データセンター

High-Voltage Direct Current (HVDC): High-voltage direct current power distribution reduces conversion stages and improves efficiency in high-density AI data centers.

Intermediate Bus Converter (IBC): A DC-DC converter that steps distribution bus voltage to an intermediate level for efficient downstream power delivery.

Battery Backup Unit (BBU): An energy storage system that provides temporary backup power during outages or power disturbances.

Capacitor Backup Unit (CBU): An energy storage subsystem that uses capacitors to provide instant backup power and stabilize the power bus during rapid load changes.

Peak Load Shaving Shelf (PLSS): An energy storage subsystem that supplements power supplies during short-term demand spikes.

Point of Load (PoL): The final power conversion stage that delivers the precise voltage required by processors, accelerators and memory devices.

Hot Swap: The ability to add or replace hardware without shutting down a system, supported by circuitry that manages inrush current and protects equipment.

Power Supply Unit (PSU): A power converter that delivers the voltage levels required by servers, storage systems and networking equipment.

Solid-State Transformer (SST): A power electronic device that converts medium-voltage electricity to lower voltages with greater efficiency and control than traditional transformers.

Uninterruptible Power Supply (UPS): A backup power system that provides immediate power during outages until longer-duration backup sources become available.

Smart Power Stage (SPS): An integrated power device that efficiently delivers high current to CPUs, GPUs, ASICs and other compute devices.

eFuse (Electronic Fuse): A semiconductor-based protection device that safeguards electronic systems from overcurrent, overvoltage, short-circuit and inrush-current events.

DC-DC Converter: A power conversion circuit that converts one DC voltage level to another for efficient power distribution and regulation.

Wide-Bandgap (WBG) Semiconductor: Advanced materials such as silicon carbide (SiC) and gallium nitride (GaN) that enable higher efficiency, power density and switching performance than silicon.

Rack Power Density: The amount of electrical power consumed within a server rack, typically measured in kilowatts (kW).

Bus Bar: A conductive power distribution component that efficiently transfers high current between electrical distribution equipment, servers and storage systems in a data center.

Vcore (Core Voltage): The regulated supply voltage delivered to the processing cores of CPUs, GPUs, ASICs and other compute devices. 

AI data centers are designed for extreme compute density and power consumption driven by GPUs and AI‑specific accelerators. Compared to enterprise or HPC data centers, they support hyperscale deployments with rack power scaling from tens of kilowatts today toward hundreds of kilowatts or even megawatt‑class racks, requiring advanced power conversion and distribution architectures.

AI data centers use multi‑stage power architectures including AC‑DC conversion, battery backup, intermediate bus conversion, multiphase DC‑DC regulation, and point‑of‑load (PoL) conversion. This layered approach ensures high efficiency, reliability, and precise voltage regulation for CPUs, GPUs, ASICs, and memory subsystems.

Raising distribution voltage from 12 V to 48 V and toward ±400 V or 800 V HVDC reduces conduction losses, cable size, and conversion stages. Higher‑voltage DC architectures improve power density and efficiency, enabling future AI racks to scale toward hundreds of kilowatts while optimizing space and system cost.

Wide bandgap (WBG) semiconductors such as silicon carbide (SiC) and gallium nitride (GaN) enable higher switching frequency, lower losses, and higher operating temperatures than silicon. These characteristics allow AI data center power supplies to achieve higher efficiency, greater power density, and reduced cooling requirements.

Hot‑swap and smart fuse solutions enable component replacement without system shutdown, improving uptime and serviceability. They provide inrush current control, fault protection, and telemetry via PMBus, which is especially important in high‑voltage 400 V and 800 V DC bus architectures used in AI data centers.