Technical Details of the 800V HVDC Architecture

Infineon Technologies and Skeleton Technologies have signed a Memorandum of Understanding (MoU) to co-develop next-generation power delivery systems tailored for high-density AI data centers. The collaboration focuses on an 800V High Voltage Direct Current (HVDC) architecture, leveraging Silicon Carbide (SiC) solid-state transformers and Gallium Nitride (GaN) peak-shaving sidecars. By integrating Skeleton’s advanced ultracapacitor storage with Infineon’s wide-bandgap semiconductors, the system aims to seamlessly manage the extreme transient loads generated by modern AI accelerators.

Key technical advantages of this innovative grid-to-core architecture include:

  • Significant reduction in power conversion stages, drastically minimizing overall energy loss.
  • Dynamic peak-shaving capabilities to stabilize the local grid during sudden compute spikes.
  • Enhanced power density, allowing facility operators to pack more compute racks per square meter.

Industry Impact on AI Data Centers

The exponential growth of large language models and generative AI has pushed traditional AC power distribution to its absolute limits. Today's AI clusters require unprecedented power densities, frequently exceeding 100kW per rack. The proposed 800V HVDC solution directly addresses these critical bottlenecks by eliminating bulky legacy copper transformers and reducing the physical footprint of power delivery infrastructure. This paradigm shift enables hyperscalers to deploy significantly more GPU clusters within existing facility constraints while maintaining strict thermal and electrical stability.

Future Outlook for Semiconductor Power Delivery

As AI workloads continue to scale globally, the synergy between wide-bandgap materials and advanced energy storage will become a critical competitive differentiator. Infineon and Skeleton’s strategic partnership signals a broader industry shift toward localized, DC-native microgrids embedded directly within data center environments. Over the next decade, this hybrid technology is expected to become the foundational standard for gigawatt-scale AI facilities, driving down the total cost of ownership and significantly improving the sustainability metrics of global cloud infrastructure.