Wisevoter
World

Data Centers Have Adopted Gallium Nitride for AI

New power architectures are replacing silicon to meet the extreme electricity demands of modern AI processing.

Updated on Sept. 21, 2026 in Data Centers

Isometric editorial illustration of geometric hexagonal server components organized on a motherboard, representing advanced power architecture for artificial intelligence systems.
Data centers are increasingly adopting gallium nitride power architectures to overcome the efficiency limitations of traditional silicon in supporting high-density AI processing hardware. AI Illustration. Upload story photo >

Live Poll

Do you expect the expansion of AI infrastructure to increase your household's electricity costs?

Data centers are shifting to gallium nitride (GaN) power architectures to support the surging energy requirements of artificial intelligence. Traditional silicon power devices have reached their functional limits as AI accelerators demand increasingly higher power density.

Why it matters

Silicon MOSFET technology only improves by about 20 percent per generation, leaving it unable to handle the gigawatt-scale power demands of current AI hardware. Adopting GaN improves total power conversion efficiency by 5 percent from the input to the GPU core.

GaN technology achieves 3 MHz switching frequencies in synchronous buck converters and enables direct 800 V DC distribution to server racks. The industry is targeting future switching frequencies of 10 MHz and current densities of 5 A/mm².

The players

EPC

The company is a manufacturer of gallium nitride-based power management devices used in high-performance computing.

EE Times

This is a long-standing trade publication that covers electronics engineering news and technology developments.

The details

Data centers are now integrating power conversion directly onto server boards to eliminate bulky power supply drawers that cannot manage the intense current levels. GaN devices, which have eight years of deployment history, are manufactured on silicon substrates using existing production equipment.

Timeline

  1. EPC seventh-generation GaN devices have been deployed over an eight-year history on AI data center cards.

  2. Individual GPUs are projected to dissipate 5 kW of power within 2 years.

  3. The article was originally published in an e-guide in July 2026.

  4. The information was published on the EE Times website in September 2026.

The Tech Race

This transition marks a definitive move away from silicon-based power architectures as the industry faces performance ceilings that threaten AI scalability. By moving to GaN, firms are prioritizing power density over legacy manufacturing methods to remain competitive in the global AI hardware race.

As data centers achieve higher efficiency, they can support the dense computing power necessary for faster and more complex AI models. Users may experience improved performance and broader access to advanced artificial intelligence tools as data center hardware becomes more energy-dense.

The takeaway

The move to gallium nitride represents a necessary evolution in power electronics to sustain the rapid advancement of artificial intelligence systems. Engineers and industry stakeholders should monitor switching frequency developments as they serve as a leading indicator of future GPU power capacity.

Further reading

For more information on infrastructure trends, see the latest updates in Data Centers.

Live Poll

Do you expect the expansion of AI infrastructure to increase your household's electricity costs?