800 VDC AI Data Centres: What Changes for Commissioning?

The Current Public Position: Emerging, Not Universal

NVIDIA publicly describes a gradual evolution from today’s AC distribution toward an 800 VDC architecture for future AI factories. Its published material presents the concept as a way to reduce conversion stages and distribute higher-voltage DC closer to very high-density compute. Power-system OEMs and semiconductor manufacturers have also announced development work around conversion, distribution, protection, sensing and switching.

Those announcements establish a credible industry direction, not a universal installed base. Current AI data centres use a range of AC distribution voltages, rack power arrangements, backup architectures and internal DC rails. Near-term projects may retain conventional facility AC while introducing high-voltage DC only within a sidecar, pod or rack-level subsystem.

For commissioning teams, the correct starting point is the selected OEM and project architecture. Public roadmaps are useful for planning, but the issued design, interface control documents and approved test plan determine what must actually be energized and tested.

Why the Industry Is Looking at Higher-Voltage DC

The basic engineering motivation starts with P = V × I. For the same power, increasing voltage reduces current. Lower current can reduce the conductor cross-section and parallel runs required, lower I²R losses, and make busway, distribution and connection systems more compact. At very high rack power, those physical benefits can be substantial.

Higher-voltage DC may also allow the power chain to be reorganized. Instead of repeated AC-to-AC and AC-to-DC conversion stages, a design may centralize conversion and distribute DC closer to the rack, where another conversion stage supplies the low-voltage rails used by compute components.

This is not a free efficiency gain. Protection, isolation, grounding, fault energy, switching, connectors, insulation coordination, measurement, maintenance and worker safety all require an end-to-end DC design. The absence of a natural current zero crossing makes DC interruption a different problem from AC interruption. Every component and temporary interface must be rated and tested for its actual duty.

How Today’s AI Rack Power Architecture Works

Many current facilities distribute three-phase AC through transformers, switchgear, UPS systems, busway and rack-level equipment. Conversion to DC occurs in or near the IT equipment, and additional conversion stages produce the lower-voltage rails required by processors, memory, networking and controls. Exact arrangements vary by platform and facility.

This architecture benefits from a mature ecosystem of AC-rated equipment, protection practices, temporary load banks and field procedures. At higher rack densities, however, the current, copper, conversion equipment and physical space required close to the rack can become increasingly difficult to manage.

An emerging design may keep much of the upstream AC plant while changing the final distribution segment. Another may convert medium-voltage AC to high-voltage DC at a more centralized point. The commissioning strategy needs to distinguish which parts of the power chain remain conventional and where the new DC boundary begins.

What an 800 VDC Architecture Could Change

Moving the DC boundary upstream can change the equipment and interfaces used between the facility source and the rack. Conversion systems, DC busway, distribution boards, protection devices, disconnects, connectors, pre-charge functions, sensing and control may all become part of the critical path.

Protection deserves particular attention. The project must understand prospective fault behaviour, available fault energy, detection, interruption, isolation and safe discharge. Switching and connectors must be DC-rated for the operating voltage and current. Measurement systems must capture steady-state and transient behaviour without introducing an unsafe temporary interface.

Backup power may also be integrated differently. Energy storage could connect to a high-voltage DC bus rather than sitting only within a conventional AC UPS arrangement, but topology and operating philosophy will vary. Commissioning cannot assume that familiar AC sequences map directly to a new DC architecture.

The Commissioning Challenge

Temporary commissioning equipment must evolve with the production architecture. An AC load bank connected upstream can still prove facility generation and distribution, but it cannot by itself validate a new 800 VDC path, its protection, or the rack-side interface.

A future high-voltage DC temporary load may need controlled load steps, project-specific connection hardware, pre-charge or inrush management, DC-rated isolation and switching, fault containment, high-speed measurement, safe discharge and clear grounding arrangements. The way several loads are grouped across a pod or hall may also differ from today’s AC practice.

The test objective matters. Applying steady DC load is one task; validating protection and fault response is another and may require purpose-built test equipment, simulation or OEM procedures. Thermal emulation adds a further layer because electrical loading must remain coordinated with the liquid-cooling condition at very high rack density.

AMOR and its technology partners are tracking and developing test approaches for next-generation high-voltage DC AI infrastructure. AMOR does not currently publish an 800 VDC production load-bank rating or certification on these pages; any future capability will be documented against the exact equipment and project requirement before it is offered.

800 VDC Does Not Eliminate Facility-Level Commissioning

Even if the final rack-side distribution changes, the facility still has an upstream electrical and mechanical system to prove. Utility interfaces, generators, transformers, switchgear, controls and heat-rejection infrastructure remain essential. UPS and energy-storage functions may be reconfigured rather than removed, depending on the selected architecture.

MW-scale LV/MV load banks therefore remain relevant. They can establish demand on the facility source and backup systems while a separate purpose-built setup addresses the high-voltage DC segment. The load plan should show clearly which equipment proves each boundary and where measurement is taken.

Integrated testing must also account for transitions between those boundaries. A change in source, storage mode or distribution state may affect both the upstream AC system and the downstream DC bus. The owner and CxA will need a coordinated procedure based on the actual control and protection philosophy.

Electrical and Thermal Commissioning Become Even More Interdependent

As rack power increases, nearly all of that electrical energy ultimately becomes heat that the cooling infrastructure must remove. Higher-voltage distribution may improve the electrical delivery path, but it does not reduce the need to validate CDU capacity, secondary-loop flow, pressure, temperature, controls and facility heat rejection.

Extremely high-density racks compress the time and space available for systems to respond. A power transition can change thermal output, pump operation, control states and backup requirements. Conversely, a cooling transition may require the electrical load to remain stable long enough for the project team to observe recovery.

Commissioning equipment should therefore be planned as a coordinated electrical-and-thermal platform. Rack emulators reproduce local demand and heat. High-capacity liquid load banks exercise CDUs and cooling zones. Facility load banks prove the wider electrical source and distribution. Future DC-specific equipment will have to connect those layers without creating an unverified test boundary.

What Owners and CxAs Should Start Asking Now

Projects considering next-generation AI platforms should make the temporary-load strategy part of early design coordination. Waiting until late commissioning can leave the team without a safe connection point, suitable protection, enough equipment capacity or a workable method to coordinate power and cooling load.

The following questions are not an official standard. They are a practical starting point for design and commissioning conversations while architectures continue to mature.

• What rack electrical architecture and maximum rack power are being specified?
• Is the rack interface AC or DC, and at what nominal voltage and operating range?
• Where does conversion to high-voltage DC occur, and who owns that equipment boundary?
• What redundancy and backup-power architecture applies across the AC and DC paths?
• What temporary load connection, isolation and protection interface will be available?
• How will load steps, safe discharge and measurement be controlled?
• How will rack electrical load and liquid-cooling load be coordinated?
• Which commissioning stages require full design load, and which can use partial or simulated conditions?

Preparing the Commissioning Strategy for the Next Generation

The most useful action today is not to assume that every project will become 800 VDC. It is to preserve interface clarity and testability as designs evolve. Owners, CxAs and engineers should require an explicit commissioning boundary, temporary-load concept, protection philosophy and coordinated thermal plan for the selected architecture.

Test equipment needs to evolve as quickly as compute infrastructure, but published capability must remain precise. Current AC rack emulators, liquid load banks and MW-scale load banks already address important parts of the facility. High-voltage DC introduces an additional equipment and safety problem that should be engineered against documented OEM and project requirements—not marketed ahead of verification.