Enteligent unveils PowerCore stack for AI data centres

Enteligent has introduced the PowerCore stack for AI data centres, designed to move power from an 800 VDC rack supply to GPU voltage rails.

The three-stage DC/DC conversion architecture is intended to support higher rack compute density as operators prepare for rising power demands from AI workloads.

The design spans the rack, server and GPU board. Enteligent says this coordinated approach is meant to reduce the space needed for power delivery and buffering while managing heat close to the processor.

AI infrastructure is pushing data centre power systems towards much larger rack loads. Enteligent cited industry forecasts that machine-learning systems could require more than 500 kW per IT rack before the end of the decade, while suppliers are also developing 800 VDC infrastructure for megawatt-scale racks.

That shift has increased attention on how electricity is delivered from facility power systems to chips. As rack power rises, the physical space taken up by low-voltage busbars, power shelves and transient buffering becomes a more significant design issue.

Three stages

Enteligent divides the conversion process into three modules. The first, RackCore, is a 100 kW rack power shelf that converts 800 VDC to a 50/48 VDC bus. This stage removes the need for supercapacitor and battery backup shelves, according to the company.

The second module, MaxCore, sits at server level and converts 50/48 VDC to a regulated 7 VDC rail. It uses a multi-phase, multi-path architecture intended to improve efficiency and simplify system design.

The third stage, ChipCore, sits near the GPU and converts the intermediate rail to 1.2/1.7 VDC. The module has 16 phases and supplies the final point-of-load or vertical power stage, the company says.

The modules can be assessed as a full rack-to-chip chain or used individually with compatible downstream power designs, giving equipment makers some flexibility in how they adopt the architecture.

System approach

The stack is based on treating the power conversion path as a set of linked electrical domains rather than a series of isolated components. In practice, that means considering voltage, current, frequency, distance and electromagnetic emissions together across the full conversion chain.

Enteligent argues this matters because, at AI-scale power levels, the most effective design is not always the one with the fewest conversion stages. Trade-offs must be judged across conversion loss, conductor current, transient response, heat generation, filtering, board space and system certification.

The company says the architecture could enable up to 50% greater rack compute density. That figure is tied to reducing the hardware associated with buffering and power delivery, which can otherwise occupy valuable rack space in high-density systems.

Heat is another constraint the design is intended to address. By coordinating conversion across multiple stages, Enteligent says it can reduce thermal pressure around GPUs, where power and cooling demands are already intense.

Sean Burke, Chief Executive Officer of Enteligent, described the announcement as part of a broader shift in data centre power design.

"As AI systems move toward much higher rack power, every conversion stage has to work as part of one system," said Sean Burke, Chief Executive Officer of Enteligent.

"The PowerCore stack draws on our DC-to-DC experience in high-density power electronics and computing to give AI infrastructure and server OEMs a modular path from 800 volts at the rack to the GPU," Burke said.

The launch places Enteligent in a market where interest in direct current architectures is growing as data centre operators and hardware makers look for ways to manage larger AI clusters. The move towards 800 VDC distribution reflects efforts to limit losses and practical constraints in conventional approaches as rack power climbs.

Enteligent, based in Silicon Valley, develops direct current power infrastructure for data centres and electric vehicle charging. Its product line includes rack-to-chip power delivery systems as well as DC-coupled charging equipment.