Vertiv and Deloitte team up to help enterprises choose where AI should run

Vertiv and Deloitte Center for Integrated Research introduced a collaboration on October 14, 2026, to help enterprises decide where AI workloads should run—and whether their facilities can support them. Alongside it, a new guide developed by Vertiv and STL Partners connects business needs to deployment locations, physical infrastructure models and reference designs.

The Designing Infrastructure for Enterprise AI guide is aimed at business, IT and facilities leaders making decisions before committing to an infrastructure plan. Vertiv’s announcement argues that available power, cooling, connectivity and operational capacity can shape both where an AI workload is deployed and how quickly it reaches sustained production.

Start with the job, not the building

Its four-step sequence puts workload requirements ahead of the final choice of physical model.

1. Define the intended business outcome and the AI workload that will serve it.
2. Assess the workload’s scale, performance and operating requirements.
3. Determine the most appropriate deployment location for that workload, using the requirements established in the previous step.
4. Match the workload and location to a physical infrastructure model and reference design, which supplies a starting point for further planning and technical validation.

The research groups workloads into six archetypes—categories with similar performance, operating and infrastructure needs. They cover model development, retraining and four categories of production use. Those groupings give planners a consistent way to compare workloads before choosing supporting infrastructure.

Organizations need to consider business value, governance, operating ownership and infrastructure together.

Nicholas Merizzi, principal and Silicon2Service U.S. Lead, Deloitte Consulting LLP

Cloud experiments and production have different needs

The guide does not advocate moving every AI workload into—or out of—the cloud. It treats enterprise AI as a portfolio: exploratory work may stay in public cloud, while sustained or business-critical work may favor a hybrid setup or capacity managed by the enterprise.

Other workloads may need faster responses, greater control, proximity to data or particular model performance. The guide says those requirements may point toward edge infrastructure or another controlled environment.

That leaves several options in play: public cloud, colocation, enterprise-owned capacity, hybrid environments and edge infrastructure. The guide’s comparison is workload-specific rather than a blanket recommendation for one hosting model. A company can therefore use different operating environments for different parts of its AI portfolio.

The approach follows an earlier Vertiv–Deloitte discussion published by Vertiv on September 22. In that DCD AI Week interview, Vertiv’s Martin Olsen described simulation, engineering and copilot workloads as having different needs. Deloitte’s Diana Kearns-Manolatos added privacy, security and response-time requirements to the factors shaping where AI should run.

A reference design still needs a site assessment

The guide identifies four physical infrastructure models: Edge AI Cluster, Small AI Cluster, Large AI Cluster and High-Density AI Cluster. These models are starting points for decisions about power, cooling, controls, connectivity, resilience, operational support and future growth—not finished engineering plans for every location.

Each proposed site still needs an assessment of available power, cooling, connectivity, operating capacity and room to grow. Vertiv says checking these constraints early can help identify barriers before they delay deployment.

The full guide is available on Vertiv’s website. Beyond the workload categories and physical models, it includes a four-phase roadmap for planning, deploying and evolving enterprise AI infrastructure.