STMicroelectronics has expanded the power supply structure for AI data centers from a 50V-centric model to include 12V and 6V. This reflects the assessment that as high-density AI servers increase, different intermediate voltage paths are required depending on rack configuration, GPU placement, and cooling methods, even within the same 800VDC base.

ST announced on March 24 that it has unveiled 800VDC-12V and 800VDC-6V solutions developed based on the NVIDIA 800VDC reference design. This announcement is a follow-up expansion to the existing 800VDC-50V solution and was introduced at NVIDIA GTC 2026.

The background of this expansion lies in the increasing power density of AI infrastructure. In large-scale training clusters, inference servers, and high-density computing environments, designs are being made to accommodate the coexistence of 50V, 12V, and 6V buses, depending on the GPU generation and server form factor. This aligns with the trend of ensuring scalability while reducing power loss.

The 12V architecture focuses on reducing power conversion steps by eliminating the existing 54V intermediate stage. The company explains that this reduces system losses and copper usage, and simplifies the power supply structure for next-generation GPUs. The new power supply board was also designed to match this direct conversion structure.

The 6V architecture places the power stage closer to the GPU. This reduces copper usage and resistance losses, and improves responsiveness to rapid load changes. It has been proposed as a method to increase power delivery efficiency, particularly in servers with ultra-high-density GPU configurations.

ST previously showcased a prototype of a power supply system equipped with a GaN-based LLC converter that operates directly at 800V in October 2025. With the addition of 12V and 6V, ST has further expanded its 800VDC power conversion portfolio for AI data centers, including its existing 50V solutions.