Grounding, Bonding, Earthing, Lightning Protection, SPD & EMC

Grounding is the electrical subsystem most likely to be treated as paperwork and most likely to bite. It is invisible when it works, it touches every other discipline (substation civil, power distribution, structured cabling, mechanical, networking), and almost every choice is poured, bonded or buried before commissioning, so revisiting it is expensive. Every one of its choices is a decision with consequences. The regime you pick at design time decides whether a single earth fault trips a 50 MW hall, whether a technician survives touching a frame during a fault, and whether a first ground fault on an ungrounded DC bus is a logged warning or a silent landmine waiting for the second fault.

The AI-factory era changed the stakes in three ways. First, the load is no longer a benign collection of dual-corded servers. It is a phase-coherent population of accelerators whose synchronized power transitions inject common-mode noise and switching transients into every reference plane they touch, and those transitions must be bounded by the amplitude-versus-duration waveform of the rack you are buying, not by its peak rating. Second, the dense-rack generation is migrating off the grounded 415/480 VAC neutral onto a ±400/800 VDC bus, which has no AC neutral. Mt. Diablo’s ±400 V output can carry +400 V, a midpoint (common) conductor and −400 V plus PE. Its two-wire 800 V output carries 800 V and a return that is grounded, except in the two-wire PSU option, which keeps both safety-isolated from PE. Neither has an AC-style residual-current device; the analogues are insulation monitoring and DC ground-fault or leakage detection. Third, the racks became wet: conductive coolant, metal manifolds, CDUs and dripless quick-disconnects are now bondable objects with their own touch-potential and stray-current concerns. The earthing system has to absorb all three at once.

The first and most consequential fork is which system-earthing regime the LV distribution runs. IEC names each regime with a two-letter code. The first letter is the source-to-earth relationship (T = directly earthed, I = isolated or impedance-earthed). The second is the relationship of exposed conductive parts to earth (T = locally earthed, N = bonded to the source neutral). The mainstream data-center answer in IEC regions is TN-S: a single point of earthing at the source, with separate protective-earth (PE) and neutral (N) conductors carried all the way to the load. North America’s equivalent is a solidly grounded wye with a separate equipment-grounding conductor (EGC) per NEC, which is functionally TN-S under another name. The real fork is whether any part of the facility deliberately departs from a solidly earthed source toward IT (isolated or impedance-earthed), and the dense-rack DC bus pushes exactly that question to the front (covered below).

The choice runs through every protective device. A solidly earthed TN-S system clears an earth fault as a high-magnitude short circuit: breakers and fuses see it and trip, and supplementary residual-current devices (RCDs/GFCIs) catch the lower-magnitude faults that overcurrent misses. A TT system (source earthed, loads earthed to a separate local electrode rather than the source neutral) usually cannot rely on overcurrent for earth faults, because the fault loop runs through soil, so it normally uses RCDs. It is common where a facility cannot trust the quality of the utility’s earth. An IT system is the deliberate opposite. The source is isolated or impedance-earthed, so a first earth fault draws only capacitive and leakage current, plus any current through the intentional earthing impedance, and the load keeps running as long as the first-fault protection conditions are still met. An insulation-monitoring device raises an alarm so crews can find and clear the fault before a second one lands on another live conductor (a different phase or the neutral). That second fault is a real short through the fault-return network, and it has to clear automatically. That continuity property is why IT earthing is the conceptual ancestor of the ungrounded DC bus, and why hospitals and process industries have used it for decades.

The first loop cannot meet the 0.20 s target. Minimum fault current is 207 V / 0.25 Ω = about 0.83 kA, below the 1,000 A the device needs. The crossover is 207 V / 1,000 A, which falls just under 0.21 Ω: any worst-case loop above that misses the target. Cutting the loop to 0.20 Ω gives about 1.0 kA, just above the operating current, so the revised conductor and bonding design passes this screen with little margin. Confirm it against the real device curve, or choose a device that clears at the proved minimum fault current.

If a changed supply or a hotter, longer PE path raises the loop back to 0.25 Ω, the design fails again. Do not substitute an earth-electrode resistance target: this TN fault loop returns through the metallic PE path, while TT, isolated DC and HRG each need their own circuit and protection treatment. Commissioning receives the hot-loop calculations, device tolerances and clearing curves, PE thermal duty and bonding continuity tests. The substation GPR, step/touch and transferred-potential study is separate (next section). IEEE 80-2013/Cor1-2015 is published but inactive-reserved, and its scope excludes DC quantitative design.

Before any white space exists, the on-site MV/HV substation (Chapter 4.2) needs a buried ground grid: a mesh of bare copper conductors and driven rods that gives fault current a low-impedance path back to the source and controls the voltages a person can be exposed to during a fault. When a ground fault dumps current into the earth, the whole grid rises in potential relative to remote earth. This is ground-potential rise (GPR), and on a stiff fault it can reach kilovolts. The danger is less the absolute rise than the gradients it creates: the touch voltage between a grounded structure and the earth a person stands on (hand to feet), and the step voltage between one foot and the other across the soil. IEEE Std 80-2013 is the last published AC-substation grounding guide used to bound both against tolerable body-current limits, which derive from fault-clearing time and body weight. IEEE marks it inactive-reserved. Its revision, P80, is an active project whose current PAR was approved in September 2025.

The decisive, irreversible input is soil resistivity. It can vary by orders of magnitude between a wet clay site and a dry rocky or desert one, so measure it before grid design instead of assuming it. The survey contractor works to IEEE 81-2025, which superseded the 2012 edition: run Wenner four-pin traverses in several directions, with probe spacings wide enough to fit a two-layer soil model. High-resistivity soil forces a larger, denser grid, more rods, deeper electrodes, or imported low-resistivity backfill or ground-enhancement material, all to reach the target grid resistance and keep touch and step voltages inside the IEEE 80 limits. Get this wrong and the consequences are poured into concrete. A substation energized over an under-designed grid is a step-and-touch hazard that is enormously expensive to retrofit, and a high grid resistance also degrades fault clearing and lifts GPR onto every bonded metallic path leaving the site: fences, cable shields, pipework, even the structured-cabling bonding network in the halls. This is a power-bound, civil-first decision. The grid goes in with the substation, on the same critical path as the interconnect, and it is one of the least reversible items in the build.

Inside the halls the governing idea is equipotential bonding: tie every exposed conductive surface to a common reference so they rise and fall together during a fault or transient, limiting potential differences for people and sensitive signals. That covers rack frames, busway enclosures, cable trays, containment, raised-floor structure and now liquid-cooling metalwork. Conductor impedance, GPR and transferred potentials still need calculation. The structured-cabling discipline formalizes this as a bonding network: a telecommunications main grounding busbar (TMGB) at the entrance, secondary busbars (TGBs) per space, and bonding conductors tying racks and pathways back to it. In North America the reference is ANSI/TIA-607 (telecommunications bonding and grounding), currently TIA-607-E (2024). The equipment-and-electronics rationale and the high-frequency reference-plane concept come from IEEE 1100 (the Emerald Book).

For high-density, high-frequency halls the upgrade from a radial (single-point) bond to a Signal Reference Grid (SRG) matters. An SRG is a fine copper mesh under or bonded across the floor, with racks tied to it at multiple points. At the frequencies of switching converters and fast signal edges, a long single bonding conductor is an inductor, not a short. A mesh keeps the reference impedance low across a wide band, suppresses common-mode noise, and stops fault or transient energy from finding a path through signal cabling. A single-point bond is cheaper and adequate for low-density legacy IT, but a dense AI hall wants a meshed SRG: its load slams common-mode current into everything, and long management and monitoring copper crosses the floor. Retrofitting one under a live, wet, 132 kW/rack floor is miserable, so it is a design-time call.

This chapter is the canonical home for DC-bus grounding in the guide. The dense-rack generation moves distribution from a grounded AC wye to ±400 VDC or 800 VDC delivered from a sidecar power rack (Chapter 4.7), and the platform you choose fixes the earthing scheme. Mt. Diablo 1.0.0 connects its ±400 V midpoint to protective earth through either high resistance or a solid contact (§8.1.4.1), and a two-wire 800 VDC output can ground its return either way (§8.1.4.2). Only the two-wire PSU option keeps the 800 V conductor and return safety-isolated from protective earth (§8.2.1). Each topology therefore has a different first-fault and protection basis.

The unipolar 800 VDC bus fed from a solid-state transformer (SST) sits outside Mt. Diablo. There is no midpoint to bond, and the SST specification does not allow a solidly grounded return. The OCP Solid State Transformer Specification (Rev 0.3.0, effective 22 June 2026; Google, Microsoft and NVIDIA) defines its output as an IT system: floating or high-resistance grounded, with detection and protection coordination throughout the LVDC distribution. It defers the grounding-impedance and insulation-monitoring details (§5.4) to a later revision. NVIDIA's monopolar 800 VDC power-rack reference design feeds a unipolar bus of the same kind. Because the rack is not an SST, that specification's grounding requirement does not apply to it. SemiAnalysis put the rack at about 660 kW in its May 2026 report, and NVIDIA targets availability in H2 2026. Ask the vendor how the return is grounded as shipped. On a floating or HRG unipolar bus, the first fault references the bus to earth at the fault point, as the deep dive below explains. Until the SST specification fills in §5.4, write the detection into the purchase order: an insulation-monitoring device or ground-fault detector at the source, fault location per feeder, and an acceptance test of both.

OCP Mt. Diablo 1.0.0 (effective October 8, 2026) settles a conflict in v0.7.0, where one clause required ground-fault detection for both midpoint schemes and another called it optional for solid grounding. §8.1.4.2.3 now requires a ground-fault detector for solid and high-resistance midpoint grounding alike, with voltage-balance detection for the high-resistance case and residual-current detection on every output cable to the IT rack. Write that into the purchase specification per output and test it at acceptance, because a unit built to v0.7.0 may not have it. The midpoint bond you select governs first-fault current, continuity, component insulation and output-protection coordination. The isolated two-wire PSU option must keep its own safety isolation and protection basis, not inherit the ±400 V midpoint rules.

Surge-protective devices defend the facility against microsecond-scale overvoltage impulses: lightning-induced surges and utility or capacitor-bank switching events. A GPU load stepping hundreds of kilowatts in milliseconds is not an SPD problem. It causes sags and swells lasting cycles, far below any clamping threshold, and storage, rectifier controls and ride-through settings handle it (Chapter 4.5 and Chapter 4.10). The usual principle is cascaded, coordinated staging. A device built for direct-lightning duty often clamps too high for electronics far downstream, so designers stage separate devices. Combined Type 1+2 or 1+2+3 arresters can instead do both jobs in one stage: they carry the rated 10/350 µs impulse current and still protect the equipment behind them, inside the coordination limits the maker declares. Type 1 (IEC: Class I, tested with the 10/350 µs waveform that mimics a direct lightning current) sits at the service entrance and takes the brunt. Type 2 (Class II, 8/20 µs) sits at downstream distribution boards and handles residual and switching surges. Type 3 (Class III) sits close to sensitive equipment as a final clamp. Choose separate stages or a combined device from the equipment’s impulse withstand, the installed protection level (including connection-lead voltage), conductor distance, surge duty and the manufacturer’s energy-coordination data. This Type 1/2/3 sequence uses IEC test classifications, not UL location definitions, and the project's risk and coordination study picks the actual AC and DC devices and where they go.

Two standard families govern. IEC 61643 is international: IEC 61643-01 carries the common requirements, IEC 61643-11:2025 remains the AC-specific product standard, and IEC 61643-41:2025 is the one to specify for DC low-voltage SPDs on a ±400/800 VDC bus. UL 1449 is the North American product standard. Its 5th Edition, published in 2021 and last revised with ANSI approval on 21 October 2025, is current, and the NEC requires SPDs to be listed. UL 1449 as published covers 50/60 Hz AC circuits to 1,000 V and photovoltaic systems to 1,500 V DC, nothing broader. For a non-PV 800 VDC bus, the SPD’s listing or evaluation must expressly cover that application, voltage, earthing scheme and fault duty. A listing for AC or PV alone does not cover it.

Getting coordination wrong is subtle and common. If upstream and downstream devices are not energy-coordinated, a downstream Type 2/3 device can end up clamping a surge the upstream Type 1 should have absorbed, and it fails. Or the upstream device’s let-through voltage is too high for the downstream device’s rating, and the failure cascades. Coordinate SPDs as a system (let-through voltage, nominal discharge current, short-circuit current rating) with the right disconnect and fusing. They must also be maintainable. MOV-based devices can degrade under repeated impulses, sustained overvoltage and heat, endurance varies with the suppression technology and the applied duty, and some spark-gap designs are reported to take repeated impulses without measurable ageing. So choose by the product’s published endurance, and keep status monitoring, isolation and replacement provisions and the manufacturer’s replacement criteria, because a facility that cannot take a hall down will not be able to swap a part in a hurry. SPDs are also a primary reason the lightning-protection and bonding systems must be unified: a strike without a low-impedance bonded path will find one through your SPDs and your signal cabling.

The lightning-protection system (LPS) on the building envelope is air terminals, down-conductors and a low-impedance connection to the earthing system. IEC 62305 governs it internationally, and NFPA 780 in North America. Its job is to intercept a strike, conduct it to earth, and equipotentialize the structure so the energy does not pass through the building and its electronics. Chapter 6.3 details the physical coordination: where air terminals sit, how down-conductors route around a liquid-cooled roof plant, and how the LPS earth ties into the substation grid. The electrical obligation here is a coordinated equipotential network. The LPS earth and in-hall bonds, the substation grid, remote earths and incoming metal are connected or isolated according to the GPR and transferred-potential study. An accidental separation can force lightning current through signal cabling and the electronics at each end, and an indiscriminate metallic tie can import a remote fault potential. Draw the intended bonds and isolation boundaries before pouring the grid.

EMC ties these together. Know which media are exposed. Optical links are galvanically immune, and short high-speed copper SerDes links (NVLink cables, DACs and AECs inside a rack or to its neighbor) ride the rack bond. The exposure is long copper (out-of-band management Ethernet, RS-485 and Modbus runs to the EPMS and BMS, telecom pairs) and metallic armor or strength members that cross bonding zones. For those, common-mode noise control decides reliability, and at low frequency the grounding and bonding system is the EMC system. Three mechanisms matter: a low-impedance meshed reference (the SRG above) that gives common-mode currents somewhere to go; correct shield bonding of STP copper and fiber armor, bonded at the right points to drain induced currents without creating ground loops; and segregation of noisy power runs from sensitive signal pathways. This practice lands here, not in the network chapters. Metallic armor and STP shields are bonded to the SRG wherever they cross a bonding-zone boundary: at the building entry, at each equipment-room boundary and at the equipment interface. Any deliberately floated end is documented as a decision, not left to the installer. EMC performance is largely determined by the bonding topology poured into the floor, and filters added at the end cannot recover it. That is why the SRG, the LPS and the SPD staging have to be designed as one system.

Grounding does not decompose into independent subsystems you can hand to separate trades. The substation grid sets the GPR that the in-hall bonding network must survive. The bonding network is the EMC reference every copper signal path rides on. SPD staging works only with a low-impedance bonded path to the same earth the LPS uses. The DC-bus monitoring philosophy makes sense only against the AC earthing regime it descends from. The forks compound:

• Solidly earthed vs isolated/IT/HRG: first-fault current, clearing and possible continuity are set per circuit, with separate solid-midpoint, HRG and isolated DC studies.
• Single-point bond vs meshed SRG: single-point is cheaper and fine for low-density legacy IT. A meshed reference lowers high-frequency impedance in a dense, wet hall; fit one where the EMC and bonding study calls for it. Retrofitting one later is costly.
• Unified earth vs split/clean earths: earths that are separate and uncoordinated can develop transient potential differences, which show up as EMC failures or lightning damage. Unify and equipotentialize; segregate by topology, not by separate electrodes.
• SPD coordination as a system vs device-by-device: energy-coordinated Type 1/2/3 staging with monitored, replaceable modules, against an uncoordinated cascade that fails on the surge it was bought to stop.

Each is a density-ramp-sensitive decision. The substrate (buried grid, floor SRG, bonded cooling network, service-entrance SPD provisioning) is poured or bonded early and is painful to revisit, so it must accommodate the ramp to 600 kW-class racks and an 800 VDC bus even when today’s hall is lighter. Reserve the reference-plane and earthing headroom you cannot retrofit, and defer only the device-level spend you can swap later.