Supercapacitors: The AI Power Revolution Catalyst, The Next Epic Price Surge Sector?

Driven by the rapid scaling of GB300 and subsequent Rubin platforms, the global supply-demand gap for supercapacitors used in AI servers has become undeniable. Musashi’s planned annual production capacity of 6.5 million units falls far short of the rigid demand for 15-18 million units in 2026. Price increases have become an inevitable outcome of the supply-demand mismatch. NVIDIA, starting with GB300, has integrated electrolytic capacitors into the power rack, which can reduce peak grid demand by about 30%. The next-generation Rubin platform will substantially increase energy storage capacity compared to previous generations, marking the upgrade of storage components from an ancillary function to a core system component for AI compute clusters. This architecture-level transformation initiated by leading compute platforms is reshaping the entire value distribution pattern of the power supply chain from the source.

From an industry space perspective, the global supercapacitor market is entering a high-speed expansion window expected to last several years. In 2025, the global market size is USD 2.80 billion, projected to reach USD 3.29 billion in 2026, and further grow to USD 9.51 billion by 2032, with a CAGR of 19.4% from 2026 to 2032. China accounts for 42.7% of the market, making it the world's main market, and domestic manufacturers are facing a historic substitution opportunity.

1. What’s happening? Supercapacitor boom imminent

The core points of this report are as follows—

① Demand leap: AI chips (such as NVIDIA's Rubin series) generate 2-3 times the instantaneous power peak during inference and training switching, which has become the "number one killer" to data center power distribution systems. Supercapacitors, with millisecond-level response, are rapidly replacing traditional lead-acid/lithium backup systems and becoming a necessity in computing infrastructure.

② Supply reshaping: With domestic breakthroughs in high-performance coconut shell activated carbon and carbon nanotube coating technologies, the BOM cost of supercapacitors has entered a 5-year downward trajectory. The global shipment volume is expected to surpass 85% YoY growth in 2026.

③ Value revaluation: Focused on "explosive growth from 0 to 1 in orders"; and a deep examination of the "asset attribute within new power systems," the industry value chain is evolving from traditional manufacturing to "technological infrastructure."

The global supercapacitor market enters a high growth channel

The global supercapacitor market is at a historic turning point. GB300 NVL72 is currently the benchmark for AI hardware power density. This platform uses a fully liquid-cooled rack-level architecture, integrates 72 Blackwell Ultra GPUs and 36 Grace CPUs, and single cabinet power reaches 130-140kW. Compared to traditional CPU cabinets at about 10kW, GB300 boosts single cabinet power by more than tenfold.

Starting with GB300, NVIDIA officially integrated capacitive energy storage components into the power rack. According to the official statement, this scheme can cut grid peak demand by up to 30%. In Vera Rubin NVL72 (expected mass production in the second half of 2026), energy storage capacity increases significantly compared to the previous generation (20x). The Rubin Ultra version with the Kyber rack architecture pushes single cabinet power to about 600kW, and the supporting 800V DC power system is being built in cooperation with more than 20 partners.

The generational evolution releases two key signals: First, from GB300’s 140kW to Rubin Ultra’s 600kW, more than 4x growth in two years, the demand for rigid power buffering per cabinet is systematically amplified. Second, for the first time, NVIDIA officially gives a clear multiplier: energy storage capacity is "20 times that of the previous generation," which means supercapacitors are evolving from "optional" to "mandatory," and from "secondary" to "core system components"—this industry logic is now irreversible.

China has become the most important single market for supercapacitors globally. Business Insights data shows China’s market will reach US$1.2 billion in 2025. Combined with localization trends in the AI data center supply chain, domestic manufacturers' dual advantages in scenario adaptation and cost control will further consolidate China’s core status in the global supercapacitor industry.

Why are supercapacitors irreplaceable?

Compared to traditional backup schemes, supercapacitors have four core advantages in AI scenarios: response speed—microsecond level (lead-acid/lithium batteries are millisecond level); cycle life—can reach 100,000-1,000,000 cycles, far above batteries' hundreds to thousands; wide operating temperature (-40°C to 85°C), suited for high-heat high-density cabinets; safety—the physical energy storage mechanism fundamentally eliminates thermal runaway risk.

Supercapacitors do not replace lithium batteries or diesel generators, but rather form a "three-layer complementary power pyramid": supercapacitors serve as "microsecond to second instant buffers," lithium batteries cover "second to minute-short support," and diesel generators provide "minute-to-hour long-term protection." In AI cabinets, all three are indispensable, together forming a multi-level backup power system meeting three types of power demand: "fast but short—slow but long—long but slow."

AI chips present typical high-frequency spike load characteristics during training and inference: synchronous computation stages bring sharp power ramp-up, followed by rapid drop-off. The chemical reaction response speed of traditional lead-acid batteries is at second level and cannot smooth out AI chip 2-3x instantaneous power peaks; mainstream lithium battery systems have switch delay over 200ms and cannot meet millisecond-level emergency response for grid outages. Supercapacitors, with millisecond-level response time, can provide instant power compensation during drastic load changes, smooth load surges, and ensure stable system operation.

2. Why is it important? Epic price hikes incoming

Understanding the market scale for supercapacitors requires two perspectives: total market size and AI server-specific market size.

With the dramatic increase in GB300 NVL72 single cabinet power, AI servers’ requirements for transient supply stability leap two levels, and supercapacitors and BBUs have officially upgraded from “optional components” in the GB200 stage to standard parts in GB300 era, unified into the Energy Storage Tray system.

Currently, the mainstream supply chain solution is the CESS (Capacitive Energy Storage System) developed by Musashi and Flex. Each GB300 cabinet needs 5 BBU modules and over 300 supercapacitors. Assuming GB300 NVL72 shipments reach 50-60 thousand racks in 2026, GB300 is expected to require 15-18 million supercapacitors in that year. GB300 PSU mass production ramp-up is ongoing, and leading power suppliers like Megmeet have received batch orders for multi-power PSUs plus supercapacitors, with delivery scheduled through June 2026. Rubin cabinets' supercapacitor quantity is expected to further rise to about 250 units per cabinet, with global demand possibly breaking 30 million units in 2028.

The core supply bottleneck is currently the only global mass supplier of hybrid supercapacitors for AI data centers—Japan’s Musashi. Musashi plans to expand annual production capacity from just 200,000 in 2024 to 6.5 million by Q3-2026. Even with full production, 6.5 million a year faces a huge gap of 9-10 million versus demand for 15-18 million units in 2026.

Supply chain information shows Jianghai Co., Ltd. has capacity for about 2.5 million supercapacitors, enough for about 7,000 GB300 cabinets, and has secured a major order for 5 million units from Flex, needing 2-3 years to fulfill. As of 2026 Q3, Musashi plans for 6.5 million per year using third-generation LIC (20kW per board), needing upgrade to fourth-gen LIC (40kW per board) for Rubin platform. Capacity ramp itself is time-bound. New lines, process validation, equipment delivery pace means even if 6.5 million/year is reached, in 2026 it covers only about one-third of target demand.

With such significant supply gaps and downstream customers rushing to secure capacity, price increases for supercapacitors are an inevitable industrial logic.

In terms of price, current server supercapacitors average about $2 per unit, with industry gross margin about 34%; AI server-grade products, due to higher technical specs, average about $2.6 per unit, with gross margin about 37%. Should prices rise, the profit uplift for related companies is two-fold: product unit income increases, and gross margin expansion further boosts net profit elasticity.

NVIDIA has already collaborated with Renesas, Delta, Flex, Eaton, GE Vernova, Siemens, and more than 20 partners to build the next-generation 800V DC power ecosystem. Once established, supercapacitors will move from "cabinet component" to "data center-level infrastructure component," with batch procurement far surpassing current forecasts. If platform-level locking effect occurs, supercapacitors will no longer be just consumables for single cabinets, but standardized long-term configurations for AI infrastructure. Matrix, its supercapacitor unit impact resistance reaches 100G, far above industry standards.

Why is this timing important? Because we’re at a critical moment for supply-demand matching.

Currently, the supercapacitor market is in a state of "tight balance but structural shortage." As Musashi’s 6.5 million/year capacity is released in Q3 2026, Skeleton’s 12 million/year runs at full, and domestic suppliers expand, 2026-2027 will see a global supercapacitor capacity concentration window.

3. What next? Supercapacitor industry chain panorama

The supercapacitor industry chain can be divided into three core segments: upstream (materials supply), midstream (cell manufacturing & module integration), and downstream (application scenarios). Materials cost accounts for more than 60% of total costs; the performance and price of materials directly determine final product competitiveness.

① Upstream materials: cost core and domestic substitution ongoing

Upstream materials include four major categories: electrode materials, electrolytes, separators, and current collectors, among which:

1) Electrode materials: cost share about 35-40%, determining energy and power densities. High specific surface area activated carbon is mainstream; new graphene composites can improve energy transmission efficiency by more than 40%. Domestically, high specific surface area activated carbon is highly localized, but some high-end raw materials (such as high-performance viscose-based activated carbon fiber) still rely on imports, bringing some "choke-point" risks.

2) Electrolyte: about 25-30% of cost; domestic matching is relatively mature. Leading Chinese suppliers like Capchem now occupy important share in the domestic supercapacitor electrolyte market.

3) Separator: about 15-20% of cost. Ultra-thin ceramic coated separators can reduce thermal runaway probability to one fifth of traditional.

With advances in domestic electrode, electrolyte, and separator materials, overall supercapacitor costs are dropping gradually.

② Midstream manufacturing: cell → module → system technology depth

Midstream’s core competitiveness is dual: "unit performance + module integration ability." From a tech route, supercapacitors are categorized as EDLC (electrical double layer capacitor), pseudocapacitors, and hybrid supercapacitors (HSC/LIC).

Hybrid supercapacitors (HSC/LIC) are now the mainstream solution in AI server applications. HSC adopts a hybrid structure, with activated carbon as the positive electrode and graphite, plus prelithiated technology, as the negative, offering both high power and energy density. NVIDIA’s supply chain uses Musashi & Flex’s CESS solution, employs lithium-ion supercapacitor (LIC), integrating 28 LICs as one 1U module, supplying 15-21kW, with a 6-year lifespan.

Server supercapacitors are moving from "board-level power hold devices" to "cabinet-level power buffer units." In AI cabinets, supercapacitors are increasingly packaged as modules or rack units with monitoring, and are managed energy buffers integrated within the power rack alongside charge/discharge and power management systems.

③ Downstream applications: five major scenarios driving demand expansion

Main downstream supercapacitor scenarios include:

From a competition view, global leaders race for capacity while domestic substitution speeds ahead. The supercapacitor market shows an "oligopoly + emerging players catching up" competition pattern. According to QYResearch, main global supercapacitor players include Maxwell Technologies (USA), Skeleton Technologies (Estonia/Germany), Musashi Energy (Japan), Nippon Chemi-Con (Japan), LS Mtron (Korea), Jianghai Co., Ltd, (China).

China is now the world’s biggest supercapacitor consumer market, though high-end products are still mainly imported.

We divide the supercapacitor industry evolution into three phases:

Phase one (2025-2026): AI data center-driven explosion. NVIDIA 800VDC power architecture officially published, GB200/GB300 clusters scale up, supercapacitors become standard for MW-level compute power. Musashi, Skeleton, Jianghai Co., Ltd. and others concentrate their capacity releases. This is the current core phase.

Phase two (2027-2030): Multi-scenario penetration. As energy density increases and costs drop, supercapacitors migrate beyond AI data centers into commercial-industrial storage, EV fast-charge assist, and other domains.

Phase three (post-2030): Infrastructure for energy. Supercapacitors become core infrastructure for grid frequency regulation and voltage support, with disruptive technologies such as solid-state electrolytes further expanding industry scale.

Finally, a summary of the key conclusions.

Conclusion one: The system-level conflict of MW-scale compute power supply is the root driver of the supercapacitor boom. AI clusters’ demand for power buffer components is non-compromisable, and supercapacitors are the most mature technology and clearest industrialization path solution available.

Conclusion two: 2026-2027 is the global supercapacitor capacity concentration window; first-moving companies gain the biggest marginal benefit. From the capacity expansion rhythm, Skeleton (starting Q4 2025), Musashi (expanding to 6.5m units Q3 2026), Jianghai Co., Ltd. (continuous expansion) are the first to benefit.

Conclusion three: Domestic substitution is timely. Chinese companies have three core advantages: (1) Integrated layout across the industrial chain, with the main aluminum electrolytic capacitor business providing stable cash flow and customer base; (2) Mass production capabilities for both EDLC and LIC, among the few suppliers in China of both technologies; (3) Deeply bound with core clients like Delta and Flex, with clear order guidance.

Conclusion four: China is the world’s biggest supercapacitor market, but the domestic proportion of high-end products is still low; enormous room for substitution. Policy (New Energy Storage Action Plan) and industry (AI supply chain independence) are dual catalysts.

Risk Statement and DisclaimerThe market has risks, investment requires caution. This article does not constitute individual investment advice, nor does it take into account the special investment goals, financial conditions, or needs of individual users. Users should consider whether any opinions, viewpoints, or conclusions in this article are appropriate for their particular circumstances. Investing based on this, responsibility is at your own risk.