Summary
- Cambridge GaN Devices and NXP have formed a long term collaboration around GaN based system solutions.
- The work targets data centre, automotive, and industrial power conversion markets where efficiency and power density are under pressure.
- Rising AI rack densities are making power electronics part of the wider data centre infrastructure constraint.
Cambridge GaN Devices has formed a long term collaboration with NXP Semiconductors to develop GaN based power conversion solutions for data centre, automotive, and industrial applications.
The collaboration gives NXP access to CGD’s gallium nitride products, early access to next generation CGD developments, and CGD’s process and technology expertise.
CGD will gain access to NXP’s processor and analogue product portfolios, system know how, and global commercial reach.
The companies are targeting system level power conversion rather than individual devices, a useful distinction as AI racks place heavier demands on every conversion stage between the grid connection and processor load.
CGD says GaN semiconductors can switch at higher frequencies and achieve greater efficiency than competing technologies.
Its ICeGaN technology is designed to support higher switching frequency, higher power density, and simpler use with standard drivers compared with more complex discrete GaN implementations.
The data centre power density figures in CGD’s material show the scale of the pressure on electrical design.
According to the company, a rack that might have consumed 40kW in 2022 can now draw 200kW or more, with some future racks expected to require 1MW or more by 2030.
Efficiency moves deeper into the power chain
Data centre power is often discussed at grid and substation level, yet facility efficiency is shaped by a chain of conversion stages inside the building and the IT equipment.
Power moves through high voltage infrastructure, transformers, switchgear, UPS systems, distribution, server power supplies, voltage regulator modules, and point of load conversion before it becomes useful compute.
Losses at device and module level can become meaningful when multiplied across thousands of servers, because avoidable heat pushes up cooling demand and reduces usable electrical headroom.
Higher switching frequencies can support smaller passive components and more compact designs, which can affect footprint, weight, thermal behaviour, and packaging close to the load.
Adoption will depend on whether the technology is robust, manufacturable, easy to design around, and supported by system level documentation that equipment makers can use at scale.
NXP’s role gives the collaboration more reach across processors, analogue portfolios, reference designs, and customer channels.
CGD’s Cambridge base adds a UK supply base angle, since domestic semiconductor and power electronics capability can feed into the infrastructure demands created by AI.
The automotive element also gives the collaboration a second route to scale, as EV traction inverters and data centre power systems share pressures around efficiency, density, reliability, thermal performance, and cost.
For data centres, the power bottleneck is not confined to grid connections, because inefficient conversion, thermal limits, space constraints, and component reliability can all reduce usable capacity.
If GaN based systems can improve efficiency and density at scale, they will sit alongside substations, UPS systems, batteries, liquid cooling, and heat rejection in the physical response to AI demand.
