Navitas collaborates with NVIDIA on 24kW, 800V-to-12V power delivery board for next-gen AI infrastructure - Semiconductor Today Magazine

The AI compute scaling bottleneck has shifted from transistor geometry to power architecture. Navitas' design win with NVIDIA on a 24kW, 800V-to-12V board is not a component-level transaction—it is a declaration that the 48V data center distribution era is over. The technical implication is structural. GaN's switching-frequency advantage at the 800V node delivers a step-change in power density over SiC alternatives, shrinking the physical footprint of power stages inside already thermally constrained GPU racks. This ripples upstream into GaN epitaxial wafer supply and power-module packaging capacity, and downstream into NVIDIA's entire system topology: UPS, transformers, and liquid-cooling loops must all be re-architected around an 800V bus. On competition, Infineon, ON Semiconductor, and STMicroelectronics hold deep high-voltage power portfolios, but the GaN window in AI data centers is narrow. NVIDIA's role as de facto architecture definer means its supplier choices will harden into industry standards within two product cycles. Late entrants face a binary: conform or be excluded. Within 12–24 months, 800V will likely become the default for new AI facility builds, and GaN penetration in data-center power devices will jump from single digits into double digits. Power delivery is about to overtake HBM and advanced packaging as the binding constraint on AI infrastructure.