Schneider Electric publishes 800 VDC arc-flash study for AI data centers

Schneider Electric has published a new study examining arc-flash risks in the 800 VDC power architectures emerging in AI data centers. Conducted using ETAP software and digital twins, the analysis indicates that appropriate system architecture and protection strategies can reduce risks to levels comparable with those found in conventional AC distribution systems.

The rapid expansion of artificial intelligence is increasing both power demand and rack density in data centers, prompting the industry to explore new electrical distribution solutions. Schneider Electric has responded by publishing a study on how arc-flash risks can be assessed and managed in 800 VDC power architectures, which are considered an important part of this transformation.

Based on design approaches adopted by some of the world’s leading hyperscale data center operators, the study compares two emerging 800 VDC architectures with different configurations. Its findings show that the effects of an arc-flash event can vary significantly depending on system architecture, capacitor placement, reverse-current control and fault-clearing time.

According to the research, arc-flash risks in 800 VDC systems remain manageable even under demanding scenarios dominated by capacitor discharge. In many cases, the level of risk is comparable with that of AC power systems commonly used in data centers.

Why is 800 VDC gaining attention?

As the power requirements of processors and accelerators used in AI applications increase, conventional electrical distribution infrastructures are finding it more difficult to supply high-density racks efficiently. The industry is therefore turning to new power architectures capable of reducing energy losses and supporting megawatt-scale racks.

Led by NVIDIA and supported by energy technology companies including Schneider Electric, the transition to 800 VDC data center infrastructure is expected to enable large-scale data centers and so-called AI factories to operate IT racks rated at 400 kW and above.

Higher operating voltages, however, require a more detailed understanding of fault behaviour, protection coordination, equipment design and safe maintenance practices. While arc-flash analysis is already standard practice in AC data centers, the industry has yet to establish a broadly accepted standard or guideline for managing the electrical hazards faced by personnel working with converter-fed 800 VDC systems.

Manish Kumar, Executive Vice President of Schneider Electric’s Secure Power and Data Center Business, said 800 VDC distribution represented a major shift in data center design and introduced new safety considerations that needed to be examined thoroughly.

“800 VDC power distribution represents a significant change in data center design, but it also introduces new safety considerations that need to be studied comprehensively. Our work with some of the world’s leading hyperscale data center operators provides engineers and safety professionals with one of the first practical frameworks for assessing arc-flash risks.”

Kumar added that the approach provides a structured method for understanding fault behaviour, establishing safe working practices and designing effective protection systems.

Rack- and facility-level architectures examined

Schneider Electric’s research evaluates two separate 800 VDC configurations representing the main approaches emerging across the industry: rack-level and facility-level architectures. The analysis combines standards-based methods, transient simulations and system-level modelling.

At rack level, the study examined a “sidecar” configuration, also known as a power rack. Under conservative assumptions, the case study found that even without a protection device, incident energy remained well below the reference personal protective equipment threshold of 1.2 cal/cm².

The facility-level case study focused on a centralised 800 VDC architecture. In a hypothetical system without overcurrent protection, incident energy was found to be somewhat higher than in rack-level configurations.

The analysis also examined faults occurring both upstream and downstream of reverse-blocking diodes, assessing their effects on backfeed, peak current and arc-flash outcomes. When the duration of the fault contribution was limited through standard protection devices, incident energy fell to levels considered appropriate for working environments and became broadly comparable with levels found in conventional AC architectures.

Capacitor discharge is decisive in the first milliseconds

One of the study’s key findings is that an arc-flash event in an 800 VDC system is time-dependent rather than static. Capacitor discharge plays a particularly important role in determining current and incident energy during the first milliseconds of a fault.

Assessments based solely on standardised and simplified calculation methods may therefore overestimate the risk, especially in systems where capacitor discharge is dominant. Transient simulations can provide more realistic results by accounting for changes in fault current throughout the different stages of an event.

Modelling conducted with ETAP’s advanced power-system analysis software and digital twins allows variables such as system topology, converter response, switching logic and protection coordination to be assessed together. Operators can consequently develop protection strategies using more detailed and representative data.

Tanuj Khandelwal, CEO of ETAP, said industry standards remained essential for electrical and arc-flash safety but that traditional methods could produce overly conservative results because they did not always fully represent the behaviour of complex DC systems.

“To understand the actual risk, engineers need to assess system topology, fault behaviour, protection coordination, converter responses, switching logic and active protection systems together,” Khandelwal said.

Design choices determine safety performance

According to the study, safety in 800 VDC infrastructure is not determined solely by the voltage level or the use of DC distribution. Capacitor placement, reverse-current blocking components, system topology and protection equipment operating within milliseconds all have a direct effect on arc-flash outcomes.

Proper selection and coordination of standard protection devices can keep incident energy below critical thresholds. The research indicates that when these design and protection measures are implemented, overall risk remains low and can, in many scenarios, be reduced to levels comparable with conventional AC distribution systems.

Alongside the research, Schneider Electric has also been testing “live swap” technologies designed to improve maintenance safety in 800 VDC systems. The technology aims to allow designated power components to be replaced safely while the system remains in operation.

The complete findings are available in Schneider Electric’s technical paper, “DC Arc Flash Analysis: A Practical Study on 800 VDC in Data Centers.”