The rapid expansion of artificial intelligence is creating an unexpected challenge for the very data centres powering the AI boom: volatile electricity demand is putting excessive strain on critical infrastructure. Rapid fluctuations in power consumption are causing batteries, generators, turbines and cooling systems to malfunction or wear out significantly faster than expected, raising operational costs and creating new reliability risks.
AI data centres differ from traditional facilities because their workloads can cause electricity demand to rise and fall dramatically within seconds. During AI model training, hundreds of thousands of GPUs can switch workloads simultaneously, creating sudden power fluctuations that conventional data-centre equipment was not designed to absorb. Power consumption can at times spike as much as 50% above a facility’s designed capacity for brief periods.
The physical impact is already being reported across the industry. At some facilities, turbines and small natural-gas engines have developed cracks, while batteries installed to smooth power fluctuations have reportedly required replacement within months or even weeks. Such equipment failures can also create risks for the expensive AI chips and computing infrastructure housed inside the facilities.
The reliability challenges are also beginning to affect project timelines. A planned 2.67-gigawatt AI campus in West Texas, for instance, has pushed its expected power delivery from 2027 to 2028 partly because of additional engineering requirements. Industry experts warn that downtime can translate into significant financial losses, with the cost varying from thousands to hundreds of thousands of dollars per minute depending on the facility and workload.
Beyond individual data centres, the volatility is creating concerns for electricity grids. AI facilities can introduce highly dynamic loads that make it harder for utilities to maintain stable power flows. The North American Electric Reliability Corporation (NERC) has repeatedly warned that data centres are becoming a significant risk to grid stability, with its assessments finding that many existing data-centre load models do not adequately capture their dynamic behaviour.
The industry is now exploring technologies such as batteries, capacitors, transformers, flywheels, hydrogen fuel cells and advanced power-management systems to smooth fluctuations. Companies are also working with utilities and power specialists to design AI infrastructure that can better accommodate rapidly changing workloads.
The issue highlights a critical challenge for the next phase of AI infrastructure growth: building more computing capacity is not simply a question of adding GPUs and data centres, but also of designing power systems capable of handling AI’s unusually dynamic energy requirements. As billions of dollars continue to flow into AI infrastructure, managing power stability and equipment reliability will be increasingly important to ensuring that these investments deliver the expected returns.
