GaN Power Chips Target AI Server Efficiency
Innoscience's gallium nitride products, shown at a major open-hardware summit, aim at the power bottleneck in AI data centers.
Updated on Nov. 3, 2026 in Data Centers
Innoscience, a China-based gallium nitride (GaN) power semiconductor company, showcased its GaN power solutions for AI infrastructure at the Open Compute Project Global Summit in late October. The company also participates in the NVIDIA MGX ecosystem. The showcase is a product presentation; the extraction does not indicate a specific shipping date or availability.
Why it matters
As AI compute scales, power has become a defining constraint on data-center performance, pushing demand for more efficient power components. Innoscience's pitch centers on GaN devices, which the company says enable higher power density and lower energy losses in AI servers and data-center power supplies.
The figures cited are market projections rather than measured results: a 35% CAGR for global power GaN devices reaching about US$3.5 billion by 2031, and a 45% CAGR for infrastructure and telecom applications including AI data centers. The extraction provides no benchmark for Innoscience's own products, and the projections' underlying methodology is not stated.
The players
Innoscience
A China-based gallium nitride power semiconductor company that builds GaN power solutions and operates a vertically integrated GaN manufacturing platform with high-volume 8-inch GaN-on-silicon production.
NVIDIA MGX
NVIDIA's modular server reference architecture ecosystem, in which Innoscience participates.
The details
Gallium nitride is a semiconductor material that can switch electrical power faster and with less energy loss than conventional silicon, allowing power-conversion circuits to be made smaller and more efficient. Innoscience says its design, process and manufacturing are combined under one roof, a vertically integrated approach it describes as the world's largest GaN manufacturing platform. The company also reports leading the industry in high-volume production on 8-inch GaN-on-silicon wafers, a larger wafer format that can lower the per-chip cost of production.
Timeline
Late October 2026: Innoscience showcases its GaN solutions at the Open Compute Project Global Summit.
2031: The global power GaN device market is projected to reach approximately US$3.5 billion.
2033: The broader GaN semiconductor devices market is projected to reach approximately US$28.9 billion.
The Tech Race
Innoscience is positioning GaN power devices inside NVIDIA's MGX modular server ecosystem, a path that places its components in the AI server supply chain rather than in standalone consumer chargers. The extraction does not indicate how Innoscience's GaN products compare with competing power-semiconductor suppliers in this segment.
The extraction does not state when Innoscience's GaN solutions will be available to data-center operators or server makers, nor pricing. Any practical effect on AI server power supplies depends on the timeline for adoption within the NVIDIA MGX ecosystem, which is not given.
The takeaway
Power efficiency is becoming a limiting factor in AI data centers, and GaN is one of the materials being pushed into that role. Watch for named customer adoptions or ecosystem designs that disclose GaN power components in production AI servers.
Further reading
For broader context on how power efficiency is reshaping the racks behind AI, see Data Centers.
Source note: This article includes information reported by TrendForce.
