Artificial intelligence data centers could become a significant new market for supercapacitors as operators look for faster ways to absorb short-duration power spikes and stabilize voltage, according to market-research firm IDTechEx.
The firm forecasts that the market for supercapacitors in data centers could exceed $950 million by 2037. The estimate comes from its report, Supercapacitors for Data Centers 2027-2037.
The issue is increasingly relevant as operators install denser AI infrastructure. TNGlobal recently reported on Aolani’s planned expansion of NVIDIA-based AI infrastructure in Southeast Asia, one of several projects adding large accelerated-computing loads in the region.
AI workloads create faster power swings
AI training and inference can produce rapid changes in rack-level power demand. IDTechEx said some spikes occur on millisecond timescales, creating voltage sags and other stresses that conventional backup-power architecture was not designed to handle by itself.
Supercapacitors differ from batteries because they can charge and discharge very quickly and tolerate a high number of cycles, although they generally store far less energy for their size. That makes them more suited to short bursts of power than to long-duration backup during a grid outage.
In a data center, the technology could therefore complement rather than replace batteries and generators. A supercapacitor can respond to the immediate transient load, while longer-duration backup systems handle sustained loss of power.
Placement changes the engineering trade-off
IDTechEx identified three main deployment levels: inside or next to server racks, at the aisle or power-distribution level, and in a centralized uninterruptible power supply (UPS) area.
Rack-level placement can react close to the equipment generating the transient load and may be attractive for new AI facilities designed around high-density racks. Aisle-level systems can support groups of racks, while centralized installations may be easier to integrate with existing facility power architecture but can be less granular.
The best location will depend on rack density, electrical topology, retrofit constraints, space and the duration of the power event an operator is trying to manage. New-build facilities can design buffering into the rack architecture from the start, while existing data centers may favor more centralized approaches if replacing rack-level hardware is too disruptive.
Economics will decide adoption
Supercapacitors have been used for years in transport, industrial power and other applications that require high power over short intervals. The data-center opportunity depends on whether the cost of installing them can be justified by improved equipment protection, reduced oversizing of other power systems or better management of grid-facing demand peaks.
IDTechEx’s forecast does not establish that supercapacitors will become standard equipment across AI data centers. Operators will still compare them with lithium-ion battery buffering, flywheels, power-electronics changes and other approaches. The emerging question is where the response speed and cycle life of supercapacitors provide enough value to justify an additional layer in the power chain.
Featured image: Taylor Vick on Unsplash
Aolani and NVIDIA plan 22,000-GPU AI infrastructure expansion in Southeast Asia