GaN Talk Blog

By: Maurizio Di Paolo Emilio, Contributing Editor at Data Centre Digest

Artificial intelligence is pushing data-center power architectures into territory they were never designed to handle. As AI accelerators become more powerful, rack power is moving from tens of kilowatts toward hundreds of kilowatts and, ultimately, the 1 MW range. At that scale, delivering electricity efficiently is no longer simply a matter of selecting better power components. It becomes a system-level architectural challenge.

One of the most significant responses is the move toward 800 VDC distribution. NVIDIA is developing an 800 VDC architecture intended to support 1 MW-class racks and beyond, while large AI infrastructure projects are driving demand for power capacity measured in multiple gigawatts.

Gallium nitride (GaN) is entering a new phase of development. After establishing itself in applications such as lidar, satellite power, fast chargers, and AI power delivery, the technology is increasingly moving toward lower-voltage, higher-current applications where switching speed, low conduction losses and power density are critical.

For Efficient Power Conversion (EPC), this transition is particularly important. The company has built its GaN portfolio around enhancement-mode eGaN® technology and has consistently pushed GaN into applications where conventional silicon MOSFETs face limitations in switching performance, size and efficiency. The development of low-voltage devices below 100 V now opens additional opportunities in AI power delivery, robotics, motor drives and other high-current systems.

Encouraging developers to design high-performance Class-D audio amplifiers leveraging GaN FETs, Efficient Power Conversion (EPC) has launched the new EPC9192 reference design, enabling high power and high efficiency Class-D audio amplifiers in a modular design. audioXpress reviews the EPC9192 to confirm the performance capabilities of EPC’s 200V, EPC2307 eGaN FETs in a ground-referenced, split dual supply single-ended (SE) design, rated to deliver 700W per channel into a 4Ω load.

Qian Luo, Author, WBG TechnoBite

In humanoid robot wrists and dexterous hands, as well as compact drone propulsion systems, motor drives are moving from external controller boards into the actuator itself. At that point, the inverter is no longer an independent unit that simply delivers three-phase current to the motor; it must share a tightly constrained space with the motor, position sensor, mechanical structure, and thermal path.

This changes the design priorities for compact motor drives. Continuous current and torque capability certainly depend on the power devices, but where the DC-link capacitors can be placed, whether heat can be transferred into the housing or airflow, and how sensing and protection signals are routed all shape the final system performance.

As AI infrastructure, humanoid robots, drones, autonomous machines and space systems push power electronics toward higher efficiency and power density, gallium nitride (GaN) is moving into increasingly demanding applications. At PCIM Asia 2026 in Shenzhen, Efficient Power Conversion (EPC) will showcase its latest GaN technology while offering a preview of where the company believes the technology is heading next. The event, taking place August 26–28, will be particularly relevant for designers working on AI power delivery and intelligent motion systems. EPC will demonstrate its GaN portfolio across the power-conversion chain, from 800 VDC distribution to point-of-load conversion, while its roadmap points toward two developments with potentially broader implications: Gen 8 GaN for ultra-high-frequency, low-voltage power conversion and Trinity, EPC’s approach to highly integrated GaN motor drives.

Watch the video

Gallium nitride (GaN) is transforming power electronics from low-voltage motor drives to megawatt-class AI data centers. In this interview, EPC details how seventh- and upcoming eighth-generation GaN devices outperform silicon by up to 5x in capacitance, area, and power density, unlocking 40V-and-below markets and advanced point-of-load conversion for GPUs. The discussion spans high-performance robotics on 48 V buses, multi-level ISOP topologies for 800 V systems, rapid prototyping via reference designs, and the growing role of GaN in space applications.

The EPC91123 is a 6 kW GaN-based DC‑DC converter reference design that delivers an isolated 800 V to 12.5 V conversion for AI servers and high‑performance computing. It uses an input‑series output‑parallel (ISOP) architecture with eight interleaved LLC modules, each processing about 750 W. Primary switches are 150 V EPC2305 GaN FETs, while secondary synchronous rectification uses 40 V EPC2366 devices, enabling operation around 1 MHz. This architecture reduces magnetics size, EMI, and thermal hotspots, achieving over 98% peak efficiency and more than 97% efficiency near full load at nearly 500 A output.

At PCIM, Mike Engelhardt, creator of LTspice and QSPICE, explains why traditional SPICE—originally built for IC design—struggles with accurately simulating power devices, especially GaN FETs. He reviews historical capacitance models (Meyer, Yang–Chatterjee), their limitations for VDMOS and GaN, and describes how QSPICE introduces a native GaN MOSFET model (level 2026) with accurate charge, output capacitance, and quasi-saturation behavior. Engelhardt details numerical improvements, proper handling of transcapacitances, and performance gains that make QSPICE a faster, more robust power electronics simulator.

Artificial intelligence is moving from chatbot conversations to autonomous AI agents working together on huge data sets – and the data center is becoming an AI factory. These next generation facilities require unprecedented computer power which in turn brings new challenges of power delivery, thermal management and energy efficiency. To help address these challenges, NVIDIA has created the NVIDIA MGX™ architecture, a modular platform that accelerates the deployment of AI infrastructure. But the real potential of MGX can only be realized with equally advanced power conversion technology.