Infineon Technologies and Skeleton Technologies entered a non-binding memorandum of understanding in early September to develop power architectures for 800 VDC AI data centers. This collaboration aims to address the dual requirements of high-voltage power distribution and multi-timescale energy storage necessitated by modern AI rack designs.
The agreement outlines two primary technical tracks focused on the power chain from grid connection to processor. The first track centers on solid-state transformers designed to convert medium-voltage AC to high-voltage DC, utilizing Infineon’s CoolSiC power semiconductors alongside Skeleton’s power conversion systems. The second track targets gallium nitride-based peak-shaving sidecars, pairing Skeleton’s supercapacitor technology with Infineon’s CoolGaN devices to manage rapid load fluctuations.
Nvidia’s technical documentation emphasizes that megawatt-class AI racks require specialized storage to handle power swings that can fluctuate from 30% to 100% of load within milliseconds. The current industry standard relies on facility-level batteries for long-duration shifts, but these systems lack the response speed required for transient spikes. Supercapacitors provide the necessary high-power density to absorb these fast spikes and fill brief valleys in energy delivery.
Infineon brings significant experience in 800 VDC systems to the partnership, having worked with Nvidia on power architectures since May 2025. The company has already established similar collaborations, including a modular 2 to 5 MW solid-state transformer project with SolarEdge and the supply of silicon carbide devices for Eaton’s IEC-certified MVSST 2.0 platform. These existing relationships provide a foundation for the integration of high-voltage conversion stages.
Skeleton Technologies contributes manufacturing capacity and established storage solutions, supported by its 220 million euro Leipzig plant that produces up to 12 million supercapacitor cells annually. The company has previously demonstrated its storage capabilities through a collaboration with DG Matrix, which integrates Skeleton’s GrapheneCBU800 and GrapheneBBU800 systems into medium-voltage skids. This track record of validating storage systems under hyperscaler power profiles aligns with the performance targets of the new Infineon partnership.
The integration of CoolSiC and CoolGaN semiconductors represents a strategic shift in how power is managed at the rack level. Silicon carbide devices handle the high-voltage conversion stages with high efficiency, while gallium nitride devices enable the fast-switching required for peak shaving. By combining these semiconductors with supercapacitors, the companies aim to minimize energy loss during the conversion process while providing a buffer for the intense, rapid power demands of AI training workloads.
The shift toward 800 VDC distribution is driven by the need for greater efficiency and reduced material costs in massive AI clusters. Nvidia projects that this architecture can yield up to 5% better end-to-end efficiency and 45% less copper usage compared to traditional 54 V or 415 VAC distribution methods. These gains are critical for scaling compute density within existing physical footprints, as the physical size of power infrastructure often competes with the space required for GPU racks.
Market analysts from SemiAnalysis project that incremental AI data center capacity utilizing 800 VDC could reach 39 GW by 2030, fueling a solid-state transformer market valued at approximately 13 billion dollars. The adoption of these systems is expected to accelerate alongside the rollout of Nvidia’s Kyber rack systems in 2027. This timeline provides a clear target for the partners as they transition from architectural exploration to the development of prototypes and customer-specific testing.
The non-binding nature of the memorandum allows both companies to maintain their existing industry partnerships while exploring new integration opportunities. As the power requirements for AI training continue to evolve, the ability to combine high-voltage conversion with rapid-response storage will remain a primary focus for infrastructure providers. Future development will likely prioritize the refinement of these architectures to meet the specific thermal and power density constraints of next-generation AI hardware.
The collaboration highlights a broader trend where semiconductor manufacturers and energy storage specialists are moving closer to the rack level to solve power delivery bottlenecks. By aligning their respective technologies, Infineon and Skeleton are positioning themselves to provide a comprehensive solution for the power chain. This approach addresses the critical need for stability in environments where power demand is increasingly volatile and high-density compute is the standard.