AI is wild card in data center energy demand, say electricity researchers

Data centers could consume upwards of 9% of US electricity generation by 2030, more than double the amount currently used, although the role AI contributing to the demand, a new study has found.

In its white paper, “Powering Intelligence: Analyzing Artificial Intelligence and Data Center Energy Consumption,” the Electric Power Research Institute (EPRI) noted that one key uncertainty that could change the trajectory of data center load growth is the use of generative AI models.

“A fundamental uncertainty in projecting data center load growth comes from the broad emergence of AI technologies in business and daily life, punctuated by the explosion into public consciousness of generative AI models, such as OpenAI’s ChatGPT,” the study said. “While AI applications are estimated to use only 10%-20% of data center electricity today, that percentage is growing rapidly.”

That growth could have an outsize effect on electricity consumption, because AI can use far more energy to perform a task than non-AI algorithms.

“Early applications were estimated to require about 10 times the electricity — from 0.3 watt-hours for a traditional Google search to 2.9 watt-hours for a ChatGPT query — to respond to user queries,” EPRI researchers wrote in the study. “Evidence about how widely these tools will be used and how much they will change computational needs is just starting to emerge.”

AI was also one of the five factors most likely to influence data center energy consumption the Uptime Institute identified in a study published in January. And one company that makes power and cooling systems for data centers warned in October that existing technologies may not be able to cope with AI’s demands.

EPRI’s findings in its latest report, which is based on public information about existing data centers, estimates of industry growth, and private electricity demand forecasts by industry experts, could force organizations to re-evaluate electricity costs in the face of regional supply challenges.

The organization also looked at those regional impacts of increased AI power demand. In the US, the researchers wrote, “Fifteen states accounted for an estimated 80% of the national data center load in 2023,” with Virginia, Texas, and California leading the pack.

Demands for highly reliable power, requests for power from new generation sources without carbon emissions, and short lead times for connection of two years or less can create local and regional electricity supply challenges, it said.

Overcoming the challenges

To overcome these challenges and allow for data center expansion, the report suggests three essential strategies: Improve data center efficiency and flexibility; coordinate closely between data center developers and utilities regarding power needs, timing, flexibility, and delivery constraints; and develop better modeling tools to anticipate and accommodate data center growth without affecting grid reliability.

Based on its research, EPRI outlined four scenarios for the development of US data center electricity consumption through 2030, with annual growth rates ranging from 3.7% to 15%.

These were based on the assumption of limited public uptake of AI tools coupled with high gains in data center efficiency at one end of the spectrum and rapid expansion of AI applications with fewer efficiency gains at the other.

“The U.S. electricity sector is working hard to meet the growing demands of data centers, transportation electrification, crypto-mining, and industrial onshoring, while balancing decarbonization efforts,” David Porter, vice president of electrification and sustainable strategy at EPRI, said in a statement.

The data center boom, he added, “requires closer collaboration between large data center owners and developers, utilities, government, and other stakeholders to ensure that we can power the needs of AI while maintaining reliable, affordable power to all customers.”

The US is not the only country with concerns about its ability to respond to growth in demand for data center capacity: recently, Singapore announced a plan for more, greener, data centers too.