Artificial intelligence (AI) data centers are widening the investment landscape for the power storage industry. While the expansion of data centers had made growing demand for energy storage systems (ESS) a key point for investors, attention is now expanding to "backup power sources" inside servers, such as supercapacitors (EDLC), that resolve rapid power fluctuations occurring in milliseconds during computation.
As big tech's race for AI leadership intensified, AI data centers expanded rapidly, centered on the United States. The global data center scale is estimated to have increased from about 36 gigawatts (GW) at the end of 2022 to 62 GW at the end of 2025. In particular, the share of investment into the United States surged about 96% over the same period, becoming the axis of growth.
Investment in U.S. data centers is leading to high growth in ESS. Because nuclear and large gas power plants take more than five years to build, ESS has emerged as an alternative to meet surging power demand. U.S. power generation capacity jumped from 13.5 GW in 2022 to 40.5 GW in 2024.
Recently, market interest is shifting from ESS to backup power sources inside servers such as EDLC. Unlike general data centers that use central processing units (CPU), AI data centers see power demand spike in milliseconds as graphics processing unit (GPU) clusters perform training and inference. If this power load is not resolved, power inefficiency occurs.
In response, big tech is focusing on building multilayered storage architectures that supply instantaneous power and stabilize voltage and frequency by using BBU (battery backup unit), lithium‑ion capacitors (LIC), and EDLC. BBUs supply power for several minutes in emergencies such as outages until emergency generators start up, while LIC and EDLC respond to sharp load fluctuations of around one second.
Jang Jeong-hun, a Samsung Securities researcher, said, "The biggest problem AI data centers face is 'load reflection,' where GPU chip power demand spikes sharply in milliseconds," adding, "BBUs and supercapacitors installed inside server racks will immediately supply momentary peak power and improve power operability."
Domestic supercapacitor companies include VINATech and LS Materials. According to Samsung Securities, VINATech is estimated to have signed a contract in May 2025 to exclusively supply supercapacitors for emergency power to Bloom Energy's fuel cell project.
LS Materials does not yet have AI-related revenue, but in Sep. 2024 it signed an MOU with Vertiv, a key partner of Nvidia and a global data center power solutions corporation, to supply supercapacitors (which the company calls ultracapacitors, or UC) for AI data center UPSs.
Cho Hyun-yeol, a Samsung Securities researcher, said, "Compared with general data centers, AI data centers are seeing sharper power fluctuations, so the trend is to add UCs in addition to BBUs in a multilayered storage architecture," adding, "In particular, when 800V DC data centers are introduced, the need for buffers between upper and lower power stages will grow, and UCs are increasingly likely to play that role."