In July last year, grid operators in the US witnessed something they had never seen before: Problems on a transmission line caused a series of voltage dips, and then – in an instant – 1,500 megawatts of load from multiple sources just disappeared.
Load losses are not uncommon, but usually from a single source. This incident on July 10, 2024, according to experts at the North American Electric Reliability Corporation, came from a number of data centres that simultaneously responded to the voltage trips and automatically switched to back-up power.
The NERC experts say the grid has not historically experienced simultaneous load losses from multiple sources at this magnitude. In this case, the lights stayed on, but experts are worried because this is just a taste of things to come.
Most data centres built till now have loads of 30 MW or less. But some under construction have loads of more than 1 GW, and they are getting bigger. Facebook owner Meta has just announced a number of multi-gigawatt data centres in the US, including a 5 GW AI data centre that Meta boss Mark Zuckerberg boasts will be the size of Manhattan.
The problem, experts say, is not just their size but their variability. The NERC notes that AI training centres are particularly random, with loads ramping down from more than 400 MNW to 40 MW, and then back up again, and then back down again, in a matter of minutes.
“Emerging large loads, data centres, AI clusters, hydrogen, crypto, aren’t just big, they’re fast, invisible, and operating on their own timelines,” writes Dlzar Al Kez, an expert who serves on the editorial board of industry journals Smart Grids and Sustainable Energy.
“The real risk is loss of control,” he writes. This isn’t just about planning. It’s about control. And right now, control is slipping. The grid still assumes load is passive. It’s not. It’s power electronic, programmable, and often strategically opaque.”
Australia has been largely immune to these impacts, although it may not be long. The debate in Australia has focused more on how much demand there will be in the future, and how to plan for that, and not so much on how quickly it might disappear over a period of seconds.
Some renewable and storage developers are looking to encourage data centres in regions where the sun shines and the wind blows, to take advantage of resources that might not otherwise be exploited because of a lack of grid capacity.
But even if such installations go “behind the meter” by tapping into a big solar farm, wind project and localised battery storage, that is still a significant issue for the grid operator. As has been seen in the rooftop solar boom, it’s what it can’t see that it fears most.
The Australian Energy Market Operator is also looking at the data centre issue, and is expected to include new data and expectations in its upcoming annual Electricity Statement of Opportunities, as well as its updated multi-decade planning document, the Integrated System Plan.
Some network operators are also reporting significant interest in new load, but in high renewable states like South Australia this comes not just from data centres, but also other green industries that are looking to exploit new technology such as electrolysers.
The NERC paper also highlights this technology as a potential issue.
“Any commercial or industrial individual load facility or aggregation of load facilities at a single site behind one or more point(s) of interconnection that can pose reliability risks to the BPS (bulk power systems) BPS due to its demand, operational characteristics, or other factors,” the NERC says in its paper.
“Examples include, but are not limited to, data centers, cryptocurrency mining facilities, hydrogen electrolyzers, manufacturing facilities, and arc furnaces.”
The NERC says hydrogen electrolyzer facilities under development could reach the multi-gigawatt scale. Initial surveys have shown that over 85% of the facility load consists of power electronic converters with the remaining load being motor driven for H2 compression, water treatment and cooling, and other plant equipment.
Al Kez says planning for the overall growth in demand will mean little for grid operators if there is little or no visibility in what’s going on.
“Planning adequacy means nothing if a 300 MW electrolyser ramps to zero in 2 seconds because its own logic deems the voltage ‘unstable’,” he writes.
“Frequency control is irrelevant if the load that tripped wasn’t visible to begin with. Restoration is compromised if blackstart islands can’t segment large loads in time.
“This is not a future scenario. It’s happening now. Quietly. Repeatedly. Systemically.”
Christiaan Zuur, from the Clean Energy Council, says data centres are a “topic of jour” in the NEM. “This may be on the way in Australia, with major implications for how we should be planning the grid as well as developing new operational protocols,” he says.
Imagine, though, if we were trying to manage this in a system that relied only on ageing, fossil fuel baseload generators.
