The spikes in demand from data centres running artificial intelligence training workloads are a significant threat to world energy stability and should be reined in by governments, the chief executive of Hitachi Energy told the Financial Times.
Previous reports about AI data centres’ power use have focused on their spiralling energy demands, but Hitachi Energy’s Andreas Schierenbeck said they also put out huge and unpredictable surges in demand that make it difficult to maintain stable supplies.
He described the spikes and troughs along with uneven renewable energy supplies as “volatility on top of volatility” that is unprecedented in any other industry.
Power spikes
AI data centres differ from conventional facilities in that they “really spike up”, said Schierenbeck, the former head of German energy group Uniper.
When an AI firm begins a training procedure using large numbers of power-hungry AI accelerator chips, energy usage spikes in seconds to 10 times its normal levels, he said.
Comparing such surges to smelters in the steel industry, he said such uses usually require a company to “call the utility ahead”.
“No user from an industry point of view would be allowed to have this kind of behaviour,” he said, adding that governments should impose similar rules on data centres.
Hitachi Energy offers AI products of its own, such as Nostradamus AI, launched last November, that offers energy forecasts for utilities, power system operators, energy producers and traders.
Formed in 2020 from the $1 billion (£730m) takeover of ABB Power Grids, Hitachi Energy is the world’s largest producer of power transformers, an essential grid component that is currently experiencing a global shortage.
Transformer shortage
Schierenbeck predicted the shortage would take three years to ease and said it was partly due to a paucity of specialist contractors who could build the reinforced flooring needed to construct transformers weighing hundreds of tonnes.
The bottleneck is yet another side-effect of the massive spike in power demand from the rise in popularity of generative AI systems since late 2022, as large tech companies collectively spend hundreds of billions of dollars on AI data centres this year.
In order to secure their expected power needs for AI, tech companies such as Google and Facebook parent Meta Platforms have been investing in nuclear power and even fusion power, a technology that has not yet reached commercial viability.