Low-Energy Fridays: Should we care about AI’s electricity consumption?

Artificial intelligence (AI) is the new technology buzzword. Last week, the House Committee on Energy and Commerce held a hearing titled “Powering AI: Examining America’s Energy and Technology Future,” which covered important topics like how AI may impact the energy sector, including the energy demands it may create. The important thing to remember when it comes to AI, though, is that like any technological advancement, the market is the best mechanism for ensuring its potential benefits outweigh costs.

One wrinkle for AI is that, like everything, it is not costless, and the cost for AI is particularly manifest in rising energy demand. The International Energy Agency estimated that AI data centers accounted for about 2 percent of global electricity demand in 2022—a number that is sure to rise as AI becomes more commonplace. The computational power it needs makes AI particularly energy-intensive, leading to criticism that the technology ought to be restricted. But this argument falls flat because electricity is not particularly scarce.

In economics, the value of new products is often weighed against the cost of its inputs—especially if those inputs are scarce. As an example, gold is a superior metal for the contacts on electronic cables because it resists corrosion and has good electrical and thermal conductivity. But because gold is expensive, most cables use nickel plating instead. In this case, the market delivers efficiency by enabling producers to weigh the scarcity of a commodity relative to its benefits and utilize substitutes where applicable. It ought to be the same for AI.

We have the means to produce massive amounts of electricity, but for artificial reasons like power plant permitting and interconnection rules, we often can’t build resources where they would be best utilized. For private-sector, energy-intensive activities like AI, electricity consumers can mitigate supply constraints by siting their data centers where power is cheap, operating their facilities flexibly, or producing their own electricity. Such actions mitigate AI’s power infrastructure requirements and can improve total system efficiency, but only in areas with wholesale and retail electricity competition. The price signals of the retail and wholesale markets inform where it is cheapest to produce electricity relative to its retail value, and sellers can then pass that price signal on to electricity customers.

Interestingly, AI can increase energy availability by overcoming inherent inefficiencies in our energy production systems. It can help electricity producers improve siting of renewable resources; help system operators better forecast weather-dependent energy production (e.g., wind power); balance the dispatch of resources in real time to minimize the need for curtailments; and better utilize least-cost electricity generating resources through dynamic line rating (optimizing transmission capacity for various weather and climatic conditions). For these reasons, AI technology is likely to deliver a net benefit to electrical systems in a healthy marketplace.

Ultimately, AI’s benefits are best achieved under market conditions. Producers whose profitability depends upon selling electricity are more likely to be interested in how to use AI effectively, while similarly competitive electricity markets will incentivize AI consumers to be more efficient in their consumption of electricity. Conversely, regulated electricity markets in which large electricity consumers aren’t exposed to the price of electricity will be less concerned about making AI cost-efficient, and such markets would have no profit motivation for using AI to improve the efficiency of those grids.

In a nutshell, competitive markets are good for innovation, and AI is no exception. Energy policy analysts see a bright future for the technology, but that future depends on successful AI innovation. Competitive enterprise has a built-in profit motive for innovation, and competitive electricity markets are better suited to stimulating AI innovation than regulated ones. Conversely, politicians trying to manage regulated monopolies are less effective at stimulating innovation—making the potential benefits of AI in the energy sector likely to accrue in competitive markets first.