Vertiv and Deloitte are highlighting a practical barrier to enterprise AI adoption: organizations must determine not only which workloads deliver business value, but also whether their infrastructure can support them reliably at scale. A new workload-based guide developed by Vertiv and digital infrastructure research firm STL Partners offers a framework for matching AI applications with deployment environments, power and cooling capacity, and physical infrastructure designs before enterprises commit to implementation.
Infrastructure Readiness Becomes a Key Enterprise AI Decision
Moving artificial intelligence from experimentation into sustained production requires more than selecting models and computing resources. Enterprises must also assess whether their facilities can deliver the power, cooling, connectivity, resilience and operational support that individual workloads require.
Vertiv Holdings and Deloitte Center for Integrated Research have introduced a collaboration focused on helping organizations address these decisions earlier in AI planning. The initiative centers on the Vertiv Designing Infrastructure for Enterprise AI Guide, developed by Vertiv and STL Partners, which provides business, IT and facilities leaders with a workload-based approach to infrastructure planning.
The guide’s central premise is that enterprise AI does not fit a single deployment model. Workloads differ in scale, performance requirements, data proximity, response times and operational criticality. Those differences can determine whether an application belongs in a public cloud environment, a colocation facility, enterprise-owned infrastructure, a hybrid deployment or an edge location.
Rather than treating cloud and on-premises infrastructure as competing choices, the framework aims to help organizations select an environment according to the requirements of each workload.
A Four-Step Framework for AI Deployment
The guide outlines four steps for connecting business objectives with infrastructure decisions.
First, organizations define the intended business outcome and the AI workload needed to achieve it. They then evaluate the workload’s scale, performance characteristics and operating requirements. The third step is to determine the most appropriate deployment location. Finally, the workload and location are matched to a physical infrastructure model and reference design for further planning and technical validation.
This sequence is intended to bring infrastructure constraints into consideration before deployment plans are finalized. An AI application may be technically feasible but still face barriers if a facility lacks sufficient electrical capacity, cooling capability, connectivity or room for expansion.
The guide also emphasizes operational ownership and governance alongside business value and technical requirements. Those factors become increasingly important when experimental projects evolve into systems that support recurring business processes or business-critical services.
Vertiv CEO Giordano Albertazzi said infrastructure considerations will play a significant role in the next phase of enterprise AI, particularly as organizations seek to provide the operating environment required by different workloads.
Nicholas Merizzi, principal and Silicon2Service U.S. lead at Deloitte Consulting LLP, similarly emphasized the need to consider business value, governance, operating ownership and infrastructure together as AI workloads mature.
Six Workload Archetypes and Four Infrastructure Models
The research groups enterprise AI workloads into six archetypes covering model development, retraining and four categories of production use. The purpose is to help organizations compare workloads with similar performance, operating and infrastructure requirements instead of planning around AI as a single, uniform category.
The framework then maps workloads to four physical infrastructure models: Edge AI Cluster, Small AI Cluster, Large AI Cluster and High-Density AI Cluster.
These models provide a starting point for evaluating power delivery, cooling, controls, connectivity, resilience, operational support and future capacity needs. The guide’s reference designs can be adapted after site assessments and more detailed engineering work.
The distinction matters because an infrastructure configuration suitable for a small deployment may not meet the demands of a larger or more computationally intensive workload. High-density AI deployments, for example, require careful evaluation of the facility’s ability to support concentrated computing loads. The appropriate design ultimately depends on workload requirements and the characteristics of the intended location.
The guide does not present a universal case for moving AI out of the public cloud. Exploratory projects may remain in cloud environments, while sustained or business-critical applications could lead organizations to consider hybrid systems or enterprise-managed capacity. Applications that require low latency, greater control or close proximity to data may be better suited to edge or other controlled environments.
Facility Assessment Moves Earlier in the Planning Process
A key part of the framework is assessing facility readiness before infrastructure decisions become difficult or expensive to change. Organizations are encouraged to evaluate available power, cooling, connectivity, operational capacity and room for future growth at proposed deployment locations.
That assessment can help identify constraints before they delay an AI project. It also connects business planning with facilities and engineering teams, whose requirements can otherwise emerge late in the implementation process.
The guide is a planning framework, not a guarantee that a proposed deployment will meet performance, cost or reliability targets. Its reference designs require adaptation to specific workloads, site conditions and detailed engineering validation. The announcement does not provide independent performance benchmarks or quantified evidence that using the framework shortens deployment timelines.
As enterprises move beyond pilots, the practical challenge is increasingly how to operate AI reliably within existing technical and organizational limits. Vertiv’s guide positions workload classification and infrastructure assessment as early steps in that process, helping organizations make deployment choices based on the needs of individual applications rather than a one-size-fits-all AI strategy.
Market Landscape
Enterprise AI infrastructure planning is shifting from experimentation-focused cloud access toward a portfolio approach that accounts for workload economics, inference demands, latency, data location and facility constraints. Deloitte’s 2026 enterprise AI infrastructure survey found that more than 70% of surveyed respondents expect to operate AI factories and edge AI deployments at scale by 2028. The survey included 515 U.S. enterprise leaders from five industries, with organizations generating more than $500 million in annual revenue.
Vertiv’s guide enters this market as a planning resource for organizations weighing cloud, colocation, enterprise-owned, hybrid and edge infrastructure. Its value will depend on how effectively enterprises translate the framework into workload-specific engineering, cost analysis and site validation.
Top Insights
- Infrastructure is part of AI strategy: Power, cooling, connectivity and operational readiness can limit production deployment.
- Workloads need different environments: Cloud, hybrid, enterprise-owned and edge models each serve different requirements.
- The framework has four steps: Define business outcomes, assess workload needs, choose a location and match it to an infrastructure model.
- Four physical models are proposed: Edge AI Cluster, Small AI Cluster, Large AI Cluster and High-Density AI Cluster.
- Validation remains essential: The guide supports planning; site assessments and detailed engineering are still needed.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
