Navitas Collaborates with NVIDIA on 800 V-to-12 V Power Delivery Board for Next-Generation AI Infrastructure - EE Journal

NVIDIA's bet on 800V DC architecture is a tacit admission: at 1000W+ per GPU and 100kW per rack, the 48V bus has hit its physical ceiling. Navitas' GaN solution is not a "power board"—it is a paradigm shift in how energy flows through AI data centers. The cascading impact runs far deeper than the headline. A three-stage 800V-to-12V topology demands GaN HEMTs switching above 200kHz at >97% efficiency, structurally outperforming silicon MOSFETs. Upstream, GaN epi-wafer and substrate capacity will be consumed by AI orders within 18 months. Downstream, Vicor and TI must redraw their PMIC roadmaps; magnetics suppliers face a full materials-system overhaul. On competition, Infineon and ST carry deep SiC moats, but GaN's high-frequency efficiency edge is architectural, not incremental. NVIDIA's selection signals that the anchor supplier for next-gen AI power stacks is set. Infineon's most credible counter: accelerate SiC+GaN hybrid modules—but the window is roughly 12 months. The 12–24 month tail: 800V migrates from NVIDIA-exclusive to OCP open standard. GaN ASP drops 30%+ on volume. "Power efficiency = compute cost" becomes the dominant TCO metric in data center procurement. Whoever locks GaN capacity locks energy sovereignty in AI infrastructure.