AI data centres turn to supercapacitors to manage power swings
People from left to right: Barry Powell, EVP and Managing Director, Electrical Products North America, Siemens Industry Inc.; Taavi Madiberk, CEO and co-founder of Skeleton Technologies; Markus Grabmeier, President of Electrical Products at Siemens AG (photo: Siemens)
An AI data centre’s processors can change their electricity demand within milliseconds. When thousands of chips move through similar stages of a workload together, their combined demand can swing by megawatts within seconds, putting pressure on the electrical systems supplying them.
Skeleton Technologies develops energy-storage systems designed to cushion those changes. On 14 October 2026, it announced a strategic investment led by Siemens, whose businesses span industrial automation, transport and electrical infrastructure. An accompanying global partnership gives Siemens exclusivity to sell and scale Skeleton’s products in the data-centre market.
For the two companies, the opportunity is to help operators extract more computing capacity from facilities whose electrical infrastructure can otherwise constrain their ambitions.
Why AI data-centre power demand fluctuates
Powering AI involves more than securing enough electricity over a day or year. Operators must also accommodate what happens from one instant to the next.
Graphics processing units, or GPUs, perform much of the computation behind modern AI. Their power requirements vary across calculation, communication and other phases of a workload. When these activities are synchronised across a large cluster, the resulting changes can create pronounced peaks and troughs.
Rapid fluctuations can disturb voltage and stress equipment between the processors and the grid. Infrastructure must be designed to handle the expected demand profile, including its peaks, even when average consumption is considerably lower.
Storage installed within the facility can help bridge that gap. It charges when demand falls and supplies energy when demand rises, smoothing the load seen by the upstream electrical system.
How supercapacitors smooth AI power demand
Supercapacitors are energy-storage devices that absorb and release energy quickly and tolerate frequent charging and discharging. Skeleton uses them to buffer the short power pulses created by AI workloads.
A lithium-ion battery stores energy through electrochemical processes. In an electric double-layer supercapacitor, charge is stored at the interfaces between electrodes and an electrolyte. This allows a fast response, although such devices generally hold much less energy for their size or weight than batteries.
The technologies can therefore complement each other: batteries suit longer periods of supply, while supercapacitors handle short, repeated bursts.
Skeleton’s particular proposition starts with the electrode material. Its proprietary Curved Graphene is a synthetic, porous carbon whose production process allows control over its properties. The pore structure helps make internal surface area accessible to the electrolyte’s ions, where charge is stored.
The company says its material provides high capacitance (the ability to store charge) and low internal resistance, limiting losses during power delivery. It combines the cells with power electronics and controls, turning a materials technology into systems that manage electrical demand.
Skeleton Technologies’ route from Estonia to AI infrastructure
Founded in Tartu, Estonia, in 2009, Skeleton developed its business long before the current surge in AI infrastructure spending. Its co-founders include chief executive Taavi Madiberk and chairman Oliver Ahlberg.
According to them, the European Space Agency was their first customer, with space-capacitor development beginning in 2011. Their first commercial product series followed in 2012, and they subsequently expanded into automotive, transport, industrial and grid applications.
Those markets provided opportunities for storage capable of delivering or absorbing power rapidly. AI data centres extend that application into another demanding environment.
Skeleton opened its €220 million factory in Markranstädt, near Leipzig, on 28 November 2025. At the opening, it said the facility was already delivering products for European electricity grids and major US cloud and data-centre operators.
Why Siemens is investing in Skeleton Technologies
Siemens has been Skeleton’s technology partner since 2022 and an investor since 2023. Its involvement extends to manufacturing: Skeleton says Siemens Digital Industries designed and automated its Leipzig factory.
The latest agreement links that industrial relationship more directly to Siemens’ customers. Siemens says nine of the ten leading global data-centre providers rely on its technology, giving it an established position in projects where electrical equipment must function as an integrated system.
Adding Skeleton’s products broadens its offering from the grid connection to the computing racks. Siemens says the systems can support higher computing density and reduce power disturbances in existing installations, potentially accelerating expansion without a complete facility redesign.
That could appeal to operators facing years of waiting for grid and facility upgrades. The potential depends on the installation, however: buffering brief peaks cannot overcome a sustained shortage of electricity.
Skeleton describes the financing as a nine-digit round without disclosing an exact amount. It says the investment will primarily support US manufacturing expansion and research and development in Silicon Valley. Separately, it plans to invest more than $200 million in operational expansion and R&D, with its US plans including domestic cell manufacturing.
GrapheneGPU: peak demand, energy savings and performance
Skeleton’s GrapheneGPU product family targets several benefits that need to be assessed separately.
Reducing peak demand can lower the grid-connection capacity required for a particular workload. Electricity savings can arise when storage replaces artificial workloads used to maintain steady demand. Additional useful computation may be possible where electrical constraints would otherwise require processors to be restricted.
Skeleton’s October announcement claims energy savings of up to 40% and double-digit GPU performance improvements. It does not provide sufficient measurement detail to establish how those figures translate across different installations.
Its earlier explanation of GrapheneGPU illustrates the importance of the baseline: avoiding unnecessary calculations consumes less electricity than running them to maintain a constant load. That comparison does not establish equivalent savings against a facility already managing demand efficiently.
A growing market for AI power management
The International Energy Agency (IEA) projects that global data-centre electricity consumption will roughly double from 485 terawatt-hours in 2025 to 950 terawatt-hours in 2030. That expansion creates a broader opportunity for companies supplying the electrical infrastructure behind computing, including systems that manage short-term fluctuations.
For Skeleton, the potential market includes new facilities and existing sites being adapted for more demanding AI workloads. Its planned US manufacturing expansion would bring production closer to major customers, while Siemens provides an established route into large infrastructure projects.
The pace of adoption will depend on costs, deployment schedules and the demands of each installation. Even so, the partnership positions Skeleton to compete for a growing share of spending beyond the processors themselves. As operators invest in more powerful computing, managing the electricity delivered to it becomes part of the investment case.
Liked this article? You can support our independent journalism via our page on Buy Me a Coffee. It helps keep MoveTheNeedle.news focused on depth, not clicks.