Vertiv & Deloitte issue guide to AI infrastructure

Vertiv has worked with Deloitte Centre for Integrated Research on a guide to enterprise AI infrastructure planning. It is intended to help organisations decide where AI workloads should run.

The framework is aimed at companies moving artificial intelligence projects from pilot stages into production. It sets out a workload-based process that links business goals and technical requirements to choices about deployment location, facility readiness and infrastructure design.

The guide, developed with STL Partners, a digital infrastructure research firm, is aimed at business, IT and facilities leaders assessing whether existing sites can support AI systems or whether a different setup is needed.

The approach starts with the intended business outcome and the type of AI workload involved. Organisations are then directed to assess the workload's scale, performance and operating requirements before selecting a location and matching it to a physical infrastructure model.

Those models are grouped into four categories: Edge AI Cluster, Small AI Cluster, Large AI Cluster and High-Density AI Cluster. According to the guide, each provides a starting point for decisions on power, cooling, controls, connectivity, resilience, operational support and future expansion.

Facility readiness is a central theme. Available power, cooling, connectivity, operational capacity and room for growth can determine whether an AI workload can remain in an existing facility or needs to move elsewhere.

Workload choices

The guide presents enterprise AI as a portfolio of workloads rather than a single deployment decision. In that view, some experimental workloads may remain in the public cloud, while more sustained or business-critical uses may be better suited to hybrid environments or enterprise-managed capacity.

It also points to cases where lower latency, closer proximity to data, tighter control or specific model performance requirements may push workloads to edge locations or other controlled environments. Public cloud, colocation, enterprise-owned sites, hybrid environments and edge infrastructure may each have a role, depending on the job being done.

The research categorises enterprise AI workloads into six archetypes, covering model development, retraining and four forms of production use. Vertiv said these groupings are intended to give organisations a more consistent basis for planning infrastructure around similar performance and operating needs.

Planning process

The planning method is set out in four steps: define the business outcome and AI workload, evaluate the workload's requirements, determine the most suitable deployment location, and align the workload and location with a physical infrastructure model and reference design.

Giordano Albertazzi, Chief Executive Officer at Vertiv, said: "The next phase of enterprise AI will be shaped not only by models and computing, but also by whether organisations can provide the power, cooling, connectivity and operating environment each workload requires. 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."

Deloitte's contribution focuses on the commercial and operating questions that emerge as AI systems move beyond trials. The collaboration reflects broader market concern that technical performance alone does not determine whether an AI application can be deployed at scale.

Nicholas Merizzi, Principal and Silicon2Service U.S. Lead at Deloitte Consulting, said: "The economics and operating requirements of AI change as workloads move into production. Organisations need to consider business value, governance, operating ownership and infrastructure together. Bringing those questions into the planning process earlier gives leaders a stronger basis for deciding where AI workloads should run and how they should be supported over time."

For data centre operators and enterprise technology teams, the message is that physical constraints may become a deciding factor in AI roll-outs. The guide argues that assessing those limits early can help avoid delays as projects move from experimentation to sustained use.

Vertiv operates in digital infrastructure, including power, cooling and IT systems for data centres, communications networks and industrial sites. The guide includes six workload archetypes, four infrastructure models and a roadmap for planning, deploying and adapting enterprise AI infrastructure.