What many people may have forgotten as they debate the great tech bubble, weigh up the challenges to Nvidia’s next chip move, or wonder if they've gone too long on US tech is that all of this technology has to be situated somewhere. Each AI application, query, system, procedure and process must be held in infrastructure and take its place in a network that allows it to perform its function unquestioningly and efficiently.

Datacentres themselves are nothing new; there have been buildings housing racks of servers for the best part of a quarter of a century now. They allowed companies to keep their computer equipment, data and records offsite, free up room in their own premises, and reduce the need to pay for the space to house them. However, more recently, with the rapid advancement of AI, a clear pattern is emerging: more space and more properties are needed to meet this unprecedented demand for the computing power required to allow artificial intelligence to function. In turn, this has supercharged the proliferation of datacentres: the very factories of a new digital economy, and arguably another revolutionary chapter in the human story.

We’ve already seen how datacentres were previously designed primarily for storage and conventional computing workloads. Today, generative AI consumes far more resources and requires far greater processing capacity to perform not only the tasks asked of it, but also the autonomous functions it carries out in its integration with everyday programmes. This amalgamation causes more power to be used, which also generates significantly more heat. As such, these datacentres have to be designed with specialised cooling systems and rack configurations, while the local power grids have to take into account the resulting higher energy demands, too. The net result is a new generation of AI-native infrastructure.

Aldric Viot, CEO of Voltekko, says this area of infrastructure has become one of the defining engineering challenges of the decade.

“Traditional air-cooling systems, long the industry standard, are increasingly inadequate for the thermal demands of GPU-intensive AI workloads,” he explains. “Immersion cooling – in which servers are submerged directly in a thermally conductive, electrically non-conductive dielectric fluid – is emerging as one of the most promising responses to this challenge.”

By transferring heat away from components up to a thousand times more efficiently than air, immersion systems can reduce a facility’s Power Usage Effectiveness (PUE) to near 1.0, eliminating the energy overhead that conventional cooling architectures impose.

“For operators running at hyperscale, this translates into material reductions in both operating costs and carbon footprint.”

It’s important to remember as well that not all datacentres are created equal. Many have different functions to perform and may also be housed in different styles of properties. Perhaps most relevant for investors is that these datacentres will be a mix of legacy facilities, AI-ready locations, and purpose-built AI campuses.

Historically, datacentres have sat at the intersection of commercial real estate and telecommunications infrastructure, but this binary view is becoming increasingly blurred.

Governments are increasingly looking at datacentres as a strategic national resource, as well as a source of capital inflows and investment into their countries. In April 2026, monthly spending on data centre construction in the US hit $50bn. The renewed push to establish sovereign AI companies and reduce reliance on software controlled by foreign states is increasingly paramount for some.

As these ambitions grow, so does the need for the infrastructure required to power them, at a time when cloud providers and AI companies are fighting for whatever spare capacity they can find. In this sense, ownership of digital infrastructure is becoming similar to the proprietary nature of energy facilities and transport networks. Now, it is no longer a property issue, but one that affects the whole tech, renewable energy and ESG architectures as the focus shifts to the construction process and the cost of bringing datacentres online.

Much has been made of the constrained nature of memory and processing chips as the world’s appetite for AI intensifies, but another issue is also in play. While the chips are integral for the tech to scale, the next bottleneck is the production of, and access to, power. Datacentres rely on a constant source of energy, which can put a strain on local grids in the process. Some construction sites face delays as hyperscalers are required to ramp up local connectivity frameworks and upgrade existing power frameworks.

It’s a reason why nuclear power has come to be seen as potential solution. Its ability to provide consistent, high-yield, low-carbon energy makes it an ideal solution, and, in the aftermath of the Iran war, less susceptible to the shocks and constraints faced by other energy sectors.

According to the World Economic Forum, up to 63 reactors are now under construction around the world. Access to energy is becoming just as important as access to chips and tech talent.

For investors, one does not simply 'buy' datacentres, but there are still a panoply of opportunities to be had further downstream. Datacentres are hugely capital-intensive, requiring significant levels of investment and financing deals, and infrastructure funds provide the long-duration cashflows that the centres need to be built. There are also real estate strategies to consider with specialist investment vehicles devised specifically for such digital infrastructure projects.

Then, there are abundant opportunities in private markets, which can provide access to the operators, developers and suppliers that are integral to their construction. Commodities and utilities – from metals and mining companies to businesses involved in power generation and grid expansion — offer another route to exposure. There is also the connectivity layer: the fibre and network infrastructure essential to data centres. Corning, the company that supplies iPhone with its robust glass screens, saw its shares rise by 270% after agreeing deals with Nvidia and Meta to supply fibre-optic cabling to their respective data centres.

The plethora of organisations and equipment that surround a datacentre’s construction and capability all ensure that investment opportunities extend far beyond the building itself.

“As regulators tighten sustainability requirements and energy costs remain elevated, energy efficiency is no longer a secondary consideration in datacentre design,” says Viot. “Efficiency is a core investment criterion, a competitive differentiator, and increasingly, a prerequisite for securing the power purchase agreements and grid connections on which new developments depend.”

Not everyone is greeting such progress with as much zeal as the companies involved. The environmental impact of constructing such large buildings, the amount of energy they need to function, as well as the amounts of water needed to cool the racks of servers are coming under increasing scrutiny. Local communities and regulators have questioned the huge amounts of resources that datacentres need to perform. Experiments to find alternatives have ranged from inserting datacentres within the sea, to potentially having them orbit the Earth in space.

At the same time, companies are realising that securing land for datacentres is not the be-all and end-all; sites must also offer a combination of regulatory certainty, political support, reliable power sources and water availability.

For many family offices and sophisticated investors, and as we’ve detailed recently in our piece about IPOs, much of the value creation in the datacentre space occurs before any public market access materialises. Large AI infrastructure projects increasingly rely on involvement at the private capital level to stay the course and remain viable, and it’s in this space where a lot of access lies. Family offices have the flexibility, pragmatism and precision to commit capital to enduring themes with long-term horizons. The issue of datacentres also has global significance with myriad cross-border flows and selection opportunities that allow the landscape to expand. As the issue gains momentum, some of the most attractive investments lie outside of the public markets altogether.

Whereas the first phase and hype surrounding datacentres focussed on models, chips and software, the next will be focussed on the more tangible and large-scale infrastructure projects that enable AI to function at scale.

As computing power becomes integral to governments and states (and their economies), investors can now go beyond the choice of either investing in AI and investing in the foundations that make this possible. Datacentres are evolving from a niche, real-estate proposition into one of the defining assets of the modern age.

At Arbra, we see every day how AI is creating demand where constraints exist. Power, cooling, land, connectivity and computing capacity are just five of the areas where opportunity can be found beyond the technology companies that dominate the headlines.