01The AC UPS architecture and its inefficiency

The traditional double-conversion online UPS is a workhorse of the AC data centre. Utility AC comes in, a rectifier converts it to DC, the DC bus feeds an inverter that produces conditioned AC out, and a battery attaches to the DC bus for backup. The topology is well-understood, well-standardised, and well-maintained by a mature vendor set (Vertiv, Eaton, Schneider Electric, ABB, Riello, Piller, Delta).

The inefficiency is structural: in the AC-in / AC-out chain, the load ultimately consumes DC (every server has an internal PSU converting AC to the 12 V or 48 V the components use). So the round trip runs AC → DC → AC → DC, with conversion losses at every stage. Total end-to-end efficiency for a realistic modern AC UPS + server PSU chain can land in the low-90 percent range under representative operating conditions. Modern UPS efficiencies vary materially with topology and load; Vertiv currently specifies Liebert EXL S1 at up to 97% double-conversion and up to 99% Dynamic Online. Each conversion event contributes some loss and adds capex, weight, and thermal load.

02The native 800 VDC UPS: what changes

A native 800 VDC UPS collapses the topology substantially. Utility AC comes in through an AC-DC rectifier that produces 800 V DC directly. The DC bus feeds the racks at 800 V (via DC distribution (see Part 3/13 on the AI power chain architecture map) on the AI power chain architecture map). A battery stack does not simply attach to a regulated 800 VDC bus. Cell voltage varies with SOC, temperature, chemistry and current, so battery-to-bus coupling is a bidirectional power-conversion problem. Where that conversion lands determines the architecture (Figure 6.3). The rack-level PSU steps down from 800 V to 48 V or lower for the actual compute silicon.

The efficiency gain is the headline result. The topology-simplification gain is often larger in dollar terms: fewer components mean less capex, less footprint, less thermal load, fewer failure points. Topology simplification comes at the cost of BMS complexity and battery-chemistry constraints, which the following sections cover.

03BMS complexity increases sharply

A lithium-ion battery cell has a nominal voltage of roughly 3.6 V (NMC) or 3.2 V (LFP). To reach an 800 V nominal DC bus, you need a series stack of roughly 220-250 cells for common 3.2-3.6 V cell chemistries, depending on pack voltage window and nominal bus definition. Each cell has to be monitored for voltage, temperature, and state-of-charge. Cell-to-cell balancing has to be maintained during charge and discharge. A single cell drifting out of tolerance can propagate through the string.

The BMS complexity comes from monitoring channels, isolation barriers, communications, contactor/precharge control, insulation monitoring, thermal sensing, fault management and functional-safety requirements — not linearly from cell count. A 48 V stack typically involves far fewer series cells and a materially simpler monitoring and isolation architecture. An 800 V stack typically involves roughly 220-250 series cells (for common LFP/NMC chemistries, subject to pack voltage window) and its BMS is a distributed sensor network with communication overhead and safety-certified isolation between cell-monitoring domains, a materially more complex BMS architecture.

04Battery chemistry choice at 800 VDC

At stationary-storage scale, LFP has become a major reference chemistry because of its cost and cycle-life characteristics, while NMC remains relevant where energy density or footprint matters. NMC and other chemistries remain relevant where energy density, footprint or specific duty cycles favour them; at 800 V, pack-level isolation, protection and thermal management become more consequential, making chemistry selection inseparable from the system safety architecture. Three considerations move the balance.

Safety. Higher-voltage series stacks accumulate more stored energy per module, so stationary-storage economics and safety requirements are already favouring LFP; at 800 V, pack-level isolation, protection and thermal management become more consequential, making chemistry selection inseparable from the system safety architecture.

Cycle life. Data-centre UPS applications draw shallow discharge cycles frequently (every utility disturbance) plus rare deep discharge cycles. LFP's generally strong cycle-life characteristics can favour it for that duty profile, subject to cell-specific degradation data and the operating window. LFP generally offers a cycle-life advantage in the shallow / high-cycle stationary-storage duty profile assumed here, although the actual comparison is cell- and operating-window-specific.

Second-life battery integration. EV batteries retired from vehicle use (state-of-health thresholds vary by application and diagnostic assessment) are an emerging supply channel for stationary storage, but their relevance to hyperscale UPS remains constrained by SOH uncertainty, warranty, certification, BMS integration and insurance. The chemistry mix in retired EV packs skews heavily toward NMC in Western markets and LFP in Chinese markets. At 800 V DC bus, second-life battery integration can require additional cell-monitoring and diagnostic headroom, which can change the BMS specification.

05Hybrid supercapacitor role: bridging the 0.1-10 second gap

Batteries can respond to sub-second events, but repeated high-power transients can make them an inefficient way to provide short-duration peak power. Extreme peak-power events (GPU cluster synchronisation transients at the millisecond scale, brief utility ride-through requirements at the sub-second scale) accelerate battery degradation disproportionately relative to the energy delivered. Hybrid supercapacitors (HSCs, sometimes called lithium-ion capacitors) fill this gap.

The AI Power Chain Capacitor Stack essays cover HSC technology in depth. The specific role at 800 VDC UPS is that HSCs can be connected alongside the battery on the DC side, either directly or through a dedicated converter, to take on short-duration power transients, taking on the sub-second transients that would otherwise be shed onto the battery. The potential result is longer battery cycle life at a moderate capex adder, and the ability to specify smaller-and-cheaper battery capacity because the peak-power headroom moves to the HSC. The battery needs two fundamentally different specifications: a power rating (MW), which sets PCS rating, current, thermal stress, protection and transient capability; and an energy rating (MWh), which sets ride-through duration, capacity, SOC window, degradation and augmentation. A 200 MW UPS does not necessarily require a 200 MW battery at the same duration if another storage layer handles the transient duty. On a 200 MW × 15 minute basis, 50 MWh is the delivered-energy target; installed battery energy will be higher after PCS losses, usable SOC window, reserve margin, degradation and temperature.

06Storage capex composition and the chemistry trajectory

The capex composition of an 800 VDC battery installation runs past cell cost alone. The component ranges are independent (cells 40-50 percent; BMS 15-22 percent; enclosure and mounting 8-12 percent; thermal management 8-14 percent; power electronics interface 10-15 percent; commissioning and controls 5-10 percent) and do not sum to 100 percent; a bill-of-materials structure stays absent. The chart below uses a separate illustrative 100 percent base case. The percentages shift with chemistry: LFP generally permits less intensive thermal-management requirements than NMC, although the actual system design remains application- and cell-specific and with voltage class (BMS goes up with voltage as noted above).

07Named vendor field at 800 VDC UPS

08Interaction with arc-fault and grounding (Parts IV, V)

Battery integration at 800 VDC interacts with the two prior essays in specific ways. The battery is an independent stored-energy source, but its contribution to DC fault current depends strongly on whether it is directly coupled to the bus or connected through a bidirectional PCS. That distinction belongs in the protection study, not merely the battery specification. A directly coupled 800 V battery stack can deliver very high fault current, with magnitude and rise time determined by the cell, pack, bus and protection impedance. Where the battery runs behind a bidirectional PCS, the PCS current limit and control/protection behaviour become part of the fault-current profile. The arc-fault protection topology therefore has to account for a battery-side source term that can differ materially from the source-side fault profile. The arc-fault protection topology (Part 4/13) has to account for this contribution; a battery-side fault has a different current profile from a source-side fault.

The grounding topology (Part 5/13) also interacts. In an IT-system architecture, the battery shares the ungrounded DC bus. A first earth fault does not create the same solid fault-current path as in a TN-S arrangement, but the battery still contributes to the DC system's fault-current and transient profile. Insulation monitoring therefore has to cover the battery-connected bus as well as the source side. Insulation-resistance monitoring has to distinguish between source-side and battery-side developing faults.

The practical implication for the operator is that the battery specification, the arc-fault protection spec, and the grounding spec have to be produced by the same engineering workstream at the utility-interface / perimeter architecture stage, not treated as a rack-level procurement problem. The corollary is that the availability architecture determines how many independent energy-storage domains the facility needs. A 200 MW facility with a 2N architecture differs from one with N+1, and neither is the same as a distributed architecture with multiple smaller batteries feeding separate DC domains. The right question moves past how big the battery is; it is how many independent energy-storage domains the availability architecture requires. Operators that try to procure batteries against a boilerplate spec without the arc-fault and grounding context can leave mismatches that only become visible during detailed protection studies or commissioning, when changes are more expensive to make.

09Operator procurement checklist

1. Chemistry decision documented against cycle-life duty profile. LFP is the working assumption for many stationary-storage and data-centre BESS applications, subject to duty cycle, safety, footprint, certification and supplier availability; NMC only where density is critical and cycle-life tolerance is well-understood; LTO for extreme-cycle grid-services applications.
2. BMS spec sized for the actual cell count at 800 V. Distributed cell monitoring, safety-certified isolation, redundant temperature sensing per module.
3. HSC augmentation modelled against actual peak-power profile. If the GPU cluster produces repeated sub-second transients above the battery's preferred operating range, model whether HSC augmentation reduces battery cycling stress enough to justify its capex.
4. Second-life battery option evaluated where capital constraint favours it. Requires additional BMS headroom and separate insurance underwriting.
5. Fault-current contribution documented alongside arc-fault protection design (Part 4/13). Battery-side fault current has a different profile from source-side and requires its own protection topology.
6. Grounding topology decision integrated with battery topology. IT-System behaves differently from TN-S on battery-side fault detection (Part 5/13).
7. Thermal management sized for chemistry-specific requirements. LFP is more thermally tolerant than NMC; under-specifying thermal management can accelerate degradation and reduce usable battery life.
8. End-of-life recycling plan committed at procurement. Regulatory position on battery recycling is tightening; take-back agreements at procurement reduce end-of-life cost.

10The reframe for anyone specifying now

The 800 VDC UPS operates as its own architecture, with capex composition, chemistry economics, BMS complexity, and interactions that all differ from the AC UPS in ways worth naming with the rest of the electrical protection design. Operators specifying a 200+ MW facility should not treat battery integration as a rack-level procurement decision late in the design. By the time it lands at the rack, the battery topology, PCS interface, arc-fault protection, grounding scheme and availability architecture may already have been constrained by decisions made at perimeter conversion and building distribution. Late discovery that these systems were designed against inconsistent assumptions is expensive at best and project-delaying at worst.

Part 7/13 moves into retrofit vs greenfield. The strategic decision that determines whether an operator gets the benefits of the 800 VDC transition on their existing facility footprint or has to build new capacity to capture them.