Schneider backs AI-era condition-based maintenance

Schneider Electric has published an IDC white paper on maintenance in AI-era data centres, arguing that calendar-based maintenance is no longer fit for purpose in many facilities.

The report says rising rack densities, multivendor estates and shortages of skilled technicians are forcing operators to rethink how they maintain critical equipment. It makes the case for condition-based maintenance, which uses monitoring and analysis of asset behaviour to identify faults earlier and reduce unnecessary service interventions.

Schneider Electric linked the findings to its EcoCare service model, which combines remote monitoring, expert oversight and predictive fault analysis. It said the approach shifts maintenance away from fixed schedules towards interventions based on equipment condition and operating limits.

IDC said the operational backdrop for data centre operators has changed sharply as AI workloads grow. The paper notes that rack power densities have increased from about 15kW per rack in standard data centres to 300kW to 600kW in AI-heavy compute zones, adding pressure on uptime and infrastructure resilience.

That shift is being compounded by the way operators are expanding capacity. According to the research, many are relying on existing installed bases, distributed campuses, on-site generation and brownfield strategies through mergers and acquisitions of local service providers, rather than building entirely new facilities.

Operational strain

The white paper also highlights the complexity of fragmented multivendor environments. Operators that acquire existing facilities can inherit equipment from multiple suppliers without a full operating history, creating challenges when integrating it into asset performance management systems.

"When operators acquire existing facilities rather than build from scratch," said Luis Fernandes, Senior Research Manager, IDC, "they introduce unknown equipment configurations from multiple vendors, with no operational history, requiring immediate integration with asset performance management systems."

Labour shortages add to those pressures. The research said the supply gap for skilled technicians has reached unsustainable levels, citing a US example where there is only one qualified person taking up a position for every seven open roles. Operators are struggling to recruit across electrical, mechanical cooling and commissioning roles, including positions that require specialist certification for high-voltage systems.

Against that backdrop, the study argues that fixed maintenance intervals are becoming less suited to the realities of AI-led data centre operations. Rather than carrying out work simply because of a date on a calendar, condition-based maintenance uses equipment data to determine when intervention is actually needed.

Schneider Electric said early adopters of AI-supported condition-based maintenance have reported fewer manual interventions, lower operating expenditure, less unplanned downtime, longer asset lifetimes and better efficiency. It added that its EcoCare offering can deliver up to a 75% reduction in unplanned downtime and a 20% reduction in operating expenditure, while also reducing risk.

Predictive model

Jerome Soltani, Global Head of Services at Schneider Electric, described the model as one focused on identifying abnormal behaviour in equipment and systems earlier. He said combining remote monitoring with AI-assisted orchestration can improve visibility into asset health and reduce disruption from unnecessary maintenance activity.

"By combining remote monitoring capabilities with AI-assisted orchestration, you can gain insights regarding the health of your assets and systems, and get an early identification of abnormal behaviour that might precipitate a failure," Soltani said.

"This ensures that downtime is minimised, but also that equipment working within specification is not disturbed or needlessly addressed."

IDC frames the issue as part of a broader shift in how operators manage infrastructure in more complex environments. Instead of treating maintenance as a routine schedule, the paper describes a model in which software-led analysis and human oversight combine to create a more continuous picture of system health.

Fernandes put that argument directly: "Your maintenance schedule doesn't know when something is failing - your equipment does."

He added: "Condition-based maintenance is an optimised operating model for AI-era infrastructure that reduces manual interventions, lowers OpEx, and extends asset lifecycle. By scaling predictive analytics to correlate behaviour across every vendor, asset, and failure trajectory, condition-based maintenance enables operators to build machine-driven, human-validated system intelligence."