The contest for leadership in AI is also a contest to build and power the data centers behind it, with the economies that host that compute capturing significant value and a measure of technological sovereignty. Global compute capacity is expected to more than double by 2030, and securing a share means connecting data centers at an unprecedented scale and speed – with connection queues in many markets now running five to ten years. Power has become a decisive constraint on where data centers are built and how fast – alongside access to fiber, gas, cooling, and water.
Much of this investment is also mobile. A growing share of AI workloads are not latency-sensitive, so the compute can locate wherever power is most competitive or can be secured first. Connection speed has therefore become a lever in the competition between markets and regions.
Locally, though, data centers often face opposition. They are challenged for raising costs for other users, lengthening connection queues, inflating grid investment, and competing for scarce low-carbon power. The concern is legitimate, but the outcome is not predetermined. Whether a data center burdens or strengthens the system depends on how and where it connects, and how it runs once it is online.
Although we firmly believe load growth is a key lever for electricity affordability - spreading fixed costs over more users, raising utilization, and letting demand materialize for which we are expanding our electricity systems - this publication does not take a position on how much data center capacity a given country should host: that is a societal and political choice. It instead starts from the premise that once a country has decided a certain amount of data center capacity is in its interest, there is a better and a worse way of hosting it. It argues that there exists a way for data centers to be online years sooner without pushing up costs for others. It rests on two key levers. First, connect in a different place than today and/or connect non-firm: accept curtailment in the hours the grid is stressed, in exchange for skipping the years-long wait for firm grid capacity. Second, build the capability to act on it: shift flexible workloads and draw on behind-the-meter (BTM) generation and storage. Done this way, grid-positive data centers can deliver measurable benefits to both the system and themselves.
In Northwest Europe, the reality is that achieving this in already congested markets will require data centers to host significant BTM capacity. That adds to their power supply costs; in the cases we have studied, though, the value of energizing years sooner because of it outweighs this added cost. Even while that may be the case, under today's connection rules in many markets, data centers retain the right to wait for a firm connection wherever they choose to site, with non-firm access merely an option - so realizing this value in practice will likely require clear signals, commercial or regulatory, that connecting firm wherever a data center wants is simply not an alternative where it would negatively impact the system. None of this is a quick fix: it requires changing today's approach to siting, connecting, and operating data centers, and today’s rules regarding BTM usage and permitting - and it will not be feasible everywhere. But where it is, grid-positive data centers should be the goal.
To quantify this, BCG has built a model that jointly optimizes load flexibility, BTM assets, and firm versus non-firm grid access to find the least-cost way to meet data-center demand. This publication focuses on Northwest Europe over a 2030-2035 horizon, and sets out what that means for each stakeholder.
The authors would like to thank the following BCG colleagues for their valuable reviews and contributions:
Diego Vermeire, Sam Vandezande, Jamie Webster, Souhail Cherqaoui-Fassi, Pierre-Armand Jaboulay, Malo Grisard, Abdelhakim Khaouiti, Mikhail Nikomarov, Jan Zenneck, and Paul Martin.
