• • • DATA CENTRES • • •
5-10 kW of just a few years ago. Integrated rack solutions now include high-density power distribution units (PDUs) or DC busbars (more suitable for higher density deployments), in-rack coolant distribution units (CDUs) and manifolds for direct liquid cooling (DLC) applications. Power distribution, from grid to chip, is undergoing a significant transformation driven by unprecedented behaviour of GPUs. Uninterruptible Power Supply (UPS) systems have been redesigned to handle ‘spikey’ and dynamic power loads of the AI workloads, which can shift from near-idle to a 150 per cent overload in milliseconds. The number of busbars above the racks is also adapting to new reference designs to provide more distributed redundancy. Cooling has also evolved. The traditional air- cooling methods are often insufficient for racks that exceed 50kW, necessitating a shift towards liquid cooling technologies. A hybrid approach that combines traditional air cooling with targeted liquid solutions, such as CDUs and rear-door heat exchangers (RDHX), can efficiently help manage these thermal densities resulting thermically neutral in existing data centres, and making them retrofittable with little CAPEX investment. As the white space fit-out evolves to integrate power and cooling systems directly into or near the racks, data centre operators can either build a smaller facility for the same capacity or, more commonly, increase compute density within the existing white space. New reference designs prioritise modularity and scalability, enabling a
insight, it becomes easier to right-size the supporting systems.
In some cases, retrofitting will be the preferred path. In others, a clean-slate build will make more sense and can be prefabricated. There is no universal template. What matters is that the design reflects what the system is being asked to do, both today and in the near future.
more flexible and efficient deployment that can adapt to the unpredictable and rapidly growing demands of AI and HPC workloads.
Design choices start with clarity Not every workload needs the same setup. The type of AI being deployed should influence how infrastructure is configured. Running an LLM for training, fine-tuning and running inference for generative AI applications requires a different infrastructure setup compared to modelling and simulation where researchers create complex simulations for engineering, finance and climate science or to digital twin renderings which require the creation and operation of photorealistic, real-time virtual replicas of physical objects and environments.
The best outcomes are achieved when infrastructure teams work closely with business stakeholders to define the use case in detail. That includes understanding the data involved, the speed required and the risks to manage. With that
Collaboration makes a difference
Edge computing deployments can be complex as they could involve multiplying sites into many locations. They require coordination across power, mechanical, IT and digital teams. Timelines can stretch, and handoffs between disciplines can lead to delays or oversights. Bringing the right partners into the process early can help align design decisions, reduce costs and keep deployment on track.
Many of the most successful projects treat infrastructure as a shared responsibility. When teams plan together, the result is a system that is easier to maintain and more responsive to change. Ultimately, AI isn’t a standalone application. It is part of a wider shift in enterprise data centre architecture. And this shift begins below the surface, in the systems that keep everything running. The more prepared those systems are, the more value AI will be able to create.
https://www.vertiv.com/en-emea
electricalengineeringmagazine.co.uk
ELECTRICAL ENGINEERING • NOVEMBER 2025 29
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 |
Page 49 |
Page 50 |
Page 51 |
Page 52
