IN Brief:
- Infineon and Skeleton will explore grid-to-core power architectures for higher-density AI data centres.
- Medium-voltage AC-to-HVDC solid-state transformers will combine CoolSiC devices with Skeleton power conversion and supercapacitor technology.
- A second development strand pairs CoolGaN devices with supercapacitors for GaN-based peak-shaving sidecars.
Infineon Technologies and Skeleton Technologies have signed a memorandum of understanding to develop higher-density power architectures for AI data centres, combining power semiconductors, power conversion systems, and high-power energy storage. The collaboration covers solid-state transformers for medium-voltage AC to high-voltage DC conversion and sidecar systems intended to support resilience and peak-power handling across the grid-to-core power chain.
The proposed solid-state transformer work will combine Infineon’s high-voltage CoolSiC power semiconductors with Skeleton’s power conversion and supercapacitor technology. A second strand will examine gallium nitride based peak-shaving systems pairing Skeleton’s supercapacitors with Infineon CoolGaN devices. The companies have not announced electrical ratings, converter topologies, development milestones, customer programmes, or a commercial availability date, so the agreement remains a development programme rather than a finished product launch.
Andreas Weisl, Executive Vice President and Chief Sales Officer of Industrial and Infrastructure at Infineon, said: “The increasing power density of modern data centers calls for new approaches to energy conversion, distribution and storage.” Accelerator racks are increasing the demand placed on conversion, buffering, protection, and thermal design, pushing those functions into a more closely connected electrical system rather than leaving them as independent facility and server components.
Infineon has been extending its AI power portfolio across several voltage levels, including high-voltage DC conversion, intermediate bus conversion, server power supplies, solid-state transformers, and protection devices. Silicon, silicon carbide, and gallium nitride devices can be applied at different points according to switching voltage, frequency, efficiency, thermal constraints, and power density.
The Skeleton collaboration pushes that design problem further upstream. A medium-voltage solid-state transformer can replace several conventional conversion stages with actively controlled semiconductor power stages, while a high-power storage sidecar can provide a local energy buffer without occupying rack volume intended for accelerator hardware. Taavi Madiberk, chief executive of Skeleton Technologies, said power density is becoming a defining constraint for AI infrastructure as operators try to supply increasingly dense GPU hardware without allowing the electrical system to consume more physical space.
Infineon has already paired SiC devices with another solid-state transformer developer. In March, the company began working with DG Matrix on its Interport platform, targeting medium-voltage conversion for AI data centres and industrial power systems. The Skeleton agreement adds high-power supercapacitor storage and a GaN peak-shaving strand, extending the design scope from conversion into short-duration power support and system resilience.
The commercial pressure behind that work is also visible in Infineon’s results. AI server and data centre demand helped lift the company to record quarterly revenue in the three months to 30 June 2026, with Power & Sensor Systems revenue rising 14% sequentially to €1.442 billion. Capacity agreements with AI customers are giving the company greater visibility over future semiconductor demand, increasing the incentive to connect discrete device development with reference architectures and system-level partnerships.
There is still substantial engineering work between an MoU and deployable infrastructure. Solid-state transformers operating from medium-voltage inputs have to demonstrate conversion efficiency, isolation, protection behaviour, thermal performance, maintainability, and fault handling at power levels where downtime is expensive. Peak-shaving systems add another control problem: the storage element, GaN conversion stage, and rack or facility power controls have to respond quickly enough to transient loads without creating instability or simply moving losses elsewhere in the system.
Device performance will therefore be only one part of the result. Hardware specifications, measured efficiency and power density, transient-response data, reliability testing, and evidence of integration with emerging high-voltage DC distribution schemes will determine whether the proposed SST and sidecar architectures are viable in production data centres. The MoU establishes the components and system capabilities the two companies intend to combine; the next useful milestone will be measured hardware rather than another increase in projected AI power demand.
