Front cover of the enterprise AI infrastructure design guide featuring Vertiv and Deloitte logos against commercial buildings

COLUMBUS, Ohio [October 15, 2026] — Vertiv Holdings Co. (NYSE: VRT), a global leader in critical digital infrastructure, has announced a collaboration with Deloitte Center for Integrated Research to help enterprises overcome key challenges in moving artificial intelligence (AI) from experimentation to production. The initiative focuses on helping organisations determine where AI workloads should run and assess whether their existing physical infrastructure can adequately support their operational requirements.

As part of the collaboration, Vertiv, together with digital infrastructure research firm STL Partners, has developed the Designing Infrastructure for Enterprise AI Guide. The guide provides a workload-based framework to help business, information technology (IT) and facilities leaders make informed infrastructure decisions before committing to deployment plans. By starting with intended business outcomes and AI workload requirements, organisations can evaluate deployment locations, assess infrastructure needs and determine facility readiness.

The underlying research highlights infrastructure readiness as a critical factor in determining where AI workloads can be deployed and how quickly they can move into production. The availability of power, cooling, connectivity and operational capacity can influence whether workloads can be supported within existing facilities or require alternative deployment models. Assessing these requirements early can help enterprises identify potential limitations before they become barriers to implementation.

The guide outlines a four-step process to support more structured AI infrastructure planning. This involves defining the intended business outcome and AI workload, evaluating its scale, performance and operating requirements, identifying the most suitable deployment location, and matching the workload and location to an appropriate physical infrastructure model and reference design.

Vertiv Chief Executive Officer Giordano Albertazzi said the next phase of enterprise AI would depend not only on AI models and computing capabilities but also on whether organisations could provide the power, cooling, connectivity and operating environments required by individual workloads. He added that starting with workload requirements would enable enterprises to make infrastructure decisions earlier, identify constraints sooner and establish a clearer path from AI ambitions to production deployment.

Meanwhile, Nicholas Merizzi, principal and Silicon2Service U.S. Lead at Deloitte Consulting LLP, highlighted the importance of considering business value, governance, operational ownership and infrastructure together as AI workloads progress into production. He noted that bringing these considerations into the planning process earlier would provide decision-makers with a stronger foundation for determining where AI workloads should operate and how they should be supported over time.

The workload-based approach recognises that enterprise AI comprises a portfolio of different applications rather than a single infrastructure requirement. Exploratory workloads may remain in public cloud environments, while sustained or business-critical applications may be better suited to hybrid infrastructure or enterprise-managed capacity. Meanwhile, workloads requiring faster response times, greater control, closer proximity to data or specific model performance may benefit from edge computing or other controlled environments.

Rather than favouring or rejecting cloud infrastructure as a whole, the guide provides a consistent framework for determining the most appropriate operating environment for each AI workload. Public cloud, colocation facilities, enterprise-owned infrastructure, hybrid environments and edge infrastructure can all play complementary roles, depending on the organisation’s requirements and deployment objectives.

The research categorises enterprise AI workloads into six archetypes covering model development, retraining and four categories of production use. These classifications help organisations identify workloads with similar performance, operational and infrastructure requirements, establishing a more consistent basis for infrastructure planning and decision-making.

Each workload can then be aligned with one of 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 requirements involving power, cooling, controls, connectivity, resilience, operational support and future capacity expansion. The corresponding reference designs can subsequently be adapted according to site assessments, workload requirements and detailed engineering validation.

Facility readiness remains essential across all four infrastructure models. Enterprises must determine whether proposed locations have sufficient power supply, cooling capabilities, connectivity, operational capacity and room for future growth. Conducting these assessments at an early stage can help organisations identify potential infrastructure constraints, reduce deployment risks and make more informed investment decisions.

Overall, the Designing Infrastructure for Enterprise AI Guide provides a comprehensive framework featuring six AI workload archetypes, four physical infrastructure models and a four-phase roadmap for planning, deploying and evolving enterprise AI infrastructure. By aligning infrastructure decisions with actual workload requirements, the guide aims to help organisations strengthen operational readiness and establish a more practical pathway for moving AI initiatives from pilot projects into sustainable production environments.

The full guide is available on Vertiv’s official website at Vertiv.com.