By Sebastian Dooley, Portfolio Manager, Principal Asset Management
The outlook for data centres is formidable. Buoyed by strong fundamentals, ever increasing demand and improving efficiency, have primed the asset class for sustained, attractive returns. However, the market is maturing. Success will no longer be guaranteed by simply deploying capital. Instead, strategic discipline, site selection, and alignment with demand drivers will define the sector’s winners and losers.
For many investors, the attractiveness of the sector is magnified by the proliferation of AI. The logic being that as AI permeates across industries, the real assets that underpin its expansion and maintenance will similarly experience exponential growth. While the expansion of AI is creating strong tailwinds for the sector at large, investors should resist being swept up in the hype.
AI model training centred facilities are emerging as a prominent sub-sector of data centres, particularly in markets with abundant land and access to renewable energy, however it is the expansion of hyperscale cloud infrastructure that remains the primary driver of demand and returns, further boosted by AI inference that is starting to come through in these locations.
These cloud-oriented data centres are positioned far more favourably than facilities developed solely for generative AI model training. Cloud-oriented facilities offer more consistent demand and sustainable returns, though investor apathy risks the sub-sector becoming the “unsung heroes” of the digital infrastructure boom.
Behind the AI hype, fundamental risks remain
The potential of AI is undeniable, though investors must be diligent in how they seek exposure.
The model training process is essential for the development of AI and relies on massive facilities that demand significant upfront capital and ongoing maintenance. Sites identified for this use are not only costly to develop out but are also more abundant than apparent near-term European demand, which can dilute competitive advantages.
Despite these structural risks, investor enthusiasm for generative AI has fuelled a surge in valuations for model training sites. This speculative momentum risks inflating asset values beyond underlying fundamentals. Sensible investors will focus on whether long-term returns can justify current prices or are being inflated by short-term hype.
This has led some observers to draw parallels between today’s boom and the fibre buildout that faltered after the dot-com bubble. While there are similarities, particularly in the surge of sites announced for generative AI training, key differences stand out. Utilisation rates remain above 90% in most mature markets, and high barriers to entry, especially through power constraints, are curbing speculative development.
The lesson: stay disciplined and resist the hype. AI focused data centres are here to stay but, while poised for long-term growth, in the near-term profits will likely prove elusive.
The real opportunity lies in the cloud
As AI continues to embed itself in more aspects of our daily lives, it will generate more and more data overall, inducing additional demand through inference. This will continue to drive the need for data processing and connectivity infrastructure, spreading the benefits of AI across the entire data centre ecosystem, not just for costly, AI-specific sites.
Moreover, older facilities will not become obsolete as the technology advances, as these sites will still handle the traditional data centre workloads which will continue to grow alongside AI. Given the rapid expansion in this technology, it is more likely that a model training sites grows redundant. Take investors’ concerns around the DeepSeek announcement earlier this year. The low-cost, high efficiency model provoked investors to question the need for such an energy-intensive ecosystem to house and power complex AI models. Whilst the particular advancement DeepSeek initially seemed to bring now seem overstated, these concerns will persist as the technology continues to improve.
However, for investors in cloud-oriented facilities, the improved efficiency signalled an acceleration in AI adoption and demand for computing infrastructure, bolstering the sector’s growth.
Ultimately, AI training facilities remain an emerging segment, representing a growing, yet still modest share of overall demand. Investors focused on the AI segment exclusively risk missing the broader upside of AI’s expansion and while taking on inflated valuations and unpredictable returns.
Location restrictions enable opportunity
AI facilities are often remote, not requiring the same connectivity as cloud-oriented sites. As such, owners and operators struggle to benefit from the structural factors that continue to bolster cloud data centres.
Across Europe supply/demand imbalances are forcing vacancy rates to record lows in the already saturated FLAP-D markets, as well as secondary hubs like Berlin or Madrid. With land and power increasingly constrained, existing facilities become ever more valuable, even when equipped with aging mechanical, electrical, and plumbing (MEP) infrastructure.
The imbalance is largely driven by cloud-based facilities constrictions, which often must be built within specific geographical clusters to be effective. This makes it easier to forecast both where and at what rate demand will grow. Investors are now looking to create additional value by retrofitting existing data centres to support new workloads, meet new regulatory requirements and improve efficiency.
Considering these risks, it will be these network-dense facilities and flexibly designed assets in key availability zones that offer the most outsized risk-adjusted returns. These remain critical for cloud and connectivity demand, while poor site selection increasingly risks creating stranded assets.
The investment case for data centres is undeniable, though investors must be diligent in how they allocate to the sector. Cloud-oriented sites will continue to be reliably valuable assets, and considered investors will be those who prioritise location over hype. There are solutions out there, and the winners of the digital infrastructure boom will be those who keep their ears to the ground and pay close attention to the real drivers of momentum.
Sebastian Dooley, Senior Fund Manager, Principal Asset Management – Sebastian is the Senior Fund Manager covering the European Data Centre sector at Principal and is based in London having joined in February 2017. Previously he was responsible for managing a pan-European logistics separate account and was the Assistant Fund Manager for the Principal European Core Strategy. Prior to joining Principal, Sebastian worked in the Corporate Finance department of Telereal Trillium where he analysed M&A and current portfolio opportunities. Sebastian holds a Masters in Natural Sciences – Physics and Economics from Durham University and is a CFA® charterholder.
