Should We Worry About AI Energy Consumption?

At home, to make some toast, you probably used slightly less than 100 BTUs.

Total energy consumption in the United States during 2023 was close to 93.6 quadrillion BTUs (a lot of toast).

Let’s take a look at where that energy comes from and how we use it. Then we can judge AI projections.

U.S. Energy Consumption

We use and waste a lot of energy. One quad–a quadrillion BTUs–is the energy equivalent of approximately eight billion gallons of gasoline.

Annually, we get a new flowchart from the Lawrence Livermore National Laboratory. Showing where our energy comes from and where it goes, the chart is a handy snapshot of how we power our homes, businesses, and vehicles.

We can see where the totals come from by following an energy pathway. Starting on the lefthand side, our energy resources are renewable, nuclear, and fossil. Next, their destination is our homes, our businesses, industry, and transportation. Then, on the far right, rejected and services refer primarily to efficiency. Your car’s gasoline drives the wheels as an “energy service” and comes out of your car’s exhaust as “rejected energy:”

You might find this 7-minute explanation of the flowchart helpful:

AI Energy Consumption

The IEA (International Energy Agency) tells us that “there is no AI without energy–specifically electricity for data centres.” At 1.5% of global energy use last year, AI’s energy needs could more than double in 5 years.

The mind-boggling statistic though is 100,000. The electricity needs of one AI data center equals 100,000 (and maybe 20 times as much) households. So, where will it come from?

Axios graphed their conclusions:

Although we switch from BTUs (from Livermore) to terawatt hours (from Axios), you get the general idea (though not the conversion that is daunting):

Where?

Our Bottom Line: Tradeoffs

As economists, we can always conclude with tradeoffs. Remembering cost is defined as sacrifice, we can ask about the energy “cost” of the spread of AI. Summarizing the 304-page IEA report, Axios focused on the emissions cost between demand from AI generation and AI’s climate benefits from its energy consumption solutions.

Unsure, the IEA said, “The widespread adoption of existing AI applications could lead to emissions reductions that are far larger than emissions from data centers — but also far smaller than what is needed to address climate change…”

Returning to our title, Should We Worry…?

Yogi Berra has the answer.

“It’s tough to make predictions, especially about the future.”

My sources and more: Thanks to Axios for alerting me to the new AI grid report. From there, it made sense to start with the Livermore website and their flowcharts. Then, the perfect complement was this IEA (International Energy Agency) report on AI.

Please note that several of today’s sentences were in a previous econlife post.