AI Server Power Supply Specifications

Modern AI server PSUs typically deliver 10–12 kW per unit, achieve >97% efficiency, and use high-voltage DC distribution with advanced Si, SiC, and GaN semiconductors for high power density and rapid load response. Power Rating and DensityAI server PSUs are designed for high-performance workloads, with reference designs delivering 12 kW per unit and power densities exceeding 100 W/in³, enabling compact integration in server racks and AI data centers (Infineon, onsemi, Yole).

Future A

I racks are projected to require up to 1. 5 MW per rack, necessitating high-voltage bus architectures to reduce current and copper volume (Texas Instruments).

Efficiency and Standards

Efficiency is critical due to high energy consumption.

Modern PS

Us achieve 97–98% peak efficiency, with 80 PLUS Titanium compliance requiring >90% efficiency across light to full load and a power factor ≥0. 95. Active power factor correction (PFC) is standard, often implemented with 3-level flying capacitor PFC, interleaved boost, or bridgeless totem-pole topologies.

Voltage and Bus Architecture

To minimize distribution losses, AI server PSUs are moving from 400V to 800V DC bus systems, reducing current from 20 kA to ~1. 25 kA for high-power racks, which allows smaller, lower-density busbars while maintaining efficiency (Texas Instruments, Yole). Typical low-voltage outputs for compute chips are 97% peak, 80 PLUS Titanium compliance.

Voltage: 800V DC bus for distribution,