Vertiv and the Deloitte Center for Integrated Research have introduced a framework to help enterprises determine where AI workloads should run and whether existing physical infrastructure can support them. The guidance is designed to reduce the risk of AI projects stalling between pilot and production because power, cooling, connectivity or operational capacity are insufficient.
The Vertiv Designing Infrastructure for Enterprise AI Guide, developed with STL Partners, starts with the intended business outcome and the AI workload rather than with a blanket cloud-versus-on-premises decision.
Organisations assess a workload’s scale, performance and operating requirements, select a deployment location, and then match it to an appropriate infrastructure model and reference design.
“The next phase of enterprise AI will be shaped not only by models and computing, but also by whether organizations can provide the power, cooling, connectivity and operating environment each workload requires,” said Giordano Albertazzi, chief executive officer of Vertiv. “By starting with the workload, enterprises can make infrastructure decisions earlier, identify constraints sooner and build a more credible path from AI ambition to production.”
The framework classifies enterprise AI into six workload archetypes: model development and experimentation; operational retraining and recalibration; generative and interactive applications; pattern recognition and anomaly detection; predictive applications; and real-time control and autonomous systems.
Each workload can then be aligned with one of four physical infrastructure models: an Edge AI Cluster, Small AI Cluster, Large AI Cluster or High-Density AI Cluster. These models provide a starting point for decisions on power, cooling, controls, connectivity, resilience, operational support and future expansion.
For Asia-Pacific enterprises, the framework is particularly relevant where AI adoption must contend with constrained grid capacity, high energy costs, tropical cooling demands, data-residency requirements and limited space in legacy facilities.
Exploratory workloads may remain in public cloud, while business-critical, latency-sensitive or data-proximate workloads may require hybrid, enterprise-owned or edge capacity.
“The economics and operating requirements of AI change as workloads move into production,” said Nicholas Merizzi, principal, Silicon2Service US lead, Deloitte Consulting. “Organizations need to consider business value, governance, operating ownership and infrastructure together.”
Vertiv says early facility-readiness assessments can identify constraints before they become deployment barriers. Its reference designs are intended to be adapted through site assessment, workload requirements and detailed engineering.
AI production planning must begin with infrastructure: enterprises should assess whether a site can support a workload’s power density, thermal profile, connectivity and growth trajectory before committing to deployment.
