How will AI affect energy demand — and your utility bill — in California? – San Diego Union-Tribune

Artificial intelligence may revolutionize practically every facet of the economy in the coming years. It’s also expected to lead to a spike in energy demand. But determining how much the AI boom will affect California ratepayers is still cloudy.

“AI is something that stretches across and will impact every single sector of the global economy and pretty much every single industry,” Kevin Dennean, technology sector equity analyst for UBS, said after talking to investors in La Jolla earlier this summer. “I’m really hard-pressed to think of a single sector where we won’t see AI and generative AI put to work.”

The implications for the energy landscape abound.

On the positive side, artificial intelligence promises to enhance productivity, improve energy efficiency and spur the development of alternative power sources. But it also raises concerns. Will AI’s appetite for electricity add another layer of stress — and ratepayer expense — to a grid that’s already under stress?

Analysts at the International Energy Agency have called AI and energy “the new power couple.”

What is AI?

The term “artificial intelligence” dates to the 1950s but it has taken on greater significance with the development of what’s called Generative AI that produces human-like content such as text, images, video, code and even music.

ChatGPT, developed by San Francisco tech company OpenAI, launched on Nov. 30, 2022, and caused a sensation by producing coherent responses to questions or prompts from a user’s computer by processing gigabytes of data, scanning billions of words and using models that adaptively learn over time.

ChatGPT attracted 100 million users in its first two months, making it the fastest-growing consumer application in history.

Other organizations and companies have developed their own models, systems and chatbots, creating a new wave of technology to usher AI into the mainstream.

“We will look back at this date (of ChatGPT’s launch) as the watershed moment for a new innovation cycle,” UBS analysts said in a 104-page report on AI released in June, “one that may prove to be the biggest and most profound in human history.”

But AI requires data centers to carry out that work — and those data centers need power to keep them running.

The International Energy Agency estimates that an internet search with AI uses as much as 10 times the amount of electricity as a traditional Google search.

Similarly, data centers in some cases require three to eight times the amount of electricity to operate as conventional data centers.  

Analysts from Wells Fargo see U.S. electricity demand rising as much as 20 percent by 2030, in part driven by AI.

By themselves, the roughly 2,700 data centers operated largely by tech firms such as Google, Microsoft, Meta and Apple consumed a little more than 4 percent of all electricity in the U.S. in 2022. The Electric Power Research Institute estimates that will more than double to 9 percent by 2030.

Others predict more modest increases.

Whatever the percentages, a burgeoning AI sector means more data centers. What does that mean for power supplies to the electric grid — and for utilities and their customers’ monthly bills?

Any impact yet in California?

Even before ChatGPT made headlines, California policymakers had already invested time and ratepayer dollars into the state’s goal of deriving 100 percent of the state’s electricity from carbon-free sources by 2045, if not sooner.

In a similar vein, Gov. Gavin Newsom four years ago issued an executive order mandating the elimination of all new sales of gasoline-powered passenger vehicles in California by 2035.

If data centers suck up even more electricity, will there be enough juice to meet demand? And what would that mean for residential utility bills that have swollen as much as 110 percent in the past decade?

The California Energy Commission says it’s looking into it.

“The short answer is that load growth — from EVs, data centers, AI, etc. — is accounted for and projected through the CEC’s demand forecast process,” the commission said in an email to the Union-Tribune.

The demand forecast is part of the Integrated Energy Policy Report the CEC produces every two years and updates every other year.

Planning forecasts are also used by the California Independent Systems Operator, which manages the electric grid for about 80 percent of the state; and the California Public Utilities Commission that oversees investor-owned utilities, such as San Diego Gas & Electric, Pacific Gas & Electric and Southern California Edison.

All three power companies also submit load forecasts for their respective service territories to state regulators.

Asked specifically about potential rises in demand because of AI and data centers, staffers said the commission is collecting additional data from utilities and other sources “to improve our forecasts of load growth.”

A scoping document for this year’s Integrated Energy Policy Report will call out data centers for particular study.

At the same time, though, the CEC said “there are a number of challenges involved in forecast data center construction” and what that would mean for load growth.

One factor to consider is electricity prices.

Ironically, California’s high-price environment for electricity may provide a buffer. At least at this early stage of AI growth, data centers and their construction are centered in other parts of the country where the cost of energy is lower than the Golden State’s:

According to the U.S. Energy Information Administration, the average price of electricity for industrial customers in Virginia, Texas and Georgia were each below 9 cents per kilowatt-hour in June 2024. In California, it came to 22.54 cents.

Gary Ackerman, a utilities and energy consultant with more than four decades of experience in power issues, sees trouble ahead.

“If data centers only were about electric energy from any source then the problem would be much reduced. However, that’s not the case,” Ackerman said. “Data center owners want ‘all clean’ energy sources, pitting their interests against all other consumers wanting the same thing. There will be a scramble for clean energy, thereby driving up renewable energy prices and all customer bills.”

Other issues

To feed AI’s hunger for electricity, some tech companies are signing their own deals to ensure they have access to big, reliable sources of energy.

Amazon Web Services recently signed a long-term agreement with a power company in Pennsylvania that will draw 960 megawatts of electricity directly from the Susquehanna nuclear  plant to power what’s called a “hyperscale” data center that Amazon plans to build.

Recently, the Wall Street Journal reported a development firm paid $136 million for a 2,100-acre site outside Phoenix that the company plans to turn into a massive data center complex.

Natural gas producers think their product can supply a good chunk of the demand that AI businesses require.

“This is a major shift for our country, and it’s a major shift for the natural gas market to be able to keep up with this,” Alan Armstrong, CEO of natural gas giant Williams, told analysts earlier this year.

But using fossil fuels to help meet AI load in California raises alarm bells from environmental groups.

So does the fact that energy demand from AI and data centers has increased greenhouse gas emissions at some tech companies.

The Graphic Processing Units (or GPUs) used for large language models at AI sites such as ChatGPT generate more heat and need more energy than the typical processing units used in  personal computers.

Google, for example, reported 48 percent higher emissions last year compared to 2019.

“We cannot meet our statewide climate and environmental goals if we keep those (natural gas) plants online,” said Shana Lazerow, legal department co-director at Communities for a Better Environment. “Many of the data centers have backup generation powered by natural gas or diesel, or some combination of the two.”

Are worries about AI just an overreaction?  

Some think the effects of an expected AI boom may be exaggerated.