STMicroelectronics announced the expansion of its 800 V DC power conversion portfolio with two new architectures: 800 V to 12 V and 800 V to 6 V, aligned with NVIDIA 800 V DC data center reference design. These new solutions complement the previously introduced 800 V to 50 V architecture and support the industry transition toward higher-efficiency, high-density AI infrastructure.
The move reflects the rapid evolution of AI data centers toward higher voltage distribution to reduce losses and improve scalability. By expanding to 12 V and 6 V intermediate rails, ST addresses different server architectures, GPU requirements and thermal constraints across hyperscale deployments.
The 800 V to 12 V conversion stage enables direct power delivery from rack-level systems to voltage domains used by advanced AI accelerators. By eliminating the traditional 54 V intermediate stage, it reduces conversion steps, lowers system losses and simplifies integration, contributing to improved overall rack efficiency and reduced copper usage.
The 800 V to 6 V architecture is designed for systems requiring power delivery closer to GPUs. This approach minimizes resistive losses and improves transient response, which is critical for high-performance AI training clusters and dense compute environments. It also enables further reduction in conversion stages, improving efficiency and system-level performance.
ST emphasized that 50 V, 12 V and 6 V intermediate buses will coexist depending on server design, GPU configuration and cooling strategies. This flexible approach allows system designers to optimize power delivery architectures based on specific application requirements.
The new solutions build on ST’s earlier demonstration of a GaN-based 800 V converter operating at 1 MHz with over 98 percent efficiency and power density exceeding 2,600 W/in³. The expanded portfolio integrates ST’s technologies across silicon, silicon carbide and gallium nitride power devices, along with analog, mixed-signal and microcontroller solutions, providing a complete ecosystem for next-generation AI server power architectures.
Original – STMicroelectronics