Table of Contents
GaN could be a game-changing semiconductor material that brings energy efficiency to AI data centers
Gallium Nitride (GaN) is increasingly used in data centers for improving power supply efficiency, increasing power density, and reducing energy consumption. With faster switching speeds and lower losses, it opens the door to smaller, more efficient power converters so that data center engineers can fit more servers in the same space, and at a lower operational cost.
Like aerospace, automotive, industrial and defense industries, the AI data center challenge is to squeeze more power out of tight spaces and to address growing thermal constraints.
In terms of high-temperature chips, GaN transistors have performed at 1472 Fahrenheit (800 degrees Celsius), surpassing what had been achieved previously with Silicon carbide chips (at 600 degrees).
Its smaller form factor and ability to handle higher voltages and temperatures with lower resistance and faster switching speeds translates into less wasted energy, and less heat. “
“AI is driving a rapid transformation in data center architectures, and GaN is at the forefront of this change,” said Johannes Schoiswohl, the head of GaN business at Infineon Technologies, in a statement to RCR Tech. “Building on its traditional strengths in power supply units (PSU), GaN is now being adopted in battery backup units (BBUs) and intermediate bus converters (IBCs) – critical systems that benefit from GaN’s high-power density and efficiency.”
He notes that with the acceleration of AI adoption, GaN’s ability to support innovative topologies makes it uniquely qualified to meet the increasing power density requirements and form factor constraints in AI data centers. He explains that Infineon is leveraging GaN to enable next-gen high-density AI powertrains, “delivering up to 4x higher power density which results in higher overall data center efficiencies enabling a lower carbon footprint.” He added that for overall AI power supply solutions, Infineon expects a to generate revenue of around €1.5 billion in this area in the 2026 fiscal year.
As Schoiswohl pointed out, the power-supply units (PSUs) are an important piece, in many instances considered the backbone of data centers. As a result, PSUs designed for traditional server configurations are struggling to keep up with the demands of GPU-based AI accelerators. That’s where GaN comes in, with efficiency, power density, and thermal performance that is superior to traditional silicon-based components. That’s why GaN is increasingly used in PSUs and power conversion systems in data centers.
“GaN provides a better switch, with less switching losses for a given size, so the RDs on per unit area for a wide bandgap device. Is much lower than for a silicon switch, explained Andy Smith, director of training at Power Integration, during PCIM 2025. “The transistors for involved for a silicon carbide or GaN are much smaller.”
Higher-frequency switching enables device simplification and improved efficiency, as well as opportunities for the development of bidirectional topologies and greater functional integration that will shorten design cycles and improve system efficiency.
Reality Check
While GaN provides significant efficiency gains for data centers, its unique material physics introduce reliability challenges that standard silicon (Si) testing cannot fully address. To bridge this gap, manufacturers are moving toward integrated power stages and advanced packaging to ensure robustness in high-stress environments.
For now, data center engineers are sometimes addressing shortcomings by using hybrid approaches that combine silicon with wide-bandgap (WBG) semiconductor materials such as silicon carbide (SiC) and GaN to improve switching performance and thermal stability under heavy AI workloads.
As the cost of incorporating GaN into conventional electronics comes down, it is expected to find more commercial applications. For example, researchers from MIT recently came up with a new 3D chip fabrication process to integrate high-performance GaN transistors onto standard silicon chips. Their method involves building many tiny transistors on the surface of a GaN chip, cutting out each individual transistor, and then bonding just the necessary number of transistors onto a silicon chip using a low-temperature process that preserves the functionality of both materials.
That triggered the launch of a spinout company, Vertical Semiconductor, to commercialize the technology for AI data centers.
By integrating high-performance GaN transistors onto standard silicon CMOS chips, the cost comes down and the technology becomes more scalable and compatible with existing semiconductor foundries. As that happens, GaN will from niche applications to mainstream adoption for energy efficiency, digitalization, and decarbonization across many sectors.
