A direct comparison of conventional AC power distribution and emerging 800 VDC architecture for AI data centers — conversion stages, efficiency, equipment maturity, and the practical reasons most facilities today still use AC even as 800 VDC gains momentum.
Two Architectures, Same Destination
Both AC and 800 VDC architectures exist to solve the same problem — getting power from the utility connection to the DC voltage a server's internal components actually use — but they take different paths to get there, with real tradeoffs in efficiency, equipment maturity, and design complexity. Understanding the comparison matters because most facilities being designed and built today still use conventional AC distribution, even as 800 VDC gains real momentum for next-generation designs, and the reasons for that gap are practical, not just inertia.
Conversion Stages: Where the Difference Actually Shows Up
Conventional AC architecture, described in AI Data Center Power Distribution Explained, moves power through several stages of AC distribution and voltage transformation, with the UPS itself performing an internal AC-to-DC-to-AC conversion (see Data Center UPS Systems Explained), and the server's own power supply performing one final AC-to-DC conversion at the rack. An 800 VDC architecture, covered in What Is 800 VDC in AI Data Centers?, removes one or more of these conversion stages by distributing power as DC much further into the facility — fewer conversions generally means lower cumulative conversion losses and less waste heat from the power system itself, which matters more as GPU rack power density climbs (see How Much Power Does an AI GPU Rack Use?).
Why Most Facilities Still Use AC Today
Despite 800 VDC's efficiency advantage on paper, conventional AC architecture remains the practical default for most new data center construction right now, for reasons that have nothing to do with the underlying physics. AC switchgear, protective devices, and UPS systems represent decades of industry standardization, widespread equipment availability, and well-understood safety and protection practices. DC fault behavior is genuinely different from AC fault behavior in ways that affect protective device design and safety practices, and the supply chain for DC-rated equipment at data center scale is still maturing relative to the deeply established AC equipment ecosystem. For an operator building a facility today, AC architecture carries lower execution risk even if it carries somewhat higher long-run conversion losses.
Where the Calculus Shifts Toward DC
The case for 800 VDC gets stronger specifically at the extreme power densities modern AI GPU deployments demand, where even small percentage efficiency gains translate into meaningful absolute power and cooling savings at facility scale, and where a facility is being designed from scratch for a long operating life during which those savings compound. This is why 800 VDC discussion concentrates specifically around new, purpose-built AI data centers rather than general-purpose or retrofit projects — the case for taking on the DC transition's execution risk is strongest where the efficiency payoff is largest and where the facility isn't constrained by existing AC infrastructure already in place.
Not a Universal Answer Either Way
Neither architecture is unconditionally "better" — AC remains the lower-risk, better-understood choice for the majority of facilities being built today, while 800 VDC represents a genuine efficiency opportunity that's likely to see growing adoption specifically in the highest-density, purpose-built AI facilities where its advantages matter most and where operators have the scale and technical resources to manage the transition's execution risk. The practical answer to "which one" depends heavily on a specific facility's power density, expected operating life, and how much execution risk an operator is willing to accept in exchange for lower long-run conversion losses.