• • • DATA CENTRES • • •
AI IS REDRAWING THE ELECTRICAL BOUNDARY OF THE DATA CENTRE
BY MIKE JAANUS,
BUSINESS UNIT DIRECTOR, GLOBAL COMMISSIONING
whether grid connection queues or the rise of private substations and behind-the-metre generation. What matters more for engineers is what
M
happens to the discipline once generation and storage move inside the building’s distribution network, becoming part of one system, protected and coordinated as one. Engineers who work on conventional data
halls and AI clusters describe the change as behavioural. When thousands of GPUs synchronise across a training step, or a job pauses to checkpoint, demand can swing sharply within seconds, and the electrical system underneath must keep up. This sits on top of the AI electricity demand trend, projected to rise significantly over the next few years. Traditional data halls drew a flat, predictable
curve that let engineers size protection and switchgear against a known ceiling and move on. An AI cluster produces something closer to a heartbeat, and that difference changes how the electrical system needs to be designed.
Protection settings no longer
assume a quiet, external grid A protection scheme built around a stable utility connection assumes fault current comes from one direction, that the grid absorbs disturbances and facility equipment behaves as a predictable load. None of that holds once a facility owns generation
uch of the public conversation around AI and power distribution has focused on where electricity comes from,
and storage sitting electrically close to the compute load. Onsite generation contributes fault current of its own, battery systems introduce transient behaviour many protection studies were never built to model, and GPU power electronics introduce harmonics a purely mechanical study will miss. This is not a paperwork problem. Get the
coordination wrong and the facility risks disturbing the piece of network it sits inside, at a scale that draws attention beyond the campus.
Resilience is decided earlier,
and by different people Redundancy used to come down to how many independent routes a facility has back to the grid. That question still applies but is no longer sufficient on an AI campus with private generation and storage. Resilience depends on how those assets behave together under real conditions. Storage needs to absorb a load spike fast enough to protect upstream equipment, and the facility needs to hold its footing through a transition between grid supply and island mode without protection settings tripping unnecessarily. That means resilience decisions are being made
earlier, by engineers who understand generation and storage dynamics as well as IT load profiles. That is a different skill set to one most electrical teams were built around five years ago, with the industry still building.
HV systems are one design
problem now, not two One team used to design the incoming high-voltage interface and hand over a stable supply, whilst another designed the internal distribution and compute infrastructure against it. That worked because the HV side rarely changed once fixed. It doesn’t work when the HV interface is sized around an AI load profile that must account for peaks appearing.
24 ELECTRICAL ENGINEERING • SEPTEMBER 2026 Projects that do this well tend to be where
engineers are in the same room from the earliest design stages, rather than handing a finished spec across a boundary. This brings implications for how design teams are structured and how early HV expertise is brought in.
The underlying shift Together, these threads point to the same conclusion: a data centre that generates and stores its own power, running a compute load unlike a traditional cloud building, is now an energy system with a complex computing system sitting on top of it. It needs to be designed and protected as a single system, then verified as one. That is the independent assurance work Global Commissioning carries out, highlighting how early collaboration from conceptual design is becoming the norm. That will not sit comfortably with an industry
used to treating power as an input to be procured rather than a system to be engineered. It is, however, the direction AI load profiles are pushing the discipline, and today’s facilities will be judged against it long after this hardware is replaced. The industry has a narrow window to shape these design decisions now or spend the next five years catching up.
www.global-cxm.com
electricalengineeringmagazine.co.uk
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
