Executive Summary
- Giuseppe Forgione, Sales Director, Regional Strategic Accounts, Northern Europe at Vertiv, sat down with Lauren Raybould, Chief Editor of DCi, to talk about how AI deployment is forging a new data centre ecosystem.
- Success is increasingly measured by the total time from initial planning to the moment an AI cluster begins generating output, often referred to as ‘time to token’. Meeting this objective requires a coordinated deployment model.
- No single company has all the answers because the reference architectures are evolving as quickly as the chips. The companies that thrive will be those that manage the complexity of thermal management and energy requirements through collaboration. The industry must operate as a collective to meet the deployment windows required by the market.
The data centre industry is currently navigating a fundamental pivot. How has the rise of AI factories changed the approach to infrastructure strategy?
The surge in demand for sovereign AI is prompting a total rethink of how hardware is deployed at scale. We’re seeing an evolution from traditional deployment hierarchies toward high-density environments that legacy methods struggle to support. National and industry-specific AI data centres must now keep data within jurisdictional boundaries while maintaining hyper-efficiency – driven by a critical need for speed and technical density that the sector has never seen before.
What is your organisation’s role in the AI data centre ecosystem?
Vertiv is a global leader in critical digital infrastructure and plays a pivotal role in the AI data centre ecosystem. We deliver the essential power, cooling, and integrated infrastructure solutions required to support the demanding high-density AI and high-performance computing (HPC) workloads of today and tomorrow. By partnering with hyperscalers, colocation providers, and enterprises on sovereign AI initiatives, we provide end-to-end power and thermal management systems that accelerate deployment timelines, maximise energy efficiency, and support the highest levels of reliability and performance.
Data centre construction once followed a predictable and linear hierarchy. Why is this model no longer effective in the AI era?
In the past, disciplines like power, cooling, and IT hardware operated in distinct channels with minimal contact. Today, AI infrastructure requires a much higher degree of coordination across these domains. When we look at a single rack of hardware representing millions of dollars in investment, the cost of idle hardware is simply too high. This all means that the primary goal has shifted to the speed at which a facility can start generating tokens.
You have mentioned that the metric of success has changed. Can you explain what that means for project management?
Success is increasingly measured by the total time from initial planning to the moment an AI cluster begins generating output, often referred to as ‘time to token’. Meeting this objective requires a coordinated deployment model. This involves bringing power, cooling, and hardware vendors to the same table from the very beginning. We’re now seeing high-velocity deployments where power infrastructure is delivered in months rather than years.
How does the transition to liquid cooling impact the mechanical design of these facilities?
In the industry, we’ve known for some time now that air-cooled systems can struggle to dissipate heat quickly enough to keep modern graphics processing units (GPUs) within their operational windows. In fact, IEEE Spectrum suggests that liquid cooling is up to 3,000 times more effective at capturing heat than air. I find that this shift introduces a new level of mechanical complexity to the data centre floor. It requires precise engineering of secondary cooling loops and Coolant Distribution Units (CDUs) to maintain stability.
What are the specific challenges of managing liquid cooling in a high-density environment?
The integration of these systems requires an intense focus on fluid dynamics and pressure management. Flow rates for major AI clusters can reach 120 litres per minute where even minor pressure fluctuations can compromise hardware stability. We believe that using CDUs to manage the interface between facility-side and rack-side cooling is now a baseline necessity. This orchestration allows for the thermal environment to remain stable even during peak processing loads.
How does the commissioning process change when the industry is dealing with such extreme power densities?
Before any live hardware arrives, facilities must undergo rigorous commissioning using load banks that simulate the heat and coolant flow of a real GPU cluster to confirm there are no leaks. Depending on the complexity and scale of the deployment, this process can take several months of preparation, testing and validation. The goal is to reach a state where the cooling system is a known quantity before the actual IT equipment is installed.
What role does software play in the design and operation of these high-density sites?
One of the most significant evolutions is the integration of digital twin technology at the design phase. By using digital twins to simulate the thermal dynamics of a facility, operators can conduct virtual commissioning before any hardware arrives. Vertiv indicates that these designs can reduce deployment times by up to 85%. This allows for the hardware to go live immediately when it does arrive.
How is Vertiv supporting faster AI data centre deployments while managing power and cooling challenges?
Speed is critical in today’s AI infrastructure race, and Vertiv is focused on helping customers bring projects online faster without compromising performance or reliability. We achieve this through standardised reference architectures and converged power and cooling solutions that significantly compress deployment timelines. For brownfield upgrades, our skid-mounted, plug-and-play modules allow rapid integration into existing facilities, while greenfield developments benefit from our more comprehensive industrialised platforms.
In a recent sovereign AI project, we progressed from initial engagement to delivering fully operational power skids in just a few months – a substantial improvement on traditional timelines and a clear demonstration of how an integrated modular approach can shorten time to token.
From the outset, our cross-functional teams collaborate closely with customers, aligning on proven reference designs, supporting full L1 to L5 commissioning, and supporting end-to-end integration between the IT stack and facility infrastructure. This collaborative model replaces rigid, siloed contracting structures, removing bottlenecks and enabling projects to move forward more efficiently.
Finally, what is the most important lesson for organisations looking to scale their AI capacity in 2026?
I believe success in this new era will be defined by the ability to orchestrate an integrated and coordinated ecosystem. No single company has all the answers because the reference architectures are evolving as quickly as the chips. I am convinced that the companies that thrive will be those that manage the complexity of thermal management and energy requirements through collaboration. The industry must operate as a collective to meet the deployment windows required by the market.
