Powering AI Racks Without Pulling More Copper
The AI Rack Power Problem Is a Copper Problem
Modern AI compute racks are not subtle. Where a conventional server rack might draw 5–10 kW, dense GPU clusters routinely demand 30, 50, or even 100 kW per cabinet. That appetite for power forces a reckoning inside the data center: traditional branch-circuit copper—governed by NEC/NFPA 70 Chapter 3 wiring methods—must be sized, conduit-routed, and protected in ways that grow expensive and space-consuming at high current levels. When you need to add ten new AI racks in an existing facility, the honest answer from the electrical contractor is often "we need to pull a lot more copper, and you may need to upgrade your switchgear."
There is a different answer emerging, and it lives in a part of the NEC most infrastructure teams have never read: Article 726.
What Is Fault-Managed Power (Class 4)?
The 2023 edition of NEC/NFPA 70 introduced Article 726, establishing a new circuit class distinct from the familiar Class 1, 2, and 3 circuits defined in Article 725. Article 726 governs Fault-Managed Power Systems (FMP)—sometimes marketed under terms like Digital Electricity (DE), Packet Energy Transfer, or Pulsed Power. The defining principle is elegant: rather than sending continuous high-voltage AC or DC current down a conductor, an FMP system transmits discrete, monitored energy packets. If the system detects a fault—an open conductor, a short, or unauthorized contact—it shuts off energy delivery in milliseconds, rendering the cable touch-safe even at elevated voltages.
Because of this real-time fault detection, Article 726 allows significantly relaxed wiring methods. In most cases, conduit is not required, which is precisely where the installation economics become compelling. Cable used in FMP systems must be listed to UL 1400-2, and the transmitting/receiving equipment must be listed to UL 1400-1. The cable itself resembles data-type cable in construction, which opens the door to pathways and spaces already designed to ANSI/TIA-569 standards—cable trays, ladder racks, and conduit bundles that data-center designers already know how to detail.
Real Capacity Over Real Distances
Heather Technologies partners with VoltServer, whose Digital Electricity (DE) platform is one of the more mature commercial implementations of FMP principles. VoltServer's representative specifications indicate delivery of up to approximately 450 V and approximately 2,000 W per channel over distances approaching 2 km on data-type cable. [FLAG: VoltServer DE voltage/power/distance figures—verify against current VoltServer datasheet before publication.]
For data-center applications specifically, Heather Technologies also partners with DCPacket and their Titan Platform, which was formally announced in partnership with VoltServer in December 2025. The Titan Platform targets high-density data-center FMP distribution, addressing exactly the AI rack power density challenge this article describes. [FLAG: DCPacket Titan Platform specifications—verify product availability, power ratings, and deployment parameters against current DCPacket documentation.]
How This Fits a Data-Center Infrastructure Framework
Data-center infrastructure planning does not happen in isolation. ANSI/TIA-942 provides the framework for data-center design including redundancy ratings, and Uptime Institute Tier classifications (Tier III being concurrently maintainable) set the reliability expectations that operators must meet. FMP distribution fits within this framework as a power-delivery layer that can be architected with the same redundancy logic as any other branch-circuit system—because the fault-shutoff behavior of Article 726 systems actually strengthens the safety argument for deploying them in high-availability environments.
Thermal management remains a parallel concern. ASHRAE TC 9.9 guidelines recommend IT equipment inlet temperatures in the range of 18–27°C for standard operating envelopes, and high-density AI racks push against those limits aggressively. The good news is that FMP architecture does not inherently worsen thermal load—power delivered is power delivered—but the ability to distribute power sources closer to the rack, without massive conduit runs generating their own heat and consuming airflow paths, can actually simplify the hot-aisle/cold-aisle geometry that keeps PUE (total facility power divided by IT power) in check.
Copper and Fiber Still Have Their Roles
Adopting Article 726 FMP for power delivery does not eliminate structured cabling—it frees it to do what it does best. ANSI/TIA-568.2-D governs balanced twisted-pair copper cabling, including Cat 6A and Cat 8. Cat 8 operates to 2 GHz and supports 25GBASE-T and 40GBASE-T at distances up to approximately 30 meters—exactly right for top-of-rack to end-of-row switching. When you are not asking copper infrastructure to carry kilowatts of power, you can right-size it for data.
High-bandwidth spine and inter-row connectivity belongs to optical fiber, governed by ANSI/TIA-568.3-D. For AI cluster interconnect, OM4 or OM5 laser-optimized 50/125 µm multimode fiber with MPO/MTP connectivity supports the high-lane-count parallel optics that GPU fabric demands. Singlemode OS2 fiber—conforming to ITU-T G.652—handles longer runs between data halls or to upstream network aggregation. Keeping these cabling systems documented to ANSI/TIA-606 (administration and labeling) and properly bonded per ANSI/TIA-607 (bonding and grounding, including TMGB/TGB topology) ensures that the overall infrastructure remains maintainable as the power architecture evolves.
The Practical Case for Acting Now
AI infrastructure buildouts are not slowing. The facilities teams being asked to support them are working with electrical plants that were designed for a different era. Pulling new heavy-gauge copper through congested pathways, upgrading panels, and re-routing conduit through occupied data halls carries real cost and real schedule risk. Article 726 Fault-Managed Power systems—listed to UL 1400-1 and UL 1400-2, deployed over UL 1400-2 cable in TIA-569-compliant pathways—represent a code-compliant, safety-validated alternative that deserves a place in every AI rack power design conversation.
At Heather Technologies, we work with engineers and data-center operators who are making these decisions today. If your next AI rack deployment is bumping against the limits of your existing copper electrical infrastructure, let's talk about what FMP can do for your facility.
About the author — Todd Taskerud, AWS CCP, RCDD/NTS/OSP/WD, LEED GA, is a BICSI-credentialed communications distribution designer at Heather Technologies, specializing in fiber, copper, and data-center network infrastructure.
