超级电容能否平滑AI数据中心供电的功率波动？

AI workloads can create load swings far sharper than steady-state compute, producing hard power spikes that conventional architectures struggle to absorb.

Supercapacitors, also called ultracapacitors, sit between conventional capacitors and batteries. They store energy as a surface charge rather than through a battery’s chemical reaction, which gives them higher power density than batteries but lower energy density. That trade-off suits short, frequent bursts rather than sustained runtime. Operators are testing supercapacitors alongside batteries and uninterruptible power supply (UPS) systems rather than replacing them. Batteries and UPS systems still support the sustained load and provide a longer backup window.

A September 2026 report from IDTechEx identifies three roles for supercapacitors in data centers: voltage stabilization, peak shaving, and short-term backup power. Supercapacitors charge and discharge in milliseconds to seconds – fast enough to track AI training cycles and recharge before the next spike.

Market Outlook: Small Today, Growing Fast

IDTechEx projects the data center supercapacitor market will reach $950 million by 2037, a 40.9% compound annual growth rate from 2026. That growth rate stands out against the wider category. A separate IDTechEx report from August 2025 puts total global supercapacitor revenue across automotive, grid, and industrial uses at $4.86 billion by 2036, growing 15.3% annually.

Three Designs Vying for the Rack Slot

IDTechEx groups data center supercapacitors into three categories, based on how they store charge and where they fit:

EDLCs (electric double-layer capacitors). The most commercially mature of the three, EDLCs store charge electrostatically at the electrode surface, much like a conventional capacitor.
Hybrid capacitors. These are usually lithium-ion (Li-ion) hybrids, combining an EDLC’s surface charge storage with the “rocking chair” lithium-ion chemistry in which ions shuttle back and forth between electrodes during charge and discharge.
Pseudocapacitors. These store charge through fast, reversible chemical reactions at the electrode surface rather than purely by electrostatic storage, delivering higher energy density than an EDLC but lower power density.

Backup power is the narrowest application across all three. Typical EDLC deployments sustain output for approximately 2 minutes, while some hybrids can last up to 10 minutes – enough to prevent data corruption during brief outages but not enough to replace a UPS system. IDTechEx frames supercapacitors as a complement to backup infrastructure, not a substitute.

From Pilots to Deployment: Pace and Roadblocks

The market for supercapacitors in data centers is still in its early stages, but growing.

“The market is transitioning from early evaluation to meaningful deployment, although adoption rates vary by technology type and supply chain maturity,” said Chris Butler, president of embedded and critical power at Flex, in an interview with Data Center Knowledge. “Instantaneous load transients can reach more than 150% of a power shelf’s rated capacity during certain AI compute events.”

Rack designers are building in flexibility rather than committing to one architecture. “Rack designs typically include dedicated slots that can accommodate BBUs (battery backup units), CBUs (capacitor backup units), or mixed configurations, providing deployment flexibility,” Butler said. “Supply chain concerns remain a major hurdle, particularly for newer hybrid supercapacitor technologies with limited qualified suppliers.”

Parag Nathaney, principal quantitative engineer for energy policy and power markets at Exelon, tied the pace of adoption to the reliability bar data centers operate under. He told Data Center Knowledge that the technology must meet 99.999% reliability for the power supply, which is why he sees the market as in its early stages.

“Supercapacitors are advancing toward greater deployment, but the current state is characterized by pilot and early commercialization,” Nathaney said.

Ian Wright, chief technology officer at DIMAAG, which builds battery systems designed to avoid supercapacitors, argued that the DC-to-DC conversion stage determines the response time, not the device alone. “The response speed to load changes is then determined by the DC/DC, which will be slower than either caps or batteries,” Wright told Data Center Knowledge. “If supercaps have been tried and not observed to get hot, that’s probably because they are not actually absorbing the cyclic loads.”

What’s Next and How Operators Decide

Whether IDTechEx’s forecast for supercapacitors plays out remains to be seen. Wright expects directly connected batteries, not supercapacitors, to become the primary tool for smoothing AI power swings.

Nathaney reached a different conclusion, tied to reliability rather than economics. “I think the application of supercapacitors for data centers is novel, but the long track record of the technology in other applications could lower the barrier for adoption by data centers,” he said.

Butler is somewhat more optimistic. As rack power levels scale from hundreds of kilowatts toward multi-megawatt designs, he expects CBUs to see growing use in managing transient peak demands. Over time, he anticipates workload-driven optimization will determine the mix of batteries and capacitors, with supercapacitors playing an increasingly important role in handling high-cycle, peak-power events.

“I expect capacitor-based energy storage to become a mainstream element of next-generation AI power architectures,” Butler said.

About the Author

Contributor

Sean Michael Kerner is an IT consultant, technology enthusiast and tinkerer. He consults to industry and media organizations on technology issues.

https://www.linkedin.com/in/seanmkerner/