If retail follows rooftops, what do AI data centers follow?
The baseline criteria for locating new retail sites that is used by retailers from quick service restaurants (QSRs) to big box stores is to look for new housing developments in what is referred to as “retail follows rooftops.” Car dealerships exhibit a similar strategy called “agglomeration” whereby they cluster along a highway in a gauntlet that can challenge even the most focused driver. But what factors influence the location of new artificial intelligence data centers? The easy answer is…
AI/DC Seeks E=MC2
A report by Bain and Co. states that “serving a 1-gigawatt data center requires the capacity of about four natural gas plants or around half of a large nuclear plant.” That’s an eye-popping statistic. Put in context, consider that Amazon alone is projected to build 200 new AI data centers over the next several years and more recently announced a $10 Billion investment in a data center in North Carolina. McKinsey projects U.S. data center demand (including AI) to grow from 25 gigawatts (GW) in 2024 to more than 80 GW in 2030.
If you do the math that comes to…well, we don’t have enough power plants either gas, nuclear, wind, or solar. Microsoft and Meta have secured energy contracts with Constellation Energy and specifically for nuclear power that will cover the next 20 years. The reason the timeline is so extended is that building and bringing online a new nuclear power facility is a 15-year endeavor. The U.S. has added only one nuclear power plant during that last 20 years, which is the Vogle facility southeast of Atlanta that has added only one-gigawatt of electrical generating capacity. The expected requirements for electrical generation capacity from AI data centers alone is projected to consume 9% of the total capacity of the U.S. by 2030. Today, that number is only 4%.
But it is more complex than just finding sources of electrical power. It’s jobs and the incentives that could be offered for economic development. In Huntsville, Alabama, for example, Meta’s data center expansion plans for that city will include 300 new jobs. If you take just the major hyperscalers, inclusive of Meta, Amazon, Google, IBM, Microsoft, Apple, and Oracle, each building has a minimum of 100 data centers globally, the ballpark estimate is approximately 250,000 new jobs, not counting the multiplier from adjacent supporting industries, which might add 1-2 million total new jobs, the impact is significant.
Available land may be the most limiting factor. The Meta facility in Huntsville for just the buildings alone is 3.5 million square feet or approximately 80 acres. The Stargate project announced by OpenAI, Softbank and Oracle will consume an 875-acre site in Abilene, Texas. The newly opened Meta facility in northeast Louisiana is 2,250 acres and adds 500 jobs. Finding land and the people to staff these facilities will put pressure on communities to determine how development will impact their lives.
Taken together, the site selection challenge will require the convergence of knowing the location of power, land, jobs in addition to favorable tax policy and broadband connectivity that will deliver the results of large language models (LLMs) and the promise of agentic AI.
So, while retail follows rooftops, it may also be following newly sprouting data centers as well.
Recent Comments