The three-stage rack-to-chip architecture aims to simplify high-density power delivery as AI infrastructure moves towards megawatt-scale racks.
Enteligent has introduced its PowerCore™ DC-based conversion stack, a three-stage architecture designed to deliver power from an 800VDC data centre rack supply to the low-voltage rails required by GPUs. The company says the integrated approach can enable up to 50% higher compute density from facility power to server by coordinating power conversion across the rack, server and GPU board.
The launch comes as AI workloads drive up rack power requirements, making power delivery, heat management and physical space increasingly important infrastructure design considerations. As the industry moves towards 800VDC distribution, conventional power shelves, low-voltage busbars and transient buffering hardware can create constraints on rack space and system design.
PowerCore divides conversion into three modules. RackCore™ is a 100kW rack power shelf that converts 800VDC to a 50/48VDC bus. Enteligent says its design can eliminate the need for separate supercapacitor and battery backup shelves, reducing hardware requirements while improving compute density and system reliability.
“As AI systems move toward much higher rack power, every conversion stage has to work as part of one system,” said Sean Burke, CEO of Enteligent.
MaxCore™ is mounted at the server level and converts the intermediate bus voltage to a regulated 7VDC rail. ChipCore™, a compact 16-phase module positioned near the GPU, then converts that rail to 1.2V or 1.7V for the final point-of-load or vertical power stage.
The modular architecture can be evaluated as a complete conversion chain or integrated in individual stages with compatible downstream designs. Enteligent says its system-level approach considers efficiency, current, heat, transient response, electromagnetic emissions, filtering and board space across the entire power path rather than optimising each conversion stage independently.
“As AI systems move toward much higher rack power, every conversion stage has to work as part of one system,” said Sean Burke, CEO of Enteligent. He said the stack provides AI infrastructure and server original equipment manufacturers with a modular route from 800 volts at the rack to the GPU.
The announcement underscores how power architecture is becoming a critical factor in scaling AI infrastructure. Enteligent is positioning PowerCore as a way for equipment designers to balance conversion efficiency, thermal performance, reliability and space constraints as GPU power demands increase.