New research outlines how DC-native power infrastructure can improve efficiency, scalability and revenue density in next-generation AI and GPU data centres.
Enteligent, a developer of Direct Current (DC) power infrastructure for high-density loads, has released a new technical white paper titled 800VDC-to-50VDC Power Delivery Architecture: Completing the DC-Native Power Stack for AI-Scale Data Centers.
The report examines how next-generation data centres can improve electrical efficiency and power density by adopting high-voltage Direct Current (HVDC) distribution combined with rack-level DC power conversion.
The white paper builds on research published by Enteligent in February 2026 that explored the advantages of facility-level 800VDC power delivery for next-generation AI and GPU data centres. These include improved power efficiency, lower total cost of ownership and greater scalability.
As AI workloads push rack power levels beyond 100 kW and facilities move toward multi-hundred-megawatt and gigawatt-scale deployments, traditional AC-based electrical infrastructure is approaching practical limits. A DC-native architecture that distributes power at 800VDC across the facility and converts it to 50VDC within the rack aligns more closely with the internal power infrastructure used by modern servers.
Sean Burke, CEO, Enteligent, said: “AI is fundamentally reshaping data centre power requirements. While 800VDC distribution solves the upstream limitations of traditional AC infrastructure, converting it directly to a 50VDC server bus within the rack addresses the final conversion bottleneck at the server level. The combination of 800VDC facility distribution with rack-level 800VDC-to-50VDC conversion represents a complete, DC-native power architecture that aligns with the realities of modern data centre requirements.”
The research highlights the advantages of centralising AC-to-DC conversion into fewer, higher-utilisation systems supported by compact conductor infrastructure and efficient rack-level converters that serve the IT load. The study found that the lowest total cost of ownership and highest operational efficiencies occur when HVDC is implemented as a facility-native architecture combined with rack-level 800VDC-to-50VDC conversion.
According to the report, an AI rack operating within a DC-native architecture can generate 10 to 15 times more revenue than a traditional rack occupying the same physical space.
Frank Smith, VP of Growth, Claros, said: “Power density is revenue density, and this architecture unlocks both. Every percentage point of efficiency recovered from unnecessary AC conversion translates directly into lower operating costs, reduced cooling infrastructure, and more usable compute per square foot of data centre floor space. Claros and Enteligent share the same perspective that by eliminating 15 to 20 kW of wasted heat per rack, you address one of the more persistent cost and scaling barriers for AI and GPU data centres.”
Enteligent’s white paper further shows how centraliSed rack conversion in an 800VDC-to-50VDC architecture can provide a unified electrical platform for all data centre workload types, including traditional enterprise and storage servers, GPU compute and next-generation AI infrastructure.
For more information, download the Enteligent white paper: 800VDC-to-50VDC Rack Power Architecture Completing the DC-Native Power Stack for AI-Scale Data.
