MDI Access Talk to an Advisor

UPS Batteries and Human Error: Two Outage Causes You Can Manage

Operate • November 3, 2026

Outage rates per site have fallen for five consecutive years, according to Uptime Institute's 2026 announcement, but the pace of improvement has slowed. Costs have not followed the same direction: 57% of respondents said their most recent major outage cost more than $100,000, and one in five reported more than $1 million. Two threads run through the analysis year after year. Power equipment, UPS systems in particular, is where impactful outages most often begin, and procedures that exist but are not followed are the leading driver of the human-error outages. Neither is exotic. Both respond to unglamorous operating discipline, and both are worth looking at before the next budget cycle sets maintenance scope.

What the 2026 data supports

Uptime's Annual outage analysis 2026 reports that power issues accounted for 45% of respondents' most recent impactful outages in 2025, down from 54% in 2024 (a 96-respondent question drawn from the 2025 Global Data Center Survey). That 45% covers all power causes. Among major power-related outages, the report says the most frequently cited cause is UPS failure, followed by transfer switch and generator issues, though the prose gives no percentage for UPS alone.

Human error is framed differently. Uptime does not count it as a primary cause; it identifies it as a contributing factor in nine out of ten outages. Within human-error outages, the report names failure to follow established procedures as the leading driver, with inconsistent or unclear processes also common.

These figures should not be added together. Human error is a contributing factor measured on a different basis than primary cause, so the two categories overlap. A roughly-half framing is not something the 2026 data supports, so we have not used one.

Battery lifecycle: test, trend, replace on a plan

Most UPS battery failures are discoverable in advance if someone is looking at the right data. IEEE publishes the reference practices: IEEE 450-2020 covers maintenance, testing and replacement of vented lead-acid batteries, and IEEE 1188-2025 covers the same for valve-regulated lead-acid (VRLA) batteries. Both address test schedules and procedures and give guidance on when batteries should be replaced. The schedules and acceptance limits live in the standards themselves, so use the current edition rather than a rule of thumb passed around a site.

In practice, a workable program has four parts. Baseline every string at commissioning. Keep continuous monitoring on at least cell or jar level, because a string is only as good as its weakest unit. Run periodic discharge or capacity testing against the schedule you have chosen and documented. Then replace on a budgeted cycle driven by the trend, not by the first failed test. In our judgment, the trend across readings matters more than any single reading.

Chemistry changes the calculus. In a 2018 analysis, Schneider Electric put VRLA service life at 3 to 6 years and lithium-ion at 10 to 15 years, with lithium-ion capital cost roughly 1.5 to 2 times higher. That is a vendor's modeling assumption, not an independent benchmark, so test it against your own warranty terms and site conditions.

The human side: procedures people actually use

Uptime lists gaps in training, unclear procedures and staff fatigue among the sources of human-error failures, and recommends regular testing, staff training and more disciplined change management as mitigations. We would organize that into four working elements.

  • Written methods of procedure for any task that touches the critical path, with explicit rollback steps and a named approver.
  • Change management that holds a hard line on what counts as a change, including vendor work, and requires a second person to verify isolation before work starts.
  • Training that includes rehearsal. Walking through a procedure at a desk is not the same as running it on a de-energized path.
  • Staffing and fatigue limits that account for overnight coverage, not just headcount on the org chart.

Downsides and risks

None of this is free, and some of it can backfire.

Cost. Continuous battery monitoring, scheduled testing and planned replacement are recurring line items. Replacing on a trend means retiring some batteries earlier than a run-to-failure approach would, and the savings from avoided outages are probabilistic. Lithium-ion carries the higher capital cost noted above.

Lithium-ion and fire code. Uptime calls lithium-ion batteries in UPS systems "a clear contributing factor" to the rise in major data center fires, while noting the rise may partly reflect rapid build-out rather than inherent risk. It describes very early smoke detection combined with sprinklers as the most effective way to manage battery fires. NFPA 855 is the installation standard for stationary energy storage systems, and the local authority having jurisdiction can require fire-test data such as UL 9540A before approving a layout. Insurer requirements vary by carrier, and we have no published figures to cite; in our judgment, raise the question with the carrier and the authority having jurisdiction before the equipment is specified, not after.

Over-proceduralization. A procedure for every action slows response and teaches staff to treat documentation as paperwork. Reserve full methods of procedure for work that can take down a path, and keep emergency actions short enough to execute under stress.

Procedures on paper only. A binder nobody follows is arguably worse than none, because it creates false confidence. Uptime's own finding that failure to follow procedures leads human-error outages is the warning. Spot audits, observed walk-throughs and after-action reviews are what close the gap.

When it is not worth it. Schneider's analysis notes that with a UPS near the end of its life, staying on VRLA may be the better financial choice than moving to lithium-ion. The same logic applies to a facility already slated for decommissioning or a major electrical rebuild: put the money into the rebuild, and keep the testing discipline in the meantime.

Opportunities

Done well, the payoff is measurable in three places. The first is avoided outage cost, which Uptime's survey puts above $100,000 for more than half of respondents' most recent major outages. The second is a longer planned replacement cycle: JLL notes that UPS batteries typically carry warranties of up to 15 years, so replacement is a recurring cost to schedule rather than react to, and the same article observes that delayed replacement and gaps in round-the-clock monitoring can contribute to fires. The third is auditable operations. Test records, signed methods of procedure and change logs give owners, tenants and insurers something to inspect.

A documented case: Cloudflare, Portland, November 2023

Cloudflare's own postmortem of its November 2023 control plane and analytics outage is a candid account of power and process failing together. We found no public case of a company crediting a battery or procedure program for avoiding an outage, so this is a case of harm.

Per the postmortem, a utility feed to the Flexential PDX-04 facility was cut and the facility moved to generators without telling Cloudflare. A ground fault later tripped protective systems and shut down both utility feeds and all generators. The UPS batteries were rated for about ten minutes, but Cloudflare's equipment suggested they began failing after about four, and restoring power took far longer than ten. Cloudflare relays, as unconfirmed, employee accounts that overnight staffing lacked experienced electrical expertise. The incident ran about 40 hours. Cloudflare also faulted its own architecture, including untested whole-facility failure and hidden dependencies. The lesson for owners and tenants is to ask for evidence of battery test results, runtime assumptions and notification procedures, and to rehearse the loss of the whole site.

MDI can help with building a Preventive Maintenance Program that inventories UPS and battery assets, sets test and replacement schedules and records the history, and with Commissioning Services that establish a documented baseline before a facility goes live.

Talk to one of our advisors about UPS battery and procedure risk

Sources

← Back to Insights

Talk to an Advisor

Ready to plan your next data center project?

Get in Touch
855.634.3282
info@mdiaccess.com
Services Solutions Our Work Insights About
© 2026 MDI Access. All rights reserved.