AIBRIEF
Each 1.25 MW rack uses 342 distributed gallium nitride power modules to replace multi stage conversion and rack UPS battery sidecars, with 98.5% efficiency as Enphase's design target.
AI data centers today rely on a multi-stage AC-to-DC conversion chain to feed server racks, plus rack-level battery sidecars for backup. Enphase Energy says its IQ Solid-State Transformer (IQ SST) design collapses that chain into a single step from medium-voltage AC directly to 800 VDC and ±400 VDC, the rack voltages NVIDIA and the Open Compute Project are converging on for next-generation AI servers.
A solid-state transformer is power-conversion electronics that step voltage up or down using semiconductors instead of an iron core. Enphase's design, detailed in the company's technical white paper, stacks 342 intelligent power modules per 1.25-MW rack. Each module pairs bidirectional gallium nitride (GaN) switches operating below 1 kV with a 22-nm Kestrel control ASIC. The architecture aligns with NVIDIA's 800-VDC roadmap for Rubin Ultra in 2027 and Feynman in 2028 and OCP's Mt. Diablo ±400-VDC spec.
Enphase targets 98.5% system efficiency and 99.999% availability, and builds the design without internal batteries, a configuration meant to retire rack-level UPS sidecars. Those figures and the 1.25-MW rack size are Enphase's own design targets, not independent benchmarks, and volume deliveries are targeted for 2028.
Real risks: GaN device supply depth below 1 kV, the Kestrel ASIC's 22-nm tape-out and production timing, and slippage in the NVIDIA roadmap the design is aligned with.