Intelligent Data Centres Issue 66 | Page 70

HOW DATA CENTRES CAN SATISFY AI ’ S GROWING POWER- HUNGRINESS SUSTAINABLY

Danel Turk , Data Centres Portfolio Manager at ABB , explores how data centre managers can address the rising power demands of AI while scaling operations with sustainability as a main priority . ith the rapid proliferation of AI services ,

W the International Energy Agency ’ s ( IEA ) Electricity 2024 report suggests that global data centre electricity use could potentially double by 2026 , to a staggering 1000 terawatt-hours – equivalent to the annual electric consumption of Japan .

For the data centre industry , expanding operations in the face of this demand is a necessary but financially daunting undertaking .
Before we race to scale up , we must recognise that many data centres are already leaving more than half of their available energy on the table . Over 65 % of a data centre ’ s energy is either consumed by cooling or grid distribution losses .
If we tackle energy efficiency challenges like these first , we could potentially meet most of that doubled power demand .
A data centre ’ s electrical network used to be much simpler : you connected your site to a utility feed and ran it . Today , there are many more elements at play . You can run on renewable energy . You can install batteries as a buffer for intermittency . You have more opportunities for on-site generation , and AI itself is unlocking new ways to optimise power consumption , performance and maintenance .
Modern data centre electrification relies on three main pillars : smart architecture and modularity ; digital equipment ; and sustainable growth . Through these , we can achieve the energy efficiency needed to scale up competitively in the age of power-hungry AI , while remaining environmentally responsible .
Modular construction
One aspect of smart architecture is modularity . Modularity in data centre construction supports incremental power and availability demands while streamlining both specification and installation processes . Constructing a data centre in segmented blocks ( e . g ., 20MW sections for a projected 200MW facility ) aligns growth with actual demand , enhances cost efficiency ( by minimising unused capacity ) and reduces downtime risks .
Prefabricated solutions like skids and eHouses , constructed and tested off-site , expedite installation and cost reductions , resonating well with local utilities and regulatory approvals due to their less invasive growth impact . Moving blocks of electrical equipment outside of the main data hall also allows data centres to cut down on building size – less concrete used means less CO 2 emitted .
Medium Voltage and UPS
Medium Voltage ( MV ) Uninterruptible Power Supplies ( UPSs ) can be modularly installed , where parallel wiring of multiple units ( e . g ., 10 2.5MW blocks to form a 25MW system ) boosts capacity without additional complexities . This setup not only speeds up deployment but also reduces the operational burden of having a larger number of low-voltage UPSs . Furthermore , MV systems , typically more energy-efficient than low-voltage alternatives , provide comprehensive power protection for entire data centres , promising long-term savings .
If you calculate the total cost of ownership , factor in the UPS ’ s lifetime . Even small percentages of efficiency gained in the UPS
70 www . intelligentdatacentres . com