Enteligent is targeting the escalating power demands of AI infrastructure by introducing a modular three-stage conversion architecture designed to bridge the gap between high-voltage rack supplies and GPU voltage rails. As machine-learning systems approach projected power requirements exceeding 500 kW per IT rack by 2030, the company is positioning its PowerCore stack as a solution to the physical and thermal constraints of traditional power distribution. This modular approach aims to facilitate a transition to 800 VDC distribution while potentially increasing rack compute density by up to 50% through reduced hardware buffering.
PowerCore Modular Conversion Architecture
The PowerCore stack utilizes a three-stage DC/DC conversion process to manage the transition from an 800 VDC rack supply to the specific voltage requirements of GPUs. The first stage, the RackCore™ module, functions as a 100 kW rack power shelf that converts 800 VDC to a 50/48 VDC bus. Enteligent claims this module eliminates the requirement for costly super capacitor and battery backup shelves, which the company notes have limited lifespans. The second stage involves the server-mounted MaxCore™ module, which converts the 50/48 VDC bus to a regulated 7 VDC rail using a multi-phase, multi-path architecture. Finally, the ChipCore™ module, a compact 16-phase component located near the GPU, converts the intermediate rail to a 1.2/1.7 VDC point-of-load or vertical power stage. By treating these stages as a coordinated system rather than independent components, Enteligent intends to optimize voltage, current, and electromagnetic emissions across the entire conversion chain.
Addressing AI Data Center Power Constraints
The introduction of the PowerCore stack responds to a shift in industry standards toward 800 VDC infrastructure for megawatt-scale racks, a move supported by developments from companies like NVIDIA. As power density rises, the physical space required for low-voltage busbars, power shelves, and transient buffering has become a critical design bottleneck for data center operators. Enteligent is positioning its architecture to mitigate these constraints by reducing the need for heavy buffering hardware and minimizing heat generation at the GPU level. The company suggests that at AI-scale power levels, the most effective architecture is not necessarily the one with the fewest conversion stages, but rather the one that minimizes the cumulative costs of conversion loss, conductor current, transient response, and board space. This system-level approach is designed to provide AI infrastructure and server OEMs with a modular path to manage the complexities of high-density power delivery from the facility bus directly to the silicon.
Key Takeaways
- The PowerCore stack utilizes a three-stage architecture consisting of RackCore™, MaxCore™, and ChipCore™ modules to convert 800 VDC to GPU-ready voltages.
- Enteligent claims the RackCore™ module can enable up to 50% higher compute density by eliminating the need for super capacitor and battery backup shelves.
- The architecture is designed to address rising AI rack power requirements, which Google projects could exceed 500 kW per IT rack before 2030.
TechInsyte's Take
In our view, Enteligent is making a calculated bet that the primary bottleneck for next-generation AI scaling is no longer just raw compute, but the physical physics of power delivery. By moving toward an 800 VDC-to-chip model, they are addressing the "density wall" where traditional low-voltage distribution consumes too much rack real estate and generates excessive heat. If the PowerCore stack successfully reduces the reliance on bulky buffering hardware like super capacitors, it could significantly alter the CapEx models for hyperscale operators. This signals a broader industry shift where power electronics must be designed as a holistic, system-level component of the AI stack rather than an afterthought of the facility infrastructure.
Questions & Answers
How does the PowerCore stack impact rack compute density?
The company claims the PowerCore stack can enable up to 50% greater rack compute density by utilizing a modular three-stage conversion process that requires less buffering hardware and reduces heat at the GPU.
What specific voltage conversion stages are included in the architecture?
The stack includes the RackCore™ module (800 VDC to 50/48 VDC), the MaxCore™ module (50/48 VDC to 7 VDC), and the ChipCore™ module (7 VDC to 1.2/1.7 VDC).
Why is the industry moving toward 800 VDC distribution?
As AI rack power requirements rise—with projections suggesting over 500 kW per rack by 2030—800 VDC is emerging as a standard to manage the space and efficiency constraints associated with low-voltage busbars and power shelves.
Can the PowerCore modules be used independently?
Yes, the company states that the modules can be evaluated as a complete chain or integrated individually with compatible downstream power designs.
Source: Enteligent
