GaN / SiC power
GaN (gallium nitride) and SiC (silicon carbide) are "wide-bandgap" semiconductors that switch electricity faster and waste less of it as heat than ordinary silicon. They are central to AI data centers, EVs, and fast chargers because they pack more power into smaller, cooler, more efficient converters.
What GaN and SiC actually are
Most chips that control electricity are made of silicon. **GaN (gallium nitride)** and **SiC (silicon carbide)** are different materials with a *wider bandgap* — meaning it takes more energy to push an electron into conducting. That sounds like a disadvantage, but it lets these materials hold off much higher voltages and switch on and off far faster without breaking down. They are collectively called **wide-bandgap (WBG)** semiconductors. Silicon's bandgap is about 1.1 eV; SiC is roughly 3.3 eV and GaN about 3.4 eV, and that gap is the whole story behind their performance.
How they move power more efficiently
A power converter's job is to take electricity at one voltage and deliver it at another — for example, turning wall power into the low voltage a chip needs. It does this by switching transistors on and off thousands or millions of times per second. Every switch wastes a little energy as heat. Because GaN and SiC switch faster and resist heat better, they lose less energy per switch and can run at far higher frequencies. Higher frequency means the bulky magnetic and capacitor components shrink, so the whole converter gets smaller, cooler, and more efficient. Modern GaN-based supplies routinely hit 96–98% efficiency versus the lower-90s typical of older silicon designs.
GaN vs. SiC: not rivals, but a division of labor
They are often discussed as competitors, but in practice they cover different territory. **GaN** excels at lower voltages (roughly 100–650 V) and very high switching frequencies, making it ideal for compact server power stages, laptop and phone chargers, and on-board EV chargers. **SiC** dominates higher voltages (1,200 V and well beyond) with extreme thermal robustness, which suits EV traction inverters, solar inverters, grid equipment, and high-voltage data-center feeds. Many cutting-edge designs now use *both* — GaN for the fast, lower-voltage stages and SiC for the high-voltage backbone.
Why this matters for AI and data centers
AI accelerators are extraordinarily power-hungry, and a modern GPU rack can draw tens of kilowatts. Delivering that much power efficiently — and getting rid of the waste heat — has become a binding constraint on AI buildout. The industry is shifting from traditional 12 V/48 V distribution toward **800 V high-voltage DC (HVDC)** architectures to cut losses across the rack. In 2025 NVIDIA selected Navitas to help develop its next-generation 800 V HVDC architecture for rack-scale systems, using GaN and SiC, with claims of up to ~5% better end-to-end power efficiency and sharply lower cooling and maintenance costs. Every percentage point of efficiency saved at this scale translates into enormous energy and cost savings.
Where they sit in the supply chain
GaN and SiC are *power* semiconductors, distinct from the logic chips (like GPUs) that do computation and from the optical/photonics parts that move data. They live in the **power-delivery** layer: wall-to-rack converters, server power-supply units, and the voltage-regulation modules that feed the processor. Companies like **Vicor (VICR)** build the power modules and 'vertical power delivery' that step voltage down right next to the GPU at very high current. So in an AI server, GaN/SiC devices and the modules built around them form the energy plumbing that keeps the expensive compute silicon fed and cool.
Who the key players are
The broad SiC and GaN power market is led by large diversified chipmakers: **Infineon** (the overall leader), **STMicroelectronics**, **onsemi**, **Wolfspeed (WOLF)**, **ROHM**, and **Power Integrations**, plus specialists like **Innoscience** in GaN. Among the names tied to AI-power momentum: **Navitas (NVTS)** supplies GaNFast and GeneSiC devices and is collaborating with NVIDIA on 800 V HVDC; **Wolfspeed (WOLF)** is a pure-play SiC maker that launched the first commercial 10 kV SiC MOSFET and reports growing AI data-center sales; and **Vicor (VICR)** focuses on the power-module and delivery architecture. These tickers are examples for context, not recommendations.
What's changing now
The technology is maturing fast and demand is broadening. Estimates put the combined GaN/SiC power market above ~$4 billion in 2025 with strong projected growth, driven by EVs, renewables, and especially AI data centers. Key shifts: the move to 800 V HVDC in AI racks, higher-voltage SiC parts (Wolfspeed's 10 kV MOSFET), tighter GaN+SiC co-packaging (Infineon and ROHM announced joint SiC packaging in 2025), and government-backed manufacturing expansion (onsemi secured EU support for SiC fabs). The direction is clear: as compute and electrification scale, wide-bandgap power becomes more, not less, essential.
Frequently asked
Neither is universally better — they suit different jobs. GaN wins at lower voltages (under ~650 V) and very high switching frequencies, ideal for chargers and server power stages. SiC wins at high voltages (1,200 V and up) with superior heat tolerance, ideal for EV inverters, grid, and high-voltage data-center power. Many advanced systems use both together.
AI racks draw enormous power, and efficiency plus heat removal have become limiting factors. GaN and SiC switch faster and waste less energy than silicon, enabling 96–98% efficient converters and the new 800 V high-voltage DC architectures that hyperscalers like NVIDIA are adopting. At data-center scale, even a few percent of saved energy is huge.
No. GPUs are logic chips that perform computation. GaN and SiC are power semiconductors that manage and convert electricity to feed those GPUs. They sit in the power-delivery layer of the system, not the compute layer.
Plenty you may already own. GaN powers compact fast chargers for phones and laptops and high-efficiency server supplies. SiC is widely used in electric-vehicle drivetrains, solar inverters, fast EV charging stations, and grid equipment.
Large players include Infineon, STMicroelectronics, onsemi, Wolfspeed, ROHM, and Power Integrations. AI-power-focused names include Navitas (NVTS), the SiC pure-play Wolfspeed (WOLF), and power-module maker Vicor (VICR). These are examples for context, not investment advice.
Silicon still handles most power electronics, but it is reaching physical limits. It can't switch as fast or hold off as much voltage without losing efficiency and overheating. Wide-bandgap materials push past those limits, enabling smaller, cooler, more efficient power systems where silicon would struggle.
Related companies
Related topics
Sources
- NVIDIA Selects Navitas to Collaborate on Next Generation 800 V HVDC Architecture
- Scaling AI Data Center Power Delivery with Si, SiC, and GaN (Infineon white paper)
- Wide Bandgap Semiconductors for Power Electronics: Comparative Properties of GaN and SiC (MDPI)
- GaN vs. SiC: Everything You Need To Know (Power Electronics News)
- Powering Clustered AI Processors (Vicor)
Educational explainer · not investment advice. Part of the learn series.