infrastructure that supports today’s workloads and accommodates the next stage of growth.

From rack density to facility design

At Deerns, we often see projects start with a target rack density before establishing what the site and wider infrastructure can realistically support. A smaller server hall does not necessarily mean a smaller facility: it can still require significant external plant, substantial grid capacity and carefully coordinated services.

For data centre owners and investors, the starting point should therefore be the facility as a whole. Cooling technologies can enable higher rack densities, but their value depends on how well they integrate with the electrical infrastructure, site constraints and overall operating strategy. The question is not simply how much computing can fit into a rack, but whether the wider facility can support it reliably and efficiently.

Understanding the cooling options

Liquid cooling removes heat close to its source, supporting processor loads that become difficult to manage with air alone.

2 approaches deserve particular attention:

• Direct-to-chip cooling circulates coolant through cold plates attached to processors, transferring heat without exposing the electronics directly to the liquid.
• Immersion cooling submerges compatible IT equipment in an electrically non-conductive fluid, with different implications for equipment selection and maintenance.

The most appropriate solution depends on the equipment, application and location. In our high-density data centre work in Brazil, direct-to-chip cooling is currently the predominant approach, influenced by compatibility with manufacturers’ equipment and practical operating requirements. Immersion cooling remains an option in areas where the application and maintenance model make it suitable.

Liquid cooling does not necessarily eliminate the need for air cooling. Other server components and network racks may still rely on air, depending on the equipment mix. The design therefore needs to account for the facility’s actual combination of thermal loads. In many cases, this makes a hybrid cooling strategy the most practical way to support different loads within the same facility.

Planning the whole site

Concentrating more electrical load within a server hall increases the infrastructure required to support that space. Generators, substations, electrical distribution and heat rejection equipment all need room, alongside maintenance access. A compact building therefore does not guarantee a compact site.

Our experience in Brazil illustrates the importance of local constraints. Equipment availability can determine how many generators and associated electrical systems are needed. Power and fibre connectivity must also coincide at the chosen location; regional energy availability alone does not establish a viable project.

3 crucial factors must be established before committing to capacity:

• What electrical supply can be delivered, and when
• How supporting plant and maintenance access fit within the plot
• Whether local climate, water resources and connectivity support the proposed operation

Designing for a realistic rack mix

Peak rack density and average facility density are not the same measure. An AI cluster includes computing equipment, networking, connections and space for servicing. Applying the maximum rack rating to every rack position can therefore overstate the facility’s likely power demand and usable computing capacity.

Hybrid designs can accommodate a planned mix of liquid-cooled and air-cooled equipment, but the balance needs to reflect the intended workloads and deployment strategy. Flexibility should be designed around defined scenarios rather than unlimited future capacity, with power and cooling infrastructure sized to credible changes in demand.

Resilience also needs to reflect how the facility will operate. A colocation facility serving multiple tenants may have different redundancy requirements from an owner-operated facility with capacity available elsewhere. Designing for extreme ambient temperatures adds another consideration: greater cooling capacity can increase electrical demand and equipment size, with corresponding implications for cost.

Assessing environmental performance

Cooling choices connect electricity consumption, water use and carbon emissions. Reducing evaporative water consumption can increase electrical demand, while using water for cooling can reduce that demand under suitable conditions. The environmental outcome would then depend on local water availability, climate, and electricity supply.

3 complementary indicators help inform a combined assessment across seasonal conditions:

• Power Usage Effectiveness (PUE): total facility energy divided by IT equipment energy.
• Water Usage Effectiveness (WUE): annual site water use relative to IT equipment energy.
• Carbon Usage Effectiveness (CUE): operational carbon emissions relative to IT equipment energy.

Clear communication with surrounding communities should include both total resource consumption and peak demand, helping set realistic expectations about a facility’s local requirements. While efficiency metrics provide an important measure of performance, understanding absolute demand gives a fuller picture of how the facility will interact with local infrastructure and resources.

Turning ambition into a deliverable facility

The engineering task is to establish how much useful computing capacity a site can support reliably, responsibly and within the client’s investment programme. Early discussion of constraints creates a stronger basis for decisions before land, equipment and layouts become difficult to change.

Deerns brings the whole system into view, connecting power, cooling, site, engineering and operational requirements to turn high-density ambitions into resilient, efficient and deliverable data centres.