For more than a decade, the data center industry has framed growth constraints almost entirely around power: how to secure more of it, shorten interconnection timelines, and manage rising energy costs while meeting sustainability goals. That focus remains valid, but it is no longer sufficient. A second constraint is moving rapidly from ESG discussions into operational reality: water availability. As AI workloads drive unprecedented rack power densities and thermal loads, water stress is becoming a material risk to reliability, scalability, and site feasibility; not merely a sustainability concern.

What is often overlooked is how water stress translates directly into power instability, particularly during the same extreme heat events that strain cooling systems. In that environment, the uninterruptible power supply (UPS) battery layer (already foundational to every data center’s power architecture) is being asked to absorb a wider envelope of electrical disturbances than it was originally sized for.

High-power batteries, with fast discharge response and tolerance for frequent cycling, emerge as the practical answer at that layer. The link between water stress and battery duty cycle runs deeper than it first appears, reshaping how operators think about the most embedded component of their resilience stack. Modern data centers can consume water near municipal scale, especially where evaporative or hybrid cooling strategies remain in use.

At the same time, AI deployments are pushing rack densities from historical norms of 5-15 kW into the 100 kW+ range, fundamentally changing thermal and electrical dynamics. Operators are responding with liquid cooling, higher temperature loops, rear door heat exchangers, and alternative heat rejection architectures.

These approaches can materially reduce onsite water consumption, often by shifting stress onto the electrical system through higher peak demand and faster load variability. In turn, this tightens power quality margins precisely when the grid is least cooperative.

Across these approaches, the trend is consistent: peak power draw is rising while power quality margins are narrowing. During heat waves and droughts, cooling systems are under maximum stress at the exact moment grid stability erodes. In traditional enterprise environments, brief interruptions or throttling may have been tolerable.

For AI training and inference workloads using continuous, high-utilization, and timing-sensitive applications, that margin no longer exists. Uptime expectations have not been relaxed as physical constraints have tightened. The pressure transfers downward into the layer of the architecture that must absorb every disturbance the grid sends through: the UPS battery.

Most discussions of water and power focus on generation, but for data center operators, the first impacts are felt downstream, inside the facility. Water scarcity affects grid reliability through a specific mechanism: a large share of global electricity generation remains thermally constrained and water‑dependent. During heat or drought conditions, cooling‑water intake and discharge limits force generation assets to derate just as demand peaks.

When constrained supply collides with peak demand, the disturbance signature splits into two regimes. The visible regime of rolling outages, capacity emergencies, and multi‑hour load shedding makes headlines and is rightly addressed by generators, battery energy storage systems (BESS), and microgrids. The less visible regime consists of voltage sags, frequency excursions, sub‑cycle transients, and brief interruptions. These events are far more frequent, rarely newsworthy, and disproportionately damaging to voltage‑sensitive digital infrastructure. It is this second, transient regime where the UPS battery earns its place as critical infrastructure.

Generators require seconds to minutes to start and synchronize. UPS batteries respond in milliseconds, holding the DC bus through disturbances, protecting sensitive equipment, and giving operators time to determine whether a generator transfer is warranted. In a grid becoming more transient-rich due to water-stressed generation, this millisecond layer is being asked to absorb a broader and more frequent set of events than traditional nameplate sizing assumed.

The UPS battery layer has always been foundational. What has changed is the duty cycle it is being asked to deliver, and that shift underpins growing interest in purpose-built high-power battery chemistry.

Pressure on the UPS layer is also coming from inside the facility. As rack densities increase, high-density cooling equipment, including pumps, compressors, coolant distribution units (and the variable frequency drives that power them), introduces fast load steps and high inrush currents onto the same buses that feed IT loads. When internal disturbances coincide with a stressed grid, exposure to voltage sags, spikes, frequency deviations, and nuisance transfer events increases. Millisecond-scale stress now arrives from both directions at once, and it lands squarely on the UPS battery.

Purpose‑built high‑power battery chemistries, such as UPS VRLA‑AGM (lead‑acid) and high‑power LFP (lithium iron phosphate), are designed for this operating regime rather than stretched to fit it. Key attributes include high short‑duration and continuous discharge capability to ride through back‑to‑back transients without terminal voltage sag, low internal impedance to maintain DC‑bus stability during fast load changes, and tolerance for frequent shallow cycling without accelerated aging.

In UPS applications, high‑rate VRLA‑AGM designs can support very high instantaneous and short‑duration C‑rates without reliance on active battery management systems, aligning well with power‑dense, millisecond‑response requirements. These chemistries also present meaningful sustainability tradeoffs. VRLA‑AGM benefits from a mature, closed‑loop global recycling infrastructure with high material recovery rates. LFP eliminates cobalt and nickel, avoiding associated ethical and supply‑chain risks, but relies on recycling pathways that are still developing and often less economically efficient at scale.

These differences matter because UPS chemistry decisions influence not only performance, but lifecycle impact, permitting complexity, and long‑term operability. Together, these characteristics align the UPS layer with the role it increasingly plays; owning the millisecond-to-minute window. While generators, BESS, and microgrids address what comes after.

Resilience in this environment is best understood as layered. UPS batteries manage milliseconds to minutes. Behind-the-meter or grid-scale BESS extends from seconds into hours, while generators carry load from tens of seconds into days. Microgrids, where deployed, address longer-term grid independence. Each layer has a role, and none replaces the others.

What distinguishes the UPS battery layer is immediacy: it is already embedded in every data center, paid for, racked, and operational from day one. As the transient envelope widens, upgrading the chemistry of that layer represents one of the lowest friction resilience improvements available to most operators.

Diesel generators remain a core component of resilience strategies, but their operating envelope is tightening. Emissions regulations, air quality limits, and growing community scrutiny have reshaped what constitutes “routine” generator operation, particularly during extreme heat events. In water-stressed grids, generators are called upon more often at precisely the moments when regulatory, operational, and reputational exposure is highest. The UPS battery layer cannot shorten the runtime a true outage demands (that remains the generator’s role), but it can reduce the number of times generators are called upon for disturbances that never warranted transfer. Every avoided start matters.

Operators in water‑stressed, AI‑dense markets are already recalibrating how they specify the UPS layer. Specifications are shifting toward sizing against the local transient envelope, greater emphasis on discharge characteristics such as C‑rate and impedance, and explicit performance criteria rather than stored energy alone.

UPS is increasingly treated as an operational buffer, not merely a bridge to generator transfer, but a tool to absorb disturbance classes that did not merit one. Resilience is no longer defined solely by how long a facility can run off‑grid, but by how gracefully it operates through instability that now arrives in shorter, sharper, and more frequent bursts. Operators can add generators, redesign cooling systems, and invest in long‑duration assets. What they cannot eliminate is the volatility introduced by water‑stressed generation. They can only absorb it.

That reality of stressed generation elevates the UPS battery layer as one of the most practical, immediately deployable levers operators must protect uptime as water, power, and climate pressures converge. Not because it replaces the rest of the resilience stack, but because it is already in place and ready to shoulder a wider transient envelope today. In the next phase of data center growth, success will belong not only to those who secure the most megawatts, but to those who recognize where resilience is now being tested, at the millisecond layer, and reinforce it accordingly.

Eric Hill, general manager, product and marketing, CSB Energy Technology

Akshay Viradiya, PE, senior staff data center engineer, LinkedIn

Rachel Rosenfeld, senior counsel, Norton Rose Fullbright