Why colos are city slickers and hyperscalers are country bumpkins
Datacenter building decisions tend to fall into two camps with colocation providers plumping for urban areas while hyperscalers seek sites where electricity, land, and construction costs come cheaper.
A study by the William Marsh Rice University in Texas, commonly referred to as Rice University, found that colocation facilities tend to be found in areas such as cities for proximity to their business customers, while hyperscale server farms follow a different tack.
Giants like Amazon, Google, and Microsoft prefer to concentrate massive facilities in a relatively small number of lower-density regions, where energy, land, and construction costs are lower, making economies of scale easier to achieve, the researchers say.
The study, by assistant professor of strategic management Tommy Pan Fang at Rice University and Shane Greenstein of Harvard Business School, offers a large-scale statistical analysis of datacenter location strategies across the US. The aim is to provide both policymakers and firms with a clearer understanding of how different types of facility get located in response to economic and strategic incentives.
Using county-level data, the researchers say they were able to model how population density, the mix of industries, and operating costs all help to predict where new datacenters are most likely to be sited.
Third-party providers, meaning colocation and hosting companies, locate in urban areas because that's where customers (financial, etc) are based, and they place a high value on low latency, secure storage, and compliance with regulatory standards.
But that doesn't mean that hyperscale campuses are likely to pop up in any old rural setting. As well as power and water, datacenters need network infrastructure that is sufficient to meet their requirements.
- New carbon capture tech could save us from datacenter doom
- Virginia's datacenter tax breaks cost state $1.6B in 2025
- Google snaps up datacenter power biz Intersect while xAI plans more capacity
- AI is rewriting how power flows through the datacenter
This is why areas such as Northern Virginia, the world's "datacenter capital," have proven such a hotspot for server farms, and hyperscale facilities in particular. Here, between 10 to 15 percent of global hyperscale capacity is located, at least double that of anywhere else, having benefited from early network infrastructure investment that made it a draw for digital traffic, the researchers say.
But future datacenter expansion may follow a more hybrid pattern, according to the study's authors, with massive regional hubs complemented by strategic colocation facilities where latency becomes an issue.
Previous reports found that datacenters predominantly focused on AI workloads are less sensitive to latency, and may be located some distance from cities, so long as power is available.
Meanwhile, in Europe, the availability of land and high energy costs are the main factors, alongside long delays to get connected to the grid. High demand in the main Euro datacenter hubs is driving investment in emerging secondary markets, including Oslo and Madrid. ®