Artificial intelligence is forcing Europe to confront an inconvenient truth: the limits of its energy system, not its innovation capacity, will decide where AI infrastructure succeeds or fails. In conversation with Paul Mortlock, Head of European Data Centre Capital Markets at CBRE, this article explores how power availability, pricing and delivery timelines are quietly rewriting the economics and geography of AI development.
Artificial intelligence has created a surge of demand for new data centre capacity across Europe, but the shift is not simply technological. It is architectural, financial and geographic. The legacy cloud era was shaped by availability zones, latency constraints and sprawling locations. The AI era is being shaped by megawatts, energy economics and time-to-power.
Paul Mortlock, Head of European Data Centre Capital Markets at CBRE, describes the change as a structural break rather than a market cycle. He argues that AI has altered both the physical blueprint of facilities and the risk profile of the tenants that occupy them. The result is a market that is less predictable, more volatile, and far less dependent on traditional real estate fundamentals.
“I think the most noticeable difference is the amount of power you can channel through a single rack,” Mortlock said. “High density computing means you can run a lot more power through the same square metre of footprint. The implication is that you need less floor space for the same amount of compute, and that changes the assumptions the real estate market has historically worked with.”
The shift is not theoretical. Density has real-world consequences. It generates heat, and heat forces a redesign of building systems. “Traditional air-cooled designs are not capable of cooling this hardware. The industry is pivoting towards liquid cooling to keep these higher density deployments stable. Air conditioning fans simply cannot reduce heat at the scale these racks produce.”
AI is therefore reducing space requirements, increasing power requirements, and forcing a new generation of cooling infrastructure. It is not an upgrade to the existing paradigm. It is the emergence of a fundamentally different one.
The value of land is becoming the value of energy
Mortlock believes the most important change is not technical but economic. For two decades, the value of land for data centres was defined by geography. Companies built where the hyperscale cloud built, and they built at scale. Land prices followed that gravity.
“The implication of AI is that you fundamentally need less space but vastly more power,” he said. “Sites with very large volumes of power, north of one hundred megawatts, ideally more, are extremely interesting. As long as they can deliver power quickly, then they become valuable assets very fast.”
Speed of energisation has become a competitive edge. So has certainty of supply. The new AI entrants that are driving market demand want hundreds of megawatts delivered within twelve to eighteen months. They are unwilling to wait out decade-long permitting cycles or slow grid upgrades.
“That is what has changed dramatically. Historically, the land had to be located in a specific hyperscale zone. That is not the case any more. It is about where there is a vast amount of power and where you can deliver that power quickly. Locations are more agnostic. Power and speed to power have become the determining factors of value.”
The irony is that the tenants driving that value are not hyperscale giants. They are emerging GPU cloud providers with weaker credit ratings and less predictable business models.
Mortlock acknowledges the asymmetry. “The tenants sitting inside these buildings are nowhere near as good from a credit perspective as the hyperscale cloud companies. They are not investment grade. So the valuations, the cap rates you put on their income streams, are not as aggressive as what you would put on a Microsoft covenant. There is more risk, and therefore valuations are lower.”
The AI boom is therefore a paradox. Demand is unprecedented, value creation is real, but the income stream is less stable and the timeline to failure is shorter. Investors are forced to chase scale, speed and power availability while accepting credit volatility.
Europe is running out of capacity
The constraint beneath all of this is brutally simple. Tier one markets are out of power. They are not short of space or technology. They are short of electricity.
“In tier one markets like Frankfurt, London and Paris, if you make an application to the grid now, you are going to be waiting six to eight years or more for power,” Mortlock said. “That may also be conditional on new grid infrastructure. In tier two markets, you are talking four to six years. Anything that has power before 2030 is super interesting.”
This reality sits awkwardly against political messaging that promises streamlined permitting, accelerated investment and simpler regulatory pathways. Governments across Europe have declared data centres as strategic assets and positioned AI as a national imperative.
Mortlock does not dismiss the intention, but he questions the narrative of acceleration. “Power is not being delivered faster. It costs a lot of money and it is very complex from a permitting perspective. Governments can simplify planning, but that does not create megawatts. The NIMBY effect is real. People do not want data centres next door.”
The UK has gone further than most in framing data centres as critical national infrastructure, but the economics remain unforgiving. “There is recognition from government that data centres impact GDP, and they are trying to simplify planning. But the physical constraints are significant, and they are not going away.”
The price of power is redrawing the map
Power scarcity is only one dimension. Cost is the other. AI workloads are energy intensive, and the economics of deployment are shaped by the cost of a kilowatt hour, not the cost of land. Mortlock illustrates the spread.
“In London you are broadly talking twenty pence per kilowatt hour. In Finland, it is three to five euro cents. That is a massive difference. The cheapest power is in the Nordics and Iberia, and that is driving demand into those regions.”
Norway is one of the clearest examples. With more than a thousand hydroelectric stations, it has become a magnet for inbound developers. Spain is emerging for similar reasons. A corridor between Madrid and Barcelona has attracted large scale development because of abundant solar and wind generation.
Portugal exhibits similar characteristics, while France has benefited from nuclear baseload. The UK, Mortlock argues, is disadvantaged on cost. “It is not a place where you typically see large AI deployments because the power is expensive. You do not see big hundred megawatt deployments in the UK. It is too costly.”
Even atypical developments are driven by energy, not hype. Blackstone’s plans in the north east of England, built on land that once hosted a power station, are the exception that proves the rule. Power availability made the site attractive. Cost economics will determine whether tenants follow.
A new generation of capital is taking risk, not collecting rent
The capital entering AI infrastructure is not conventional. Institutional investors dominated cloud because they could buy stability, scale and predictable income. AI attracts a different cohort.
“There is a lot of capital looking at AI, but it tends to be the type of capital that needs higher returns,” Mortlock said. “Value add or opportunistic capital. To get those returns, you need to get in earlier and take development risk, leasing risk, and capital risk.”
The risk is not just construction or market timing. It is credit. “The covenants are not investment grade. That makes the income less secure. So developers are buying land that does not have power, does not have permitting, and does not have a tenant. The earlier you get in, the higher the risk and the higher the return.”
Speed compounds the equation. AI facilities can be built faster because critical systems are containerised. “You can plug and play. You can build much more quickly than traditional cloud. That matters because demand is immediate and backlog is severe.”
A continental market defined by physics, economics and impatience
Mortlock believes the sector is still being interpreted through a cloud-era lens. This leads to confusion, optimism and mispricing.
“The market has been driven for twenty years by location and latency. That model is not relevant to AI. The driver now is megawatts, cost of energy and time to energisation. Everything else is secondary.”
The change is not gradual. It is abrupt. Density has inverted space economics. Power scarcity has inverted geographic logic. Credit risk has inverted valuation patterns. Capital flows have inverted investment timelines.
The result is a market defined by three pressures. AI workloads need more electricity than the grid can supply. Operators want power faster than the system can deliver. Investors want returns faster than tenants can guarantee.
None of this is stabilised by regulation or architecture. Every variable is still moving.
Mortlock is pragmatic. He does not present AI infrastructure as a bubble or a revolution, but as a sector operating under a new set of physical and financial laws.
“Everything the sector used to optimise for has shifted,” he said. “Power availability, cost of energy and speed of delivery are the dominant forces now. The traditional assumptions about land, valuation and location are less relevant than they have ever been.”
