Data centers have been top of mind for Missouri residents and policymakers. Here, we’re breaking down the increased use of AI applications, growth in data center construction, and resource demands of this growth. For more in depth information on any of these topics, check out our Science Notes linked below, or go straight to them here.
In Powering Data Centers, Dr. Maryluz Hoyos Ensuncho explained how data centers contain computers and servers that store and process data for various online services, including:
AI programs involve more processing and use more energy than other data center applications, driving more data center construction and power demand. In the U.S., data center electricity demand is forecasted to increase from 4% of all electricity consumption in 2024 to as much as 9% by 2030. In Data Centers Power Requirements, Dr. Hoyos Ensuncho found that a single hyperscale data center, which can require over 100 megawatts (MW) of power, can use 8.7 terawatt-hours (TWh) of electricity over the course of a year. For comparison, that’s more than one-eighth of the entire state of MO’s electricity generation in 2023 (66.7 TWh).
Hyperscale data centers are more energy-efficient than smaller scale data centers, and IT equipment has become more efficient over time. Practices like power management, virtualization, and investing in renewable energy alongside new data centers also help to reduce strain on the electric grid.
Overall, electricity demand in the U.S. is projected to increase 35-50% between 2024 and 2040, in part due to increased demand from data centers. In Addressing AI Energy Demand, Dr. Vikram Lakhanpal looked at how states are responding to the anticipated increase in demand. Three major approaches states have taken include:
Other state measures limit how much of the state’s power large load customers can use, or encourage businesses to bring their own power generation when constructing new facilities. State PUCs can also negotiate higher electricity rates for large load users or require them to pay additional fees to fund state programs.
Data Centers consume water directly through cooling processes, and indirectly through electricity consumption. Dr. Lakhanpal explores data center water consumption in Science Notes on Data Center Water Use and Closed Cooling Systems.
Data centers primarily run either cold air or cold water through server rooms to remove heat generated by computer equipment. The heat is then transferred outside, often using evaporative cooling, where water in a cooling tower absorbs the heat, evaporates, and is blown away. The cooling systems have a trade-off: air cooling consumes more electricity, and water cooling consumes more water. Data centers choose their cooling systems based on the local climate and which resources are more available.
Some data centers use a closed loop system, where water is recirculated in a closed loop between the facility and a heat exchanger. Reusing the water lowers the maintenance required in the closed loop and the water use compared to an open loop system that continuously pulls new water from a nearby source. However, removing heat from the closed loop still requires air chilling or another water loop for evaporative cooling, which requires consuming either electricity or water to remove the heat.
Water consumption refers to the amount of water a facility withdraws from the water system that does not return. Data centers can consume a large portion of the water they use:
In total, U.S. data centers consumed 17.4 billion gallons of water in 2023, equivalent to about 160,000 American households.
Dr. Lakhanpal also looked at the physical footprint and location of data centers in Data Center Land Use. He found that the average size of new data centers is growing, due to the construction of more wholesale and hyperscale data centers, which can cover hundreds of thousands of square feet. Data centers are typically built on a single story to maximize cost efficiency, though multi-story data centers have been built in dense urban areas.
Most data centers are built in metropolitan areas, to keep response times to users (latency) low. However, AI models undergo training before they’re used publicly, which uses a lot of power. Because training doesn’t require close connections to users, large data centers in remote areas where land is cheaper are ideal options to run AI training programs.
Powering Data Centers - What are the electricity demands of powering data centers?
Data Centers Power Requirements - What are data center power requirements and how are they projected to change?
Addressing AI Energy Demand - How are states responding to increased energy demands due to AI?
Data Center Water Use - How do data centers use water?
Closed Cooling Systems - How do closed cooling systems work?
Data Center Land Use - How much and what kind of land do data centers use?
Make sure you don’t miss any of our data center information!
Stay up to date by subscribing to our newsletter or following us on Facebook or LinkedIn.
Like what you’ve read? Donate to make sure we can continue to provide non-partisan research to Missouri residents and policymakers.