by Chris DePuy, 650 Group / October 6, 2026
Renesas announced its first 100V enhancement-mode GaN discrete power transistors, carrying its GaN portfolio into the power stages closest to AI server processors. The RTP100E005G1FL, RTP100E2P6G1FL, RTP100E1P8G1FL-DSC, and RTP100E1P2G1FL-DSC parts keep a silicon-compatible footprint for drop-in migration, ship in MOSFET-compatible packages with bottom- and dual-side cooling, and span RDS(on) of 5 mΩ to 1.2 mΩ.
Renesas claims up to 35% lower hard-switching FOM and 63% lower soft-switching FOM against comparable GaN parts, zero reverse-recovery charge (Qrr = 0), and 40–70% lower switching losses with as much as double the system-level power density in 800 VDC architectures. The physics: AI-rack converters historically run at a few hundred kilohertz, but 800 VDC distribution and 48V buses are pushing conversion stages toward megahertz-class switching to cut magnetics volume and shrink passives. We think the timing is pointed: rack densities on our AI infrastructure forecast path make 800VDC the near-term voltage, and the 48V-and-below conversion layer is where the next vendor fight forms — Infineon just closed its C2i acquisition for software-defined multiphase control, Flex moved 800V up the stack with EPC Power, and a silicon-footprint GaN FET is an easy second-source fight to join. We are bullish on AI power capex: every efficiency point in that layer compounds across a fleet.
650 Group researches several of the technologies we covered in this blog, including Data Center / Co-Location / Cloud, Enterprise, and more.
