The International Energy Agency (IEA) has just released a major report on the relationship between artificial intelligence and the energy sector. The report projects a dramatic rise in electricity consumption by A.I. and other constituents of the information economy, especially in the United States. This is occurring at a time when electricity prices in the U.S. are rising at rates well in excess of inflation. The prospect of still faster growth of the information economy is, therefore, stoking fears that ever-higher electricity prices are on their way.
Such an eventuality cannot be ruled out, but there are reasons to believe that A.I.’s net impact on the economics of the grid may not be negative. A central conclusion of the IEA report is that in the future, it may be possible to beneficially apply A.I. itself to the electric grid, helping to reduce both demand and costs. That may prove to be the case, but it is important to note that other forces are at work that are already having positive effects today.
In business-as-usual commercial development, architects and engineers design buildings to meet their clients’ needs, with electric hook-ups that are sized to meet anticipated electricity demand—secure in the knowledge that the local utility will deliver all of the electricity that is desired. Data center developers, by contrast, consider at the earliest stage of a project, when center scale and siting are still in play, where their electricity will come from. The result, with increasing frequency, is a non-traditional approach to the question of electricity supply. While many such approaches involve green electricity—and many press releases tout their decarbonizing effects—sustainability is just one of several considerations. Money and time often figure with at least as much prominence.
Three main types of considerations drive industry thinking in this regard. First is the reality that, more often than not, renewables-based generation yields a lower levelized cost of electricity (LCOE) than fossil-based generation. A Lazard report released last year showed that on-shore wind and solar generation beat the least expensive fossil options at the lower end of their respective LCOE ranges. The fact that wind and solar installations can be cost-effective at scales significantly smaller than those of centralized thermal plants encourages their integration with data centers. The need for downscaling also helps explain why data center proprietors are expressing interest in a type of next-generation nuclear energy referred to as “small module reactors.” The IEA report notes that 14 of the 20 largest data center operators (measured by data center power capacity) have clean energy usage that exceeds 50 percent of their total power consumption. Seven of the top 20 are at 100 percent. Amazon (AMZN) (number one on the list) reports that it currently has more than 600 renewable energy projects underway in 28 countries around the world.
Second is the reality that electricity prices in deregulated markets are extremely volatile, from day to day and certainly from year to year. An analysis done by my firm, Carbon-Neutral Consulting, showed that over the course of 2024, monthly energy charges for one of our U.S. clients varied by as much as 50 percent above or below the median level. The sure-fire ways of avoiding the risk of out-year price spikes are 1) entering into a long-term power purchase agreement; or 2) owning the generating assets that will supply your power needs. This circumstance may help explain Microsoft’s announcement last September that it was entering into an electricity-sourcing arrangement that will involve restarting the Three Mile Island nuclear plant. That plant’s generating Unit 2 was infamously shut down after a near meltdown in 1979. Unit 1 remained in service until 2019, when it was shut down because of its high costs. While it’s possible that the reported recommissioning cost of $1.6 billion is a bargain, it is also possible that Microsoft (MSFT) found the offer of relatively expensive but stably priced electricity over 20 years compelling.
Third is the reality that transmission and distribution (T&D) represent a significant proportion of the price of electricity. (Data released from the U.S. Energy Information Administration shows that T&D expense accounted for 38 percent of the average electric bill in 2022.) Data center projects that can avoid sourcing electricity through the grid will therefore bear only the electricity routing costs associated with short-distance movement of electricity over a proprietary system—presumably a much lower number. Avoiding the grid’s T&D system brings another major benefit: it allows data center developers to steer clear of T&D upgrades and interconnection queues that can delay electricity-intensive developments for years.
At the end of the day, there is certainly a risk that the ongoing increase in information economy electricity demand may cause electricity prices to rise to problematic levels. But this appears to be one potential crisis that may never arrive because of the crisis-averting tendencies embedded within it.
