
We've spoken before about how much electricity is going to be demanded by data centers in the future. According to this study, data center energy usage is expected to represent somewhere between 6.7-12% of total electricity consumption in the US by 2028. And according to McKinsey, demand for data centers is going to at least 4x by the end of this decade. So the consensus is that we are going to need more, not less, data centers in the foreseeable future.
But if data centers represent the physical infrastructure needed for our digital activities, it's both interesting and valuable to think about where this stuff wants to go, especially since tech is, in some ways, a decentralizing force for cities. Interestingly enough, they exhibit the same economies of agglomeration as many other urban activities in that they want to be near density and other data centers. Maybe even more so.
Here's an excerpt from a Harvard Business School report (2022) called "Where the Cloud Rests: The Location Strategies of Data Centers."
The study finds a pervasive urban bias in the location of third-party data centers. For example, we find that all large metropolitan areas with over 700,000 population have at least one supplier. Less dense areas may or may not have any. Moreover, local entry rises with the presence of local information industries and intensive data users, such as finance, insurance, and real estate. Because less supply locates in the areas with lower density, a high fraction of buyers in small and medium-sized locations must get their services from non-local suppliers—likely located in the closest major city. Relatedly, we also find supply of more specialty services in denser and more competitive locations. We interpret all these patterns as the result of tension between economies of scale and user preference for proximity.

