AI Power Surges Put Global Grids at Risk, Warns Hitachi Energy

Hitachi Energy CEO warns that sudden power surges from AI data centres could destabilise global grids, urging new regulations, technological innovations, and infrastructure planning.

As artificial intelligence (AI) continues to gain traction across industries, the infrastructure required to support growing AI is posing a growing and urgent threat to global power grids. According to Andreas Schierenbeck, CEO of transformer manufacturer Hitachi Energy, AI-related power surges from data centres pose a serious risk to the stability of national and global electricity networks.

In a recent interview with the Financial Times, Schierenbeck warned that the extreme volatility in energy use generated by AI training workloads is unlike anything utilities have dealt with before, due to the unpredictability of demand surges.

“AI data centres are very, very different from these office data centres because they really spike up,” he said. “If you start your AI algorithm to learn and give them data to digest, they’re peaking in seconds and going up to 10 times what they have normally used.”

This pattern of usage—massive power spikes followed by steep drops—is putting unprecedented strain on grid infrastructure, especially when paired with the unpredictable nature of renewable energy sources like wind and solar. Schierenbeck described the situation as “volatility on top of volatility,” emphasising that no other industrial sector would be allowed to draw power with such erratic demand.

Rewriting the Rules of Energy Consumption

AI training relies heavily on energy-intensive processors, including GPUs and specialised AI accelerator chips, which consume vast amounts of electricity, particularly during the model training phase. As major tech companies continue to scale up AI infrastructure, these facilities are becoming among the most power-hungry on the planet.

A recent study by Bloomberg estimated that global power consumption could increase by a factor of 10 by 2030, driven largely by the continued adoption of AI. Countries like Ireland and the Netherlands have already placed limits on new data centre construction due to concerns about the strain it puts on national grids.

But Hitachi Energy’s CEO argues that the problem goes beyond overall consumption—it is the unpredictability of usage that’s especially dangerous. He called on governments to impose operational rules similar to those applied to traditional heavy industries like aluminium smelting, where companies must notify utilities before powering up large equipment.

“No user from an industry point of view would be allowed to have this kind of behaviour,” said Schierenbeck. “If you want to start a smelter, you have to call the utility ahead.”

Grid Planning and Innovation Lag Behind Demand

The rise in AI-related demand is hitting an already strained electrical infrastructure. Hitachi Energy, which was formed out of a £8 billion ($11 billion) acquisition of ABB Power Grids in 2020, is struggling to meet a massive surge in orders. Its order backlog has grown from £10 billion ($14 billion) to £31.6 billion ($43 billion) over the past three years.

A key bottleneck? Transformers. These essential grid components, which manage voltage and ensure stable power delivery, are in critically short supply. Manufacturing them requires specialised facilities and skilled contractors capable of building reinforced flooring to support machinery that weighs hundreds of tonnes. Schierenbeck estimates it will take at least three years for the industry to catch up.

To address the crisis, Hitachi Energy plans to invest £4.4 billion ($6 billion) to expand production and hire 15,000 new workers by 2027. 

AI as Part of the Solution?

Ironically, the same AI that is creating the problem may also be part of the solution. Hitachi Energy has developed Nostradamus AI, a forecasting tool designed to help utilities, energy producers, and traders anticipate demand and optimise grid performance. Announced in late 2024, the platform uses machine learning to improve the predictability of energy investments and consumption patterns.

Still, technology alone will not solve the issue. Energy analysts from Rystad Energy have suggested that AI demand could be a stabilising force if managed properly. For instance, data centres could be required to set maximum power thresholds and align training schedules with periods of surplus renewable energy.

Meanwhile, tech giants like Google and Meta are hedging their energy needs by investing in long-term sources, including nuclear and even experimental fusion power. However, those efforts are years—if not decades—away from achieving meaningful commercial scale.