Four reasons why the AI boom is a real estate story
AI's next frontier is real estate, as land, power and connectivity become increasingly valuable.
Autheurs
Every breakthrough in AI ultimately depends on physical infrastructure. Powerful computers need secure buildings, vast amounts of electricity, high-speed connections, and access to land that can support future expansion. Yet, these resources are becoming increasingly scarce.
Paradoxically, around 80% of the cost of building a modern AI data centre sits inside the building’s servers and IT equipment, rather than the physical facility. But that investment is worthless until it has somewhere to operate.
So, for investors, the question is no longer whether demand for AI infrastructure will continue to grow, but how that growth can continue when the physical infrastructure has become the bottleneck. The answer is unlikely to be a simple race to build more data centres. Real estate players hold the key: here we look at four reasons why.
1. Squeezing more compute from digital lemons
The cheapest data centre is the one you've already built. Data centres are deliberately designed with spare capacity to ensure uninterrupted service if equipment fails or demand suddenly spikes. Today, smarter software is allowing operators to use more of that available capacity. Rather than dedicating individual servers to a single task, computing workloads can be shared dynamically across many servers, ensuring idle capacity is put to work. For owners of existing facilities, increasing the productivity of current assets can be faster, cheaper, and less risky than constructing entirely new ones.
The trend is already well underway. According to AFCOM, the professional association for data centre operators, average rack density jumped from 16 kW to 27 kW per rack in the past year (almost a 70% increase) while 69% of survey respondents expect densities to increase further over the next 12 to 36 months.
2. Brains over brawn
AI infrastructure is becoming more efficient through advances in computing, cooling, and connectivity. Perhaps the most significant development is silicon photonics. Today's AI clusters rely heavily on copper connections, but these are approaching their physical limits as bandwidth demands increase. Instead, photonics uses light rather than electrical signals to transfer data, reducing power consumption and increasing bandwidth. Meanwhile, closed-loop cooling systems continuously recirculate coolant to remove heat more efficiently, allowing increasingly powerful chips to operate within the same building. For data centre owners, that means delivering more compute from the same building and power connection.
Technology is rapidly evolving. Researchers are even exploring biocomputing—using living organoids as part of computing systems. The human brain performs extraordinary cognitive tasks while consuming roughly the energy required to run a dim light bulb. If AI is to keep scaling, it may eventually need to borrow a page from the original supercomputer: the human brain.
3. Location, location... computation
Not all AI needs to happen in the same place. Training frontier AI models requires enormous computing power but is relatively insensitive to location. These facilities cam migrate to areas with abundant land and low-cost electricity. By contrast, inference (the process of putting trained AI models to work in applications such as ChatGPT) benefits from being located close to users, where every millisecond of latency matters.
This distinction reinforces the strategic importance of global cities. While some computing capacity will migrate to lower-cost regions, the infrastructure supporting real-time AI services is likely to remain concentrated around the world's largest urban economies, where dense fibre networks and proximity to users remain enduring competitive advantages.
This is where our proprietary HALO platform adds an additional layer of insight. By combining API-fed dashboards with AI-driven analysis, HALO identifies which companies already control scarce developable land and where markets remain underdeveloped. Paired with our understanding of local power availability and how community consent is evolving, we pinpoint which data centre landlords are making the most strategic portfolio decisions, before this is reflected in market pricing.
4. Watt comes next?
In many markets, the biggest constraint is no longer the building itself—it's access to electricity. Waiting several years for a grid connection is a non-starter for scale. As a result, data centre owners are bringing power generation closer to the asset. Battery storage and renewable energy are increasingly integrated directly into data centre campuses, reducing reliance on constrained electricity networks.
At the same time, data centres are becoming more active participants in the electricity system rather than simply consumers of it. AI training workloads can often be paused or rescheduled when the grid is under pressure, helping balance electricity demand while reducing operating costs. Backup battery systems, originally installed solely for resilience, are beginning to provide balancing services to the wider grid, creating new revenue opportunities while supporting network stability.
In the AI era, a data centre is no longer just a building connected to the grid, but it is increasingly becoming an active node in the electricity network.
The physical edge
The first phase of AI rewarded companies building chips and developing models. The next phase will reward those enabling the physical infrastructure on which AI depends. That represents an important shift for real estate investors: a property's competitive advantage may be measured as much in megawatts as square metres.
For listed real estate investors, digital infrastructure can offer exposure to one of the world's most compelling structural growth themes while providing something other types of technology investments might not: portfolio diversification and ownership of scarce, resilient, income-producing assets. As AI unfolds beyond a software story, the greatest opportunity may lie not in predicting the next breakthrough model, but in owning the real estate that makes every breakthrough possible.
Subscribe to our Insights
Visit our preference center, where you can choose which Schroders Insights you would like to receive.
Autheurs
Topics