Utility bills are getting cheaper, but AI could spoil the party

• Utility bills have been shrinking after years of increases, Bank of America says.
• But the firm says electricity costs have jumped with power demand quickly rising.

Shrinking utility bills have given consumers some room to breathe, but the relief could be cut short by the massive power demands of artificial intelligence, Bank of America reported Tuesday.

In a new note, the bank cited that the average year-over-year median payment fell by 1.4% in the three months through May. It's a welcomed drop for consumers, who have faced climbing costs in the preceding two years.

While prices still remain well above 2019 levels, negative price growth is nothing to shrug at, the analysts wrote: on average, utility payments make up around 8% of income in households making less than $4,000 a month, according to internal data.

But signs are already showing that the reprieve won't be long-lasting.

Although payments have declined, electricity prices — the chief component driving utility costs — have been rising quickly, BofA noted. Through May, the year-over-year electricity rate reached 5.9%, a jump from January's 3.8.

The upswing is on account of increased power demand, fueled by a range of diverse sources.

In part, a hotter climate has pushed up electricity consumption, as households turn up their air conditions or make greater use of heat pumps.

Legislation, such as the Inflation Reduction Act and CHIPS Act, have also contributed. Not only have they bolstered the prevelance of electric vehicles, power-straining domestic manufacturing has jumped as a result, the bank said.

"Given the long timelines in building large plants there is probably much more to come," it predicted.

But AI is the front-and-center factor that will drive utility costs up over a potentially multi-decade horizon, BofA noted.

While the technology's capabilities have captivated Wall Street in recent years, making use of AI is incredibly power-sapping, the bank highlighted:

"For example, BofA Global Research shows that classifying text using AI uses around 2 kilowatt hours per one thousand tasks, but generating images using AI takes almost 3,000 kilowatt hours for the same number of tasks," analysts wrote. "For reference, the average US household uses around 886 kilowatt hours (kWh) per month."

What's more, the semiconductor chips that operate the AI software run much hotter than their predecessors, and require liquid cooling — a system that also drags on electrical power. Added is the rising expansion of cloud servers, which also help handle AI services.

By 2026, BofA expects that an additional 18 to 28 gigawatts will needed of electric capacity. Higher demand cloud servers has already caused a upswing in commercial electricity demand, such as in Virginia and Texas.

Both states are major hubs for the data centers that run the technology.

"While the gyrations of global commodity prices, especially for fossil fuels, will likely continue to impact the price consumers pay for energy, the need for extra capacity in the electricity generation system may well act as a headwind to any prolonged drop in their utility bills," the bank said.

For investors, the rising power demand has drawn a spotlight on utility stocks, projected to be the AI boom's actual winners.