AI Is Pushing Data Centre Infrastructure Towards Integrated, Prefabricated Systems

AI is rapidly reshaping the physical infrastructure behind modern data centres, with higher GPU densities, greater power requirements and more demanding cooling needs forcing operators to rethink how facilities are designed and deployed. The shift is increasingly moving beyond traditional, site-built models towards prefabricated and integrated infrastructure that can bring power, cooling and white-space systems together more efficiently, while also shortening deployment timelines.

At the same time, enterprise adoption of AI is broadening the market beyond hyperscalers and colocation providers, creating new demand for infrastructure that can support higher-density workloads while remaining resilient, scalable and efficient. In hotter markets such as the Middle East, those requirements place even greater emphasis on thermal management and the need to adapt cooling strategies to local operating conditions. Alex Brew, VP Regional Sales EMEA at Vertiv, discusses how these changes are influencing customer requirements and shaping the next generation of AI-ready data-centre infrastructure.

What is changing most significantly in the data-centre market as AI adoption accelerates?

The technology and infrastructure required to deliver AI factories and this next generation of facilities are changing significantly. Continued evolution in GPU architecture is driving greater infrastructure density, and we are also seeing mass adoption of newer technologies such as liquid cooling.

But one of the biggest changes is the shift from traditional deployment models towards more prefabricated, off-site production-based methodologies for constructing data centres.

For us, this is critical because of the time-to-token element — the speed it enables without compromising on quality. From a sustainability standpoint, it is also broadly recognised as a more progressive construction technique.

It allows customers to deploy faster and realise returns on their investments sooner, based on the token output they can generate thereafter.

We are also seeing Vertiv increasingly recognised not simply as an industrial manufacturer or product supplier, but as a system-level integration and infrastructure partner. Customers are bringing us into projects at the very inception of the idea — for example, when they first decide they want to stand up an AI factory.

What are customers looking for when they talk about AI-ready infrastructure?

We are experiencing growth across markets and product lines, but there is particular acceleration around infrastructure solutions and the move towards prefabricated construction.

This brings more of our products together into a system-level offering. We talked about power modules, and we also looked at the SmartRun system, where the power train and thermal chain converge into a prefabricated white-space infrastructure block.

We are seeing significant demand closely tied to AI-factory-type data-centre deployments and the colocation market.

What is also interesting is the demand coming from the enterprise space. Enterprises increasingly want to realise AI in-house, and that is creating new infrastructure requirements that we are also catering for.

Does that challenge look different in hotter markets such as the Middle East?

In warmer markets, the constraint is actually less about the power itself and more about the thermal chain.

When you look at the heat-rejection plant, you need to consider different and more appropriate technologies for hotter climates in order to expel heat and deal with more extreme ambient conditions.

That becomes a particularly important part of infrastructure design in these markets.

With power availability becoming a major constraint globally, could on-site generation become more important for data centres?

On-site power generation is a major topic at the moment.

Part of that is because of the time it can take in certain markets to obtain the required permits and access to power. But on-site generation can also help facilities operate more efficiently.

It can provide a better way of addressing power requirements, particularly when there are load spikes, rather than relying entirely on the grid.

One important thing to understand is that it is inherently in the operator’s interest to make the data-centre system as efficient as possible.

Is the market moving towards complete infrastructure projects, or is there still strong demand for individual products?

It is a healthy mix.

There are obviously large-scale projects that we are working on, but we also have a very strong service business. At a product level, we continue to supply other parts of the market where our technologies are integrated into broader system architectures.

The shift towards system-level infrastructure is important, but that does not mean the individual product and service businesses disappear.

As GPU densities rise, how are liquid-cooling requirements evolving?

The evolution of GPUs is driving the requirement for more and more liquid cooling.

We are now also seeing liquid cooling interface with different parts of the IT stack rather than simply the GPUs themselves.

What is important to understand is the breadth of capability required to deploy liquid cooling at scale. We are talking about interfacing liquids directly with IT, and historically those two things have not operated in such close proximity.

That has changed rapidly with the rise of AI.

We have now deployed a significant number of installations globally and built substantial experience through our service organisation. It is not just about deploying coolant distribution units (CDUs); it also involves the secondary fluid network and taking a complete system-level approach.

That broader capability is where customers increasingly see the value we can bring to their projects.