Last Updated on September 15, 2026 by Jeff Tomas
AI data centers are expanding fast, but their dense GPU systems are creating a new power problem: electricity demand can jump sharply within milliseconds. Uptime Intelligence reports that some AI configurations experience load swings of up to 150%, putting added pressure on power distribution, UPS systems, and backup equipment.
That pressure is driving interest in supercapacitors, which can respond quickly to short power spikes and dips. They don’t replace batteries, generators, or traditional UPS systems, but they can provide rack-level buffering and help stabilize power while longer-duration backup systems respond. For broader context, AI infrastructure energy demand is already shaping decisions about how data centers produce, store, and deliver electricity.
The $950 million figure refers to an IDTechEx forecast for the global data-center supercapacitor market to exceed $950 million by 2037, not the entire supercapacitor industry. The growth case becomes clearer when you look at GPU-rack power swings, fast UPS support, and the technologies being developed to manage them.
Key Takeaways
- AI GPU racks can create sharp power swings within milliseconds, adding stress to UPS systems and data-center distribution equipment.
- Supercapacitors provide fast, high-cycle buffering for short transients, while batteries and generators handle longer outages.
- Local DC-bus storage can reduce current swings, protect sensitive hardware, and limit disturbances before they reach upstream systems.
- IDTechEx forecasts the data-center supercapacitor market will exceed $950 million by 2037.
- Rising demand is also driving investment in gas-fired electricity for data center growth, renewables, nuclear power, and advanced backup systems.
What the $950 Million Supercapacitor Forecast Measures
The $950 million figure refers to a narrow data-center market, not the value of every supercapacitor sold worldwide. IDTechEx expects supercapacitors used in data-center power architectures to exceed $950 million by 2037. Its scope includes electrical double-layer capacitors, hybrid supercapacitors, UPS equipment, and power-buffering systems used in hyperscale and colocation facilities.
The growth story begins around 2026, when rising AI workloads start pushing more facilities toward fast-response power storage. IDTechEx projects a 40.9% compound annual growth rate from 2026 through 2037. Actual deployments may follow project schedules, with wider adoption expected as new AI campuses and high-density GPU clusters enter service.
That distinction matters because the forecast measures equipment and systems sold for data-center applications. It doesn’t describe current annual revenue, the amount already installed, or the value of the entire supercapacitor industry. Readers should treat $950 million as a future market projection, not a present-day sales figure.
Why AI Data Centers Need More Supercapacitors
Hyperscale AI data centers create the strongest demand because their GPU racks can change power consumption very quickly. Those sudden shifts place stress on UPS systems, converters, switchgear, and local distribution equipment. Supercapacitors can absorb and release energy in short bursts, helping smooth the load while batteries or generators handle longer interruptions.
Power buffering also fits the design of large GPU facilities. A supercapacitor bank can sit near a rack, DC bus, or UPS system, reducing the distance that fast-changing current must travel through the electrical network. As operators plan data center energy storage and backup systems, this local response can help protect sensitive computing hardware and limit disturbances upstream.
Why Market Estimates Differ
Broader supercapacitor forecasts often reach several billions of dollars because they cover many products and industries. For example, another IDTechEx estimate places the global supercapacitor market at $4.86 billion by 2036, including automotive systems, grid equipment, industrial machinery, and other applications.
Different reports produce different totals for several reasons:
- Some count only component sales, while others include complete modules, controls, and installation.
- One forecast may cover global demand, while another focuses on the US, Europe, or Asia-Pacific.
- Product definitions vary between EDLCs, hybrid devices, and integrated storage systems.
- Analysts also use different base years, adoption assumptions, revenue models, and forecast periods.
The cleanest reading is simple: $950 million is a projected data-center subsegment within a much larger global market. It shouldn’t be added to broader estimates as a separate industry.
Why AI Racks Need Faster Power Support
AI racks can change their electrical demand in milliseconds. When a large GPU cluster starts a training job, pauses between tasks, or shifts computing intensity, many processors may increase or reduce power at nearly the same time. That synchronized movement creates sharp load swings across rack power supplies, busbars, UPS equipment, and upstream distribution systems.
The problem involves power delivery, not only energy capacity. Energy capacity describes how much electricity a storage system can hold, usually measured in watt-hours. Power delivery describes how quickly it can provide that electricity, measured in watts or kilowatts. A battery may store substantial energy for a long outage, while a supercapacitor can deliver a rapid burst for a much shorter period.
How Supercapacitors Support AI Racks
Supercapacitors charge and discharge quickly because they store energy electrostatically rather than through the slower chemical reactions used by batteries. A rack-level module can absorb excess energy during a brief demand reduction, then release it when GPU loads rise. That action smooths short spikes before they travel through the wider data-center electrical system.
During a disturbance, the module can provide millisecond-to-second ride-through. This gives power converters and UPS controls time to respond, while batteries or generators prepare to carry the longer load. Supercapacitors therefore complement traditional backup equipment instead of replacing it.
They can also support peak shaving. By supplying part of a short demand surge locally, the system reduces the peak drawn from upstream equipment. That can limit voltage fluctuations, reduce stress on power supplies, and help operators size distribution hardware around a smoother load profile.
GB200 and GB300 Show the Scale
NVIDIA’s GB200 NVL72 combines 72 Blackwell GPUs in a rack-scale system with power demand reported at roughly 120 kW. NVIDIA describes the GB200 rack-scale platform as a liquid-cooled system built for tightly connected GPU workloads, where coordinated activity can create demanding power transitions.
The newer GB300 NVL72 raises the power challenge further. NVIDIA describes a GB300 power supply with energy storage that supports steadier rack power. Industry reports also claim that some GB300 configurations may require more than 300 supercapacitor units per rack. That figure should not be treated as a universal specification for every AI rack, but it shows how quickly fast-response storage can scale when many GPUs share one power system.
For operators, the goal is practical: place enough local buffering near the load to keep rapid GPU changes from becoming larger facility-wide disturbances.
Supercapacitors, Batteries, and UPS Systems Each Have a Different Job
Supercapacitors, batteries, and UPS systems often appear in the same data-center specifications, but they are not interchangeable. Each addresses a different part of the power problem, and the likely design for AI facilities is a hybrid system that combines rapid response with longer backup duration.
How the Technologies Compare
The clearest differences involve response speed, stored energy, cycle life, and operating cost. This comparison reflects the roles each technology typically fills in a data center.
| Technology | Response speed | Backup duration | Cycle life | Best use | Main limitation |
|---|---|---|---|---|---|
| Supercapacitor | Extremely fast, often within milliseconds | Seconds, sometimes up to about 90 seconds | Hundreds of thousands to more than one million cycles | GPU load swings, short ride-through, generator-start support | Low energy storage makes long outages expensive |
| Battery | Fast enough for UPS transfer and generator startup | Minutes or longer | Hundreds to several thousand cycles, depending on chemistry | Extended backup and reliable energy delivery | Heat, aging, maintenance, and frequent-cycle stress |
| UPS system | Detects disturbances and transfers or conditions power | Depends on its storage source | Depends on the storage technology | Complete backup, switching, power conditioning, and monitoring | It needs batteries, supercapacitors, or another energy source |
Supercapacitors have high power density and very low internal resistance, so they can absorb and release energy rapidly. They also tolerate repeated charge and discharge cycles far better than conventional batteries. That makes them useful when AI racks create frequent power changes rather than a single long outage. A technical review of energy storage for AI data centers also identifies supercapacitors and flywheels as strong options for fast response and high-cycle applications.
Their weakness is energy density. A supercapacitor bank can handle a short disturbance, but storing enough energy for hours would require a large and costly installation. Self-discharge can also reduce efficiency when the system must hold energy for long periods.
Why Batteries Still Matter
Batteries store much more energy in a smaller space, which makes them better suited to multi-minute backup. In a typical facility, the battery-backed UPS keeps servers running while generators start, synchronize, and accept the load. Lithium-ion batteries can offer longer service life than lead-acid designs, but all battery systems face heat management, capacity loss, and replacement planning.
The UPS is the wider architecture, not simply another type of battery. It includes power electronics, switching controls, monitoring, and the storage source. A modern AI data center may therefore use supercapacitors for rapid rack-level buffering, batteries for sustained ride-through, and generators for extended outages.
That arrangement avoids forcing one technology to handle every event. Supercapacitors manage fast, repeated disturbances, while batteries preserve energy for the longer interruptions that a capacitor bank cannot cover economically.
Who Is Building the AI Data Center Supercapacitor Supply Chain?
The AI data-center supercapacitor supply chain includes more than cell manufacturers. It also involves capacitor module suppliers, power-system companies, rack designers, controls providers, and service teams that must make the equipment work inside existing facilities.
NVIDIA creates demand through platforms such as the GB200 and GB300 NVL72. These systems are power-system customers and load drivers, not supercapacitor manufacturers. Their high-density GPU racks require fast energy buffering, which pushes suppliers to develop equipment that can handle rapid power pulses.
Component Suppliers and Power-System Integrators
LS Materials is one of the companies associated with supercapacitor development for AI infrastructure. Its work with Vertiv focuses on supercapacitor applications in UPS and data-center power systems, placing it closer to the component and module side of the chain. The exact commercial scope of each deployment still depends on the final rack design, qualification process, and customer contract.
Jianghai is also cited in market commentary as a supplier relevant to AI server power applications. Its role may involve capacitor components or assemblies that support server power conversion, although that doesn’t make the company the supplier of every supercapacitor installed in an AI rack.
Suppliers can sell products at several levels:
- Individual cells that system builders combine into larger banks.
- Packaged modules with monitoring, cooling, and balancing circuits.
- Rack-level power systems that connect directly to a DC bus or power shelf.
- Integrated UPS solutions that combine storage, switching, controls, and service support.
Vertiv operates further downstream by translating NVIDIA’s rack requirements into deployable infrastructure. Its PowerDirect 3000 provides a 33 kW, 50 VDC power shelf for GB200 and GB300 NVL72 systems, with hot-swappable power supplies and built-in capacitance for GPU pulse loads. Vertiv’s GB200 power and cooling reference architecture shows how a component need becomes part of a complete rack design.
What Buyers Must Verify
A promising capacitor module still needs to pass demanding data-center tests. Operators and equipment vendors must verify thermal performance, voltage balancing, fault behavior, control-system response, and compatibility with existing UPS units, busbars, rectifiers, and monitoring platforms.
Reliability testing should cover repeated pulse loads, temperature changes, brownouts, maintenance events, and failure of individual modules. Service contracts matter too, because operators need replacement procedures, remote diagnostics, spare parts, and predictable support over the system’s operating life.
Regional construction adds pressure to get these details right. Thailand’s expanding data center industry will increase demand for reliable power systems, especially where new facilities connect large AI loads to already-busy electrical networks.
Costs, Limits, and What the Boom Means for Data Centers
The business case for supercapacitors depends on the type of power problem a data center needs to solve. They cost more per watt-hour than batteries because they store far less energy, but they can deliver high power repeatedly with little performance loss. For AI facilities, that tradeoff can work when fast response matters more than long backup duration.
Where Supercapacitors Can Pay Off
A supercapacitor bank can absorb brief GPU load spikes, improve power quality, and protect sensitive equipment from short disturbances. It can also reduce the number of battery cycles in a UPS, which may slow battery aging and lower replacement costs. Repeated pulse loads are a strong fit because supercapacitors can charge and discharge far more often than conventional battery systems.
Local buffering may also reduce stress on rectifiers, converters, switchgear, and UPS inverters. By supporting part of a rack’s sudden demand, the system can help operators accommodate higher rack power without sizing every upstream component for the sharpest possible transient.
A hybrid design often offers the best economics. Batteries provide minutes of stored energy, while supercapacitors handle rapid power events. Research on hybrid energy storage for data centers supports this approach because each storage technology handles the duty it fits best.
Where the Limits Become Expensive
Low energy density remains the central weakness. A supercapacitor system may provide excellent power for seconds, but storing enough energy for a long outage requires a much larger installation. Self-discharge also makes it a poor choice for holding backup energy over extended periods. As a result, batteries remain more practical when runtime, rather than rapid cycling, drives the specification.
Operators must also budget for power electronics, monitoring, balancing controls, cooling, fire protection, and installation space. Supercapacitors are not simple battery replacements. Their controls must manage fast dispatch, recharge timing, voltage changes, and coordination with the UPS.
Supply-chain risk adds another concern. Specialized activated carbon and electrolyte materials can face price volatility, while vendor concentration may limit procurement options. The AI data-center power technology review highlights the gap between high power capability and relatively low energy density.
Forecasts deserve the same caution. Estimates for the broader supercapacitor market vary sharply, and the $950 million data-center projection depends on rapid AI deployment, high rack-power growth, and widespread adoption of hybrid systems. Early AI projects could be delayed, redesigned, or served by batteries, flywheels, or grid upgrades instead.
Regional rules will shape those choices too. In Thailand, clean energy sourcing requirements for data centers could affect electricity contracts, backup planning, and the value of local storage. The forecast is best read as a growth scenario, not a guaranteed outcome.
Frequently Asked Questions
These answers clarify what the $950 million forecast includes, where supercapacitors fit, and which companies are connected to the market.
Is the $950 million figure for the entire supercapacitor market?
No. The figure refers to the forecast market for supercapacitors used in data-center applications, including related modules, UPS equipment, and power-buffering systems. It does not represent the full global supercapacitor market, which also includes automotive, industrial, grid, and consumer applications.
Will supercapacitors replace batteries in data centers?
No. Supercapacitors handle very short, high-power events, such as GPU load spikes, voltage dips, and UPS transfer support. Batteries remain important because they store more energy and can provide backup for minutes or longer. A hybrid design gives each technology a suitable job.
How long can a data-center supercapacitor provide power?
The duration depends on the module’s size, voltage, stored energy, and connected load. In practice, data-center supercapacitors usually provide milliseconds to seconds of ride-through, rather than hours of backup. That short window can stabilize a rack or bridge a disturbance while batteries, UPS controls, or generators respond.
Why are AI servers harder to power than traditional servers?
AI servers use dense groups of GPUs that consume far more power than many conventional server configurations. When those GPUs start, pause, or change workload intensity together, the rack’s demand can shift quickly. These larger and faster electrical transients put extra pressure on power supplies, UPS equipment, converters, and distribution systems.
Which companies are associated with this market?
The companies have different roles. LS Materials and Jianghai are associated with capacitor and supercapacitor components, while Vertiv supplies UPS and critical-power infrastructure for data centers. Vertiv’s work on AI-ready power systems addresses fluctuating loads and rack-level power demands, as shown in its AI power infrastructure solutions. NVIDIA is an AI platform company whose high-density GPU systems create the need for faster power buffering, not a component supplier in this market.
What could slow the supercapacitor market’s growth?
High cost per watt-hour remains a major limitation because supercapacitors store less energy than batteries. Their limited long-duration capability, uncertain adoption rates, improving battery technologies, and competition from flywheels or grid upgrades could reduce demand. Delays in AI data-center construction would also push deployments and equipment orders further into the future.
AI data centers are creating a fast-growing need for power buffering as dense GPU racks produce sharp changes in electricity demand. That need may push the data-center supercapacitor market above $950 million by 2037, according to the IDTechEx forecast. The strongest takeaway is that supercapacitors address a specific problem: fast, short-duration power disturbances that can strain rack power systems and upstream equipment.
Supercapacitors are not replacements for batteries, UPS systems, or generators. They provide rapid buffering and short ride-through support, while batteries and other backup systems deliver power for longer interruptions. As data centers also respond to wider grid and storage needs, operators can review Thailand’s 14 GW energy storage target for broader planning context.
For operators and investors, the forecast should be a starting point rather than a guarantee. Each project needs evaluation based on rack power swings, ride-through requirements, backup duration, total system cost, and local grid conditions.
