The FDI angle

– Data centres accounted for $144bn in global greenfield FDI in 2024, a record-high share.
– CBRE expects 806MW in new data centre capacity to come live across Europe in 2025.

Why it matters: The surge in FDI for data centres highlights the growing importance of AI infrastructure, attracting global investments and shaping future technological landscapes.

Over the past decade, the data centre industry has increasingly found itself between a rock and a hard place. On one hand, the availability of land for new developments combined with rising energy prices and grid limitations has made starting new projects more difficult. On the other hand, the rapidly growing number of applications powered by large language models (LLMs) requires a significant uptick in data centre capacities suitable for training and inference tasks.

In the US, the ultimate manifestation of the growing artificial intelligence (AI) infrastructure demand and rising prices has been the Stargate project, a data centre venture envisioned by OpenAI and SoftBank with a planned budget of $500bn. Part of this investment may land in Europe, as the project leaders are reportedly weighing building data centres in the UK, Germany or France.

Advertisement

Stargate aims to mobilise €200bn in funding for AI projects across the continent

In the EU, the recent InvestAI initiative, announced weeks after Stargate, aims to mobilise €200bn in funding for AI projects across the continent, which includes €20bn to finance five of the so-called “AI gigafactories” — large-scale sites housing supercomputers and data centres. In a subsequent AI Continent Action Plan published in April, the European Commission set the goal of tripling the EU’s data centre capacity by 2032. 

At country level, most European governments have rushed to announce their own strategies of harnessing the AI boom, which often include significant infrastructure investments. By far the largest one at €109bn will land in France within the next five years, announced by president Emmanuel Macron at the country’s AI Action Summit in February.

This year, CBRE expects 806 megawatts (MW) in new data centre capacity to go live across Europe — a significant uptick from the 655MW delivered in 2024. The cost of these new projects is estimated at €9.7bn.

Overall, preliminary figures from fDi Markets show that data centres accounted for $144bn in global greenfield foreign direct investment (FDI) — or some 12% of the total investment, which is a record-high share for the industry.

Despite a stellar 2024 and being one of the largest industry players, Microsoft is seemingly slowing down its investment tempo. Recently, the corporation made news by pulling back on or delaying plans for data centre projects around the world. Those include sites in the UK, Indonesia, and Australia, as well as in Illinois, North Dakota and Wisconsin in the US.

Analysts at TD Cowen, who first spotted the move, said in a research note that the change of plans is most likely “tied to Microsoft potentially being in an oversupply position”. The corporation confirmed this in a statement to Bloomberg, mentioning that “we are well positioned to meet our current and increasing customer demand”. The researchers at TD Cowen connected it with Microsoft’s recent decision to forgo some new business from OpenAI, in effect ceasing to be its exclusive cloud provider and lowering its data centre capacity demand forecast.

Advertisement

Analysts predict no slowdown for the data centre industry in general, nor its AI-driven part in particular

The enormous investment campaigns announced in the US and Europe are based on the idea that the quality of the outcome of LLM models is strictly correlated with the scale and intensity of their training. In other words, the more data processing deployed for their training, the better. Nonetheless, recent breakthroughs challenge this idea. The Chinese DeepSeek model was trained for a fraction of the usual costs. If it becomes a harbinger of new, more efficient models, it may end up changing the calculus over future demand for data centre capacity too. 

The usual argument against this idea invokes the Jevons paradox, which essentially means that reducing the cost of a resource leads to an increase in demand, causing total consumption to rise.

Either way, analysts predict no slowdown for the data centre industry in general, nor its AI-driven part in particular. According to research presented by McKinsey in April, AI workloads are expected to account for 54% of all data centre capacity demand in 2025 — a figure poised to grow to 70% by 2030.

Building power-hungry data centres together with means of electricity generation appears to be an important trend in the industry, as it alleviates the strain such operations can put on the local grid. Known as creating “power couples”, the concept may transform the way data centres and grid operators work together.

“The power coupling conversation is about adding [electricity] generation at the same time and place as the load,” says Nick Turchak, an energy transition adviser. “Essentially, the goal of the concept is to assure the stakeholders that the new infrastructure won’t be subsidised by existing ratepayers.”

He goes on to explain that by building a wind or solar farm with additional battery storage together with the data centre, the facility would not add load to the existing grid. Often there would be a contract in place, limiting the possibility of the data centre to draw power from there. In addition to that, the “power couple” becomes a source of cheap renewable energy, the surplus of which gets exported to the grid. 

US-based Crusoe Energy has come up with its own data centre architecture that can be deployed at generation sites quickly and easily

Pushing the idea even further, the US-based Crusoe Energy has come up with its own data centre architecture that can be deployed at generation sites quickly and easily. One of the company’s unique offerings is capturing wasted natural gas from an oilfield (that would be flared otherwise) and converting it into electricity to power the server racks. Coincidentally, Crusoe is also the lead developer of the first data centre under the Stargate project.

In the long term, numerous researchers and start-ups are working on solutions that could make it easier to generate electricity on-site in the next two to three decades. Microsoft ran a pilot project in Ireland in 2024 using green hydrogen cells to supply energy to its data centre. Another potential game-changer would be small modular nuclear reactors that could theoretically become a stable and predictable power source for facilities located almost anywhere in the world.

While those ideas stay in the research labs, the data centre developers will have to stick to the proven recipe of finding new well-connected locations with affordable land and energy. The industry is expected to grow rapidly in the coming years (see page 40), with AI workload driving the demand alongside traditional use cases like cloud regions.

“There are multiple sources of demand for data centres supply,” says Kevin Restivo, director of data centre research in EMEA at CBRE, a commercial real estate services and investment company. “The enterprises of the world … will morph their operations over the course of time. And AI will be incorporated into those workloads, and therefore AI requirements will come from multiple classes of companies, not just neoclouds.”

With that in mind, new local champions are likely to emerge in the coming years to further expand the global interconnectivity map.

Andrii Degeler is a freelance journalist based in Amsterdam

Do you want more FDI stories delivered directly to your inbox? Subscribe to our newsletters.