Efficient Power Conversion (EPC) has released new details on its EPC91123 evaluation board, an 800V-to-12.5V, 6kW isolated converter designed to support next-generation AI data-center power architectures as part of the NVIDIA MGX AI Factory ecosystem.
Power architecture simplification
The EPC91123 enables direct conversion from 800V DC to 12.5V DC, eliminating the need for a conventional 48V intermediate stage. This simplification improves overall system efficiency by 1–2%, a critical gain given the massive power consumption of AI data centers.
The converter uses an input-series, output-parallel (ISOP) LLC topology and leverages EPC’s Gen7 gallium nitride (GaN) devices, including the 40V, 0.8mΩ EPC2366 and the 150V, 2.2mΩ EPC2305. These components are housed in compact 3.3mm x 2.6mm packages.
Performance and form factor
The solution achieves 98.2% peak efficiency and 97% full-load efficiency. It delivers high power density in a compact 104mm x 47mm x 8mm form factor, optimized for space-constrained AI server environments.
Alejandro Pozo, EPC’s director of DC-DC system engineering, notes that the ISOP topology is well suited for high step-down-ratio conversion, enabling improved transformer optimization and interleaved operation. EPC plans to extend this approach to an 800V-to-6V platform.
Strategic alignment with NVIDIA MGX
The EPC91123 is part of EPC’s broader strategy for 800V DC AI power delivery, spanning conversion to 48V, 12V, and 6V to address different system-level requirements. Advanced GaN-based ISOP topologies and Gen7 eGaN technology aim to reduce distribution losses and support scalable accelerated computing.
EPC CEO Alex Lidow states that NVIDIA’s use of 800V DC architecture is a key element in addressing efficiency and power density demands as AI compute scales. EPC is contributing multi-level GaN-based solutions to the NVIDIA MGX ecosystem.
Forward-looking significance
By eliminating a power conversion stage and leveraging GaN’s inherent efficiency advantages, EPC’s approach directly addresses the escalating power demands of AI infrastructure. As data centers push toward higher voltage architectures, such compact, high-efficiency converters will be essential for enabling scalable, cost-effective accelerated computing platforms.
