Evaluate the complete system.
Higher rack density is changing how electrical power reaches compute. NVIDIA describes a progression from existing AC facilities through hybrid rack-level conversion to more integrated 800 V DC architectures. The physical arrangement and deployment path differ between those approaches.
Schneider Electric’s evaluation principles similarly emphasise architecture, ecosystem readiness, protection and grounding, energy storage integration and operation. A voltage target alone does not resolve these decisions.
Start with the compute interface.
Identify the intended hardware generation, input requirements and deployment phases. Establish what is compatible today and what depends on future equipment. Map the electrical boundaries between the facility, power conversion, distribution and compute.
Define protection and continuity.
Ask how isolation, earthing, fault interruption, redundancy and maintenance will work in the proposed arrangement. Map storage and backup functions to the actual architecture. A battery connected to a system does not, by itself, define its continuity performance.
Qualify the complete chain.
Request an interface schedule and evidence for the selected equipment. Agree the integration owner, test procedures and acceptance boundaries. Separate supplier capability, tested configuration and complete-site performance.
Compare a credible alternative.
Assess the proposed DC arrangement against the conventional option for this site. Include space, conversion stages, losses, maintainability, expansion and operating responsibilities. Quantified benefits should follow from the project comparison.
Technical evaluation only; this is not an engineering design or a standard product offer. Suitability, supply and acceptance require project assessment.
- NVIDIA — Why scaling AI compute requires a new power architecture ↗
- Schneider Electric — Five principles for 800 VDC in AI data centres ↗
Sources checked 29 September 2026. References do not imply partnership or endorsement.
