FEATURE
AI is exposing the limits of traditional operating models
Three industry trends are accelerating this transition.
The first is workload volatility. AI training and inference create highly dynamic power profiles that fluctuate far more rapidly than traditional enterprise applications. Infrastructure now has to respond to continuous variation rather than predictable peaks.
The second is density. Higher rack power has accelerated the adoption of liquid cooling and new generations of AI infrastructure are increasing operational complexity across every part of the facility.
Steven Brown, Vice President, EcoStruxure IT Line of Business, Schneider Electric
Schneider Electric’ s own AI-ready infrastructure reflects this shift, highlighting the growing importance of power, cooling, rack design and integrated operational management as AI deployments scale.
The third is simply scale. Across every region, operators are trying to deliver new capacity faster than ever before while navigating constraints around power availability, skills and sustainability. Success increasingly depends on making better operational decisions rather than simply adding more infrastructure. platforms focused on IT assets and capacity planning and SCADA systems managed industrial processes.
Each was designed to optimise a specific operational domain and, for many years, that was enough. The problem now is that AI workloads do not behave in isolated domains.
Rapid or cascading increases in GPU utilisation immediately affect power demand. Higher electrical loads influence cooling performance.
Cooling decisions affect rack temperatures. Those changes can have implications for workload placement, infrastructure resilience and operational efficiency, often within seconds.
Understanding any one of these systems in isolation is no longer sufficient. Hyperscale, colocation, neocloud and enterprise data centre operators increasingly need to understand the relationships between them. This represents a significant shift in how today’ s data centres need to be managed.
Many organisations have responded by introducing AI-powered analytics within individual operational systems. These solutions undoubtedly provide valuable insight, but they often remain focused on a single domain.
Making individual systems more intelligent does not automatically create an intelligent data centre. The real challenge lies in connecting operational context across the entire facility.
Why operational intelligence matters
Most operators can already see what is happening inside their infrastructure. The more difficult question is understanding why it is happening and what will happen next.
If temperatures begin to rise within an AI cluster, is the underlying cause an increase in workload, a change in cooling distribution or an emerging electrical constraint?
If additional compute capacity is provisioned for a customer, will the existing power and cooling infrastructure continue to operate within safe limits?
www. intelligentcio. com
INTELLIGENT CIO EUROPE
25