Europe’s slow and steady approach to AI could be its strength

Europe is a highly fragmented market and is often said to be lurking behind the US and China when it comes to AI expansion.

But the very things that have hindered the company’s growth as a major player could give it an advantage in future-proofing the warehouses that are critical to powering the AI ​​boom.

Pankaj Sachdeva, a senior technology partner at McKinsey, told CNBC that the race is on to double, if not triple, the overall capacity of data centers built over the past 40 years, which McKinsey estimates will cost up to $7 trillion to build by 2030.

He expects the United States to account for the bulk of the activity, but that Europe will “continue to build at a fairly meaningful pace” and nearly double its existing production capacity.

“Europe is really participating in this infrastructure development and is really keeping pace, or we think we’re going to keep pace,” Sachdeva added.

To get there, the bloc will need to overcome major obstacles in access to power and regulation, experts told CNBC.

A critical bottleneck for Europe is access to electricity, with energy cost and availability shaping investment flows across the region. Scandinavia and Spain have an energy surplus thanks to hydropower and renewables, increasing appetite for data center construction, while Germany and the UK may be less attractive due to energy supply constraints.

Italy is one such winner in terms of grid congestion. According to energy think tank Ember, connection times can last up to three years, compared to the European average of four years.

The losers are once again Germany, the UK, Ireland and the Netherlands, and “basically, they either don’t have the grid capacity at the moment, or they don’t have the system, so it’s effectively a moratorium for the foreseeable future,” Jags Walia, head of global listed infrastructure at Van Ranschot Kempen, told CNBC.

Although there are significant differences between European countries, Walia said that ultimately “it will be difficult” for Europe to catch up in the short term with the United States, where deregulation and big investments have allowed it to set up its system much more quickly. He added that most European countries have about 200 to 300 data centers, while “the United States has about 5,400.”

Constraints are driving diversification away from traditional FLAP-D markets such as Frankfurt, London, Amsterdam, Paris, and Dublin and encouraging investment in well-resourced and stable data centers.

Efforts are also being made to develop projects faster. For example, in the UK, the central government overrode local governments to approve data centers that had previously been refused. Last year, the country designated data centers as critical national infrastructure, underscoring their importance to the economic agenda.

According to the International Energy Agency, energy consumption from power-hungry data centers could more than double from 460 TWh in 2022 to 1,000 terawatt hours (TWh) in 2026, with AI being the main driver.

Walia said electricity is the biggest cost factor for data centers, but a state-of-the-art facility could ease that burden.

This is a particularly troubling problem for Europe, where energy prices soared when Russia invaded Ukraine. Britain’s energy costs are among the highest in Europe, rising by around 75% since before the attacks.

This may be a deterrent to opening stores in certain locations, but operators aim to balance this with busy times on the power grid.

Kevin Restivo, head of European data center research at CBRE, said grid congestion has also sparked debate over how Europe procures electricity.

“There are a lot of speculators lining up, and those speculators are making things even more difficult because they have no intention of building data centers. They probably just want someone to have the power to flip it,” Restivo told CNBC.

For example, the UK operated on a first-come, first-served basis, and the importance of a project was not considered in deciding who received electricity first.

However, the system is currently moving to a “prepare first, connect first” process where completed projects can move to the front of the connection queue. This is partly designed to combat speculation. This reform shows how energy and infrastructure construction is forcing the evolution of old systems and setting the stage for further innovation.

At the same time, the steady pace of change allows developers to be more deliberate about what, where, and how they build. This means Europe may place greater emphasis on state-of-the-art facilities.

The quickest way for Europe to avoid these challenges, Walia said, is not to wait for new grid connections, but to ask: “Where are the good grid connections for declining industries now?” Such places can be converted from industrial hubs to technological hubs.

Europe is unlikely to lead in building AI hyperscalers or facilities for AI training, and while this race is considered largely won, the general consensus is that Europe may excel not only in facilities designed for AI inference, but also in smaller, cloud-centric connectivity-style facilities that require large amounts of fiber in and out.

In fact, there are few basic model developers on the continent, with France’s Mistral being the best known, but McKinsey estimates that 70% of all AI demand comes from inference.

As a result, Seb Dooley, a senior fund manager at Principal Asset Management, said the continent hasn’t seen “too many” announcements of large-scale AI-related data center construction, nor has it seen its “slightly overpriced nature.”

“So, actually, from our perspective, we see these regions as being well protected from any oversupply bubbles that may occur down the road,” he said, adding that the cloud is well established.

Principal Asset Management expects AI inference to occur in the same facility as the cloud, which is already the case at some cloud sites in the United States. This eliminates the speculative risks associated with other AI investments and provides investors with “fairly good returns”, the fund manager said.

It’s also an opportunity for Europe. Dooley said that with the widespread push for sovereign AI, inference will likely need to exist within Europe’s borders. However, the technical requirements are different. Densities tend to be higher than the 20 kilowatts per rack of traditional clouds, so data centers that want to run both will need to take that into consideration. Inference also requires a different cooling system.

“That means these facilities need to be designed with some degree of flexibility and robustness so that they can switch between two different systems as requirements change,” Dooley added.

So the joy of the slower, more thoughtful pace going on in Europe is that you have time to think about things like that.

The pace of AI development has led to rumors of a bubble, and if it bursts there will be a mountain of stranded assets. Even if AI continues its pace, as many believe it will, there is still a risk that data centers being built today will not be suitable in the future as AI’s technological needs change.

To help with this, investors are focusing on securing customers before construction begins. Speculatively built data centers are “mostly relics of the past,” Restivo said. Development and operating companies often lock customers into 10- to 15-year contracts, which can lead to obsolescence, he added.

However, this is different if the tenant itself is a start-up or young company. For example, Restivo said neo-cloud providers carry “significant risks” and have short contract terms of five to seven years.

“These companies are not returning capital to shareholders, they have unproven business models, and the demand for short-term capacity is very high,” he said, adding that “there is a lot of risk for developers and operators” working with neoclouds. But some debt lenders and developers are “increasingly comfortable” with these terms, Restivo added.

However, repurposing brownfield sites can be problematic. That’s when industrial plants that are still in operation are replaced with data centers, which means job losses. European policy requires developers to report data center energy and water usage and the legitimacy of specific locations.

Some member states go further. Walia pointed to proposed sustainability requirements in Spain that would require data center developers to report on their socio-economic impacts. “No one asks about that in the United States,” he said.

But Dooley predicted that tighter regulation would work in Europe’s favor in the long run, as data centers become integrated into local communities, “rather than just completely disrupting everyone’s lives sometimes,” and noted that sustainability is one area where the European Union is “very good at innovation.”

“From my perspective, what makes Europe so interesting and stands out is that it feels like a much safer investment case from a capital markets side compared to the United States,” Dooley said.

“A lot of that comes from the fact that it’s difficult to build in Europe. We have a lot of constraints, but the reality is that the harder it is to replicate something, the higher the long-term value of what you’ve got and the more likely people are to reuse it and come up with creative solutions to repurpose assets,” he added.

Ultimately, investors and developers may have no choice but to support Europe on this issue, thanks to sovereign AI being an “underappreciated” driver of data center construction, Jim Wright, manager of the Premier Mitten Global Infrastructure Income Fund, told CNBC.

Overall, Europe has an opportunity to innovate and create long-term value for both investors and citizens. Scarcity increases profitability and resilience for the former, while regulation fosters sustainable and constructive construction for the latter.

However, there is no one-size-fits-all approach to building data centers in Europe. “At this point, the industry is still in the ‘figuring out what exactly is needed,'” Dooley added.