• • • AI • • •
Managing extremes with immersion cooling
At the upper end of the density spectrum, immersion cooling presents a different electrical–thermal relationship. By submerging hardware in dielectric fluid, immersion systems remove heat uniformly across components and eliminate internal server fans entirely. From an electrical engineering standpoint, this simplifies power delivery within the server while shifting complexity to facility-level systems. Heat is extracted efficiently, but requires careful integration with pumping, heat rejection and monitoring across critical digital infrastructure. Immersion cooling is, therefore, best suited to well-defined, high-density deployments such as AI training clusters. It is less a universal solution than a specialised tool within a broader thermal strategy.
Hybrid approaches for mixed
electrical topologies Most facilities will operate mixed-density topologies with hybrid deployments for the foreseeable future. This means that air and liquid-cooled racks could be positioned directly next to each other or in adjacent rows. Supporting this requires solutions that balance electrical constraints with thermal performance. Rear-door heat exchangers, for example, remove heat close to the IT load from exhaust air before it re-enters the room. This reduces thermal load on room-level systems without altering server electrical architecture. In-row cooling provides targeted capacity where electrical density is highest, stabilising local conditions without overhauling entire halls.
These approaches allow operators to incrementally increase electrical density while maintaining thermal stability and avoiding disruptive redesigns. Before heat can be transported or rejected, it must first be effectively captured and managed within the data hall, where room-level airflow design, containment strategies and targeted cooling systems determine how efficiently thermal energy is removed from IT loads. In this context, perimeter-based air-handling units and thermal wall solutions help define airflow paths at the room boundary, enabling controlled heat collection across both raised and non-raised floor environments while supporting higher-density hybrid cooling architectures.
Plant-side systems and electrical efficiency
Once heat is captured, it must be transported and reused or rejected. This stage of the thermal chain has significant electrical implications, particularly as AI workloads drive greater variability in cooling demand.
Chilled-water systems serving AI data centres may increasingly operate with higher return temperatures and fluctuating loads. Mechanical plant must, therefore, be capable of efficient part-load operation while responding dynamically to electrical demand changes. Alongside traditional chiller-based approaches, many data centres are now adopting so-called “trim cooling” strategies, which operate at elevated
electricalengineeringmagazine.co.uk ELECTRICAL ENGINEERING • FEBRUARY 2026 23
Control systems as the integration layer
The effectiveness of the thermal chain ultimately depends on control. Sensors, analytics and automation now act as the integration layer between electrical load, thermal response and mechanical plant behaviour.
water or chip temperatures to maximise free-cooling hours and minimise compressor use, with the optimal approach depending on where a facility sits in its thermal evolution. Centrifugal chiller technologies, available in
both air-cooled and water-cooled configurations, align well with these requirements. Their variable-speed operation allows electrical input to scale with cooling demand, supporting efficient operation under transient conditions rather than fixed design points.
In facilities where electrical infrastructure is constrained, alternative drive approaches can complement grid-powered systems by redistributing energy demand and preserving capacity for IT load growth.
Direct expansion (DX) systems continue to play a critical role in modular and edge set-ups, and smaller AI clusters. DX technology uses refrigerant-based heat transfer directly within the cooling unit, eliminating the need for a chilled-water loop. This makes DX systems simpler to deploy, faster to commission and ideal for sites where water availability or plant infrastructure is limited.
At the equipment level, controls regulate flow rates, fan speeds and local temperatures. At the facility level, supervisory systems coordinate setpoints across electrical and mechanical domains, anticipating load changes and optimising energy use. From an electrical engineering perspective, this integration reduces unnecessary load, smooths demand profiles and improves fault tolerance. It also provides the data needed to plan future capacity and evaluate the electrical impact of new workloads.
Reframing heat as a core engineering factor
As AI reshapes critical digital infrastructure, heat management must be treated as a first-order engineering parameter rather than a secondary consideration. Electrical power, thermal capacity and control systems are now tightly coupled and decisions in one domain directly affect outcomes in another. Viewing cooling through the lens of the thermal chain helps engineers understand these interactions. It shifts the focus from individual components to system behaviour, enabling more resilient, efficient and scalable infrastructure design.
In AI data centres, managing heat is no longer just about removing excess energy. It is about aligning electrical and thermal systems to support optimal performance under dynamic conditions. That alignment is fast becoming a defining challenge for electrical engineering in the age of AI.
https://www.vertiv.com/en-emea/
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48
