How a data center UPS actually works — double-conversion topology, why lithium-ion has largely replaced VRLA batteries, how UPS runtime is sized, and what happens electrically in the seconds between a utility outage and generators reaching full output.
The UPS's Job: Zero Interruption, Not Long-Duration Backup
A data center UPS exists to do one very specific thing: guarantee that IT equipment sees continuous, clean power with zero interruption, even during the brief but real gap between a utility outage and onsite generators reaching stable full output. It is not designed to run the facility for hours — that's the generator's job (see Data Center Generator Systems Explained). A UPS's battery runtime is typically sized for just a few minutes, enough to comfortably outlast generator start and transfer time, not to serve as a long-duration backup power source in its own right. Misunderstanding this distinction is one of the more common mistakes people make reasoning about data center power architecture.
Double-Conversion Topology: Why "Online" Matters
Nearly all large data center UPS systems use double-conversion (also called online) topology, meaning incoming AC power is continuously converted to DC, then back to AC, before it ever reaches the IT load — the UPS is always actively in the power path, not standing by waiting to switch in during an outage. This matters for two reasons beyond outage protection: it completely isolates downstream equipment from utility voltage sags, surges, and frequency variation, and it filters out electrical noise the IT load itself generates — including the sharp power swings an AI training cluster can produce during synchronized events like checkpointing, discussed in AI Data Center Electrical Design — Complete Guide. A lower-cost UPS topology (line-interactive or standby) simply can't provide this level of isolation, which is why double-conversion has become the standard for facilities at this scale despite its higher cost and slightly lower efficiency.
The Battery Shift: From VRLA to Lithium-Ion
Valve-regulated lead-acid (VRLA) batteries were the standard UPS battery chemistry for decades, but lithium-ion has become increasingly dominant in new large-scale data center UPS deployments. Lithium-ion offers a meaningfully smaller physical and weight footprint for the same energy capacity, a longer usable cycle life (fewer full replacement cycles over the facility's life), and faster response characteristics — all of which matter more at hyperscale, where UPS battery rooms can otherwise consume significant floor space and where replacement logistics for large VRLA banks are genuinely disruptive. The tradeoff is higher upfront cost per unit of capacity, though total cost of ownership over the system's life increasingly favors lithium-ion once cycle life and footprint are factored in.
How UPS Runtime Is Actually Sized
UPS battery runtime is sized as a deliberate engineering decision, not a default — it needs to comfortably exceed the worst-case time for onsite generation to reach stable, transferable output, with margin for a failed first-start attempt on a generator (most designs assume at least one generator may fail to start and require a second attempt). A typical target is in the 3-10 minute range for a data center, though the exact number depends on the specific generation system's documented start and transfer time. Oversizing UPS runtime well beyond what's needed adds significant cost (batteries are one of the more expensive components in the chain) without a corresponding benefit, since the generators are expected to be carrying load well within the sized window.
Where UPS Systems Sit in the Redundancy Architecture
UPS redundancy follows the same N+1 or 2N logic covered in N+1 vs. 2N Data Center Power Explained — an N+1 UPS configuration has one spare module beyond what's needed to carry the load, sharing a common output bus, while a 2N configuration runs two entirely independent UPS systems as part of the facility's fully separate A and B power paths. The choice between the two tiers for UPS specifically follows the same cost-versus-fault-tolerance tradeoff as the rest of the facility's power architecture.