The ripple effect from ever expanding rack densities is moving into a new layer of data centers, forcing the industry to once again reevaluate historic norms and create new standards.
The unprecedented pace with which AI workloads are growing is quickly pushing infrastructure toward megawatt-scale racks to support accelerated computing. Traditional AC distribution systems and lower-voltage architectures cannot keep up with these extreme, concentrated power demands. As a result, the industry is actively aligning around 800 VDC architectures as a highly practical and effective path forward for next-generation AI data centers.
As the industry races toward 800 VDC architectures, safety engineering and arc flash risk mitigation must evolve just as quickly. The transition to higher voltage systems can unlock enormous token production efficiency gains, but deployment will require a clear, comprehensive understanding of electrical safety.
Addressing safety in 800 VDC systems
With higher DC voltage combined with the use of active power electronic converter, electrical fault behavior inherently becomes far more complex. Transitioning to these systems requires the industry to develop a much deeper, more rigorous understanding of fault current behavior, protection coordination, equipment design, and safe work practices for all operating personnel.
Arc flash events exist for all forms of power distribution, and they always carry the risk of potentially exposing personnel to devastating injuries in addition to severe equipment damage due to plasma energy and intense pressure waves. If we are to successfully capture the efficiency benefits of 800 VDC, we must ensure advanced safety protocols.
The gap in existing arc flash standards
While arc flash analysis is already a standard, mature practice in traditional AC data centers, no industry-wide standard or guidance currently exists to manage the specific electrical hazards that personnel may encounter in converter-fed 800 VDC systems. Existing arc flash analysis methods were primarily developed for traditional AC electrical systems and conventional, battery-based DC systems.
Consequently, these legacy calculation methods do not fully reflect the converter-fed 800 VDC architectures emerging in AI data centers, and current standards have been shown to overestimate actual arc flash risk. Traditional calculations often assume steady-state fault currents. However, in 800 VDC systems, capacitor discharge dominates the earliest milliseconds of an event, making the fault current behavior highly transient and time-dependent. Converter-fed systems behave fundamentally differently than traditional battery-fed DC systems and, as a result, conventional analysis methods may overestimate or inaccurately represent the actual arc flash risk. These new architectures demand entirely new safety methodologies.
Driving industry collaboration
Bridging this gap requires active, cross-functional cooperation across the entire technology ecosystem.
“Operating at 800 VDC voltages introduces new safety considerations that must be addressed through industry-aligned standards and procedures,” said Bin Lu, Executive Vice President of the Power Products Division at Schneider Electric. “Clear operational guidelines, including maintenance practices, arc flash mitigation, grounding strategies, and training, will ensure safe handling of 800 VDC infrastructure. Industry working groups and regulatory bodies must collaborate to update or expand applicable codes and best practices.”
Promising work is already underway within key industry organizations. The Institute of Electrical and Electronics Engineers (IEEE) is focused on ongoing work around DC power systems and refining arc flash methodologies, while the Open Compute Project (OCP) is helping to standardize next-generation data center architectures through active discussions around 800 VDC power distribution standards.
Currently, operators rely on a patchwork of standards that are not fully optimized for these next-generation designs. For instance, NFPA 70E successfully addresses electrical safety in the workplace, supported by IEEE 1584 which covers arc flash analysis for AC system. Additional analytical approaches are available to evaluate to DC arc flash behavior in Annex D of NFPA 70E, particularly in systems with time-dependent fault characteristics. These include the Stokes and Oppenlander method as well as Paukert-based formulations. Annex D of NFPA 70E is informative, it means that these specific approaches are not yet part of a formalized, standard calculation framework, which unfortunately leads to high variability in how they are applied in practice.
Applying any of these alternative analytical approaches requires detailed, granular knowledge of specific system parameters and fault behavior, including unique arc characteristics and complex system responses. Because these variables cannot be easily captured through simplified calculations alone, advanced simulation and analytical tools are typically required to accurately model true system behavior.
How Schneider Electric is developing more thorough approaches to 800 VDC arc flash evaluation
Industry standards are evolving, but operators need practical guidance now. Some of the questions the industry seeks to answer include: Is it possible to conduct an arc flash analysis today for an 800 VDC AI data center? What are the requirements? Can existing methods be used? How accurate are they? What are the challenges?
To help answer these questions, Schneider Electric conducted a first-of-its-kind analysis evaluating arc flash risk in two emerging 800 VDC architectures: rack-level architecture and centralized architecture. Simulation scenarios were based on typical hyperscaler deployment designs and used the methods defined in NFPA 70E and supporting IEEE guidance as a foundation for our work.
Our key findings from this evaluation demonstrate the following:
- Arc flash outcomes depend strongly on architecture, capacitor placement, and fault-clearing behavior.
- Transient behavior matters: In 800 VDC systems, arc flash is driven by time-dependent fault currents, with capacitor discharge dominating the first milliseconds of an event. Even under the most demanding capacitor-dominated assumptions, arc flash risk in 800 VDC systems can be managed, and in many cases, is comparable to or lower than typical AC systems.
- Better modeling using advanced software and digital twins can help accurately portray arc flash risk. Software (such as transient simulation and power system analysis tools) shows that simplified arc flash analysis methods for DC systems often overestimate arc flash risk in capacitor-dominated systems. Operators can significantly reduce arc flash risk by informing their protection strategy with advanced software and digital twins like ETAP.
Furthermore, our analysis reinforces that arc flash outcomes depend on architecture and system configuration, not on DC distribution alone, where capacitor placement, reverse-blocking devices, and millisecond-scale protection are key levers for safe 800 VDC deployments. This becomes evident when looking at the specific case study data for each architecture:
- Rack-level 800 VDC architectures: In a sidecar case study using conservative methods and assumptions, results showed incident energy well below the referenced 1.2 cal/cm² PPE threshold.
- Centralized 800 VDC architectures: The facility-level case study shows slightly higher incident energy potential than rack-level designs. The analysis also assesses the effect of system topology and demonstrates how fault locations both upstream and downstream of reverse-blocking diodes influence back-feed, peak current, and arc flash outcomes. When fault contribution is time-limited with standard protection devices, arc flash incident energy is reduced to appropriate levels for the work environments and generally aligned with common AC architectures.
Moving forward safely
The transition to 800 VDC is an impressive leap forward for AI data center efficiency, and the industry doesn’t have to compromise on safety to achieve it. With the right modeling tools, appropriate architectural choices, and rapid protection coordination, operators can confidently mitigate arc flash risks.
Read the full white paper DC Arc Flash Analysis: A Practical Study on 800 VDC AI Data Centers now to learn more.
Add a comment