Supercapacitors could buffer millisecond-scale AI power spikes, stabilise voltage, and complement batteries across racks, aisles, and UPS systems in future data-centre power architectures at scale.
IDTechEx has highlighted supercapacitors as a potential solution for managing rapid power fluctuations in AI data centres through its report, Supercapacitors for Data Centers 2027–2037: Technologies, Applications, Players and Forecasts. The report examines how these devices could support voltage stabilisation, peak shaving and short-duration backup requirements as AI workloads increase power demand.
AI training and inference can create large, short-duration changes in electricity demand, producing power spikes on millisecond-to-second timescales. Such fluctuations can contribute to voltage sags and place additional stress on conventional data-centre power infrastructure. Supercapacitors can respond rapidly by charging and discharging energy over short intervals, making them suitable for smoothing transient loads.
The technology could be deployed at several points within a facility. Rack-level systems can provide local voltage support for servers and allow operators to target only racks handling intensive AI workloads. This approach can be particularly suitable for newly built facilities where power architecture can be designed around high-density computing requirements.
Aisle-level installations, including power-distribution equipment, can support multiple racks and may offer easier integration during data-centre retrofits. Centralised UPS-room deployment can provide broader power support while requiring less equipment within individual racks. However, it may not provide the same local protection against voltage disturbances.
The choice of deployment will therefore depend on rack density, electrical architecture, retrofit requirements, available space and the duration of power events being addressed. IDTechEx expects aisle- and UPS-level deployments to gain adoption in the short term, particularly in existing facilities, while rack-level systems could become important in new AI data centres.
