Hirs: AI and Energy—Is There Enough to Go Around?

The wave of smart devices that can answer your questions, provide driving directions and steer you to buy things you need, as well as those you don’t, is behind the current surge in electricity demand. Midline estimates are that the U.S. will need at least 50% more installed electricity capacity by the end of the decade. That’s a lot. These artificial intelligence (AI) data centers need a constant supply of electricity, and the nation’s grids are unprepared.

AI has been the stuff of science fiction and the goal of computer programmers for more than 70 years. Every advance in computer processing power brings humankind closer to the goal.

We began to see it in our daily lives as search engines learned our patterns and directed us to websites aligned with our prior searches. The advent of smart TVs, with microphones that are always listening, brought us personalized advertisements for clothing, vacations and health care. 

That our credit card companies and banks sell our data to their partners means that somewhere in the cloud, there is a data entry for each of us that knows our buying habits, down to when it’s time for an oil change or trip to the dentist. 

This year’s Nobel Prize in chemistry recognized scientists who used advanced techniques, including AI, to understand and design novel proteins with applications for creating new drugs and other materials. DNA sequencing that used to take years has been reduced to minutes.

AI applications for seismic analysis and measurement while drilling, or MWD, are reducing costs for oil and gas companies. These applications of AI are typically driven by a single user. What happens when millions ask ChatGPT or some other generative AI to write a term paper that is due at 5 p.m. tomorrow? 

In his 1982 book “Critical Path,” Buckminster Fuller calculated that humankind’s knowledge doubled every century up to 1900, speeding up to every 25 years by the end of World War II. IBM predicted that knowledge would double every 12 hours by 2020.

Regardless of whether it is 12 weeks, 12 days, 12 hours or 12 minutes, the avalanche is more than individuals can manage. And presumably, these forecasters were speaking about scientific facts and verifiable studies. They couldn’t have anticipated the “alternative facts” and deliberate disinformation now spreading across the Internet.

This exposes a critical weakness of the technology. A preponderance of misinformation is often interpreted as valid by these generative AIs.

Examples abound. Term papers written with footnotes citing studies from Prague in 1957 coupled with articles from the New York Times in 1963 and the Wall Street Journal from 1974 applied to oil market conditions in 2024. It even shows up in investment analysts’ newsletters that repeat myth after myth to justify their current investment strategies. Shame! 

As Mark Twain said, “Nothing spoils a good story like the arrival of an eyewitness.” But these AIs are not yet eyewitnesses and are only beginning to gain the ability to sort out what is true and what is false. Until they can accurately discern the truth, the computing power needed to sort through fact and fiction will increase at an even higher exponential rate.

Think back to your school days. Your teacher could easily handle one or two inquiries at a time but would be overwhelmed by more than a dozen. To handle millions of simultaneous queries, the internet repositories of all this information will need to be replicated many times in addition to the new algorithms necessary to manage our simultaneous queries.  

No one can anticipate accurately how much electricity will be required by the coming AI centers, but those operating with subscription services have the advantage because they already have the cash flow necessary to construct their own power supplies.

Microsoft has contracted to restart Unit 1 at Three Mile Island, and Google has contracted to buy small modular reactors for its AI needs. Apple, Netflix, Amazon and others are sure to follow.

Those AI centers that have to rely upon grid operators to admit them will be handicapped because electricity markets such as those in California, Texas and the Pennsylvania-Jersey-Maryland region don’t reward capacity growth, meaning those markets will scramble to provide enough electricity to power the insatiable demands of AI.

Stephen Hawking was concerned that AI could lead to the collapse of civilization if it was not used as a helpful tool or if it somehow developed into the robotic scourge of science fiction. But before AI destroys civilization, Hawking’s prediction that humankind’s ever increasing use of electricity “will turn earth into a giant ball of fire” could happen first. I’ll have to ask Alexa about that.