The AI boom is becoming an energy story as much as a technology story. In the US, fast-growing demand from data centers, crypto-mining and cloud services is reportedly causing some states to slow down plans to retire coal-fired power plants.
The issue is not only how much electricity new technologies use, but how quickly that demand is arriving. Operators are looking at the grid and seeing a harder balancing act: close older coal capacity as planned, or keep more backup available while power use rises.
AI is changing the timing of coal closures
According to the Financial Times, some US states are delaying plans to close coal-fired power plants because electricity demand from new technologies is so large. AI is a major driver, but the pressure also includes crypto-mining and cloud services.
That does not mean every delayed plant will necessarily run more often. Seth Feaster of the Institute for Energy Economics and Financial Analysis notes that simply moving a shutdown date does not mean the plants will be used. A coal plant can remain available on paper while actual generation still depends on market needs, grid conditions and operator decisions.
Still, the direction is important. By the late 2020s, 54 gigawatts of coal are expected to go offline. That equals about 4% of the country's total electricity capacity, and it is 40% less than predicted last year.
In plain terms, the coal phase-out is still moving forward, but more slowly than earlier expectations suggested. The growth of AI power demand is one reason the timetable is becoming harder to hold.
Data centers are becoming a bigger load on the grid
Generative AI requires large amounts of computing power to train and run models. That computing happens in data centers, which need electricity not only for processors but also for the surrounding infrastructure that keeps those systems operating.
A study by the Electric Power Research Institute says power demand for data centers could double by 2030. If that happens, data centers could account for about 9% of total U.S. electricity demand.
The comparison with ordinary internet use shows why the shift matters. The International Energy Agency estimates that ChatGPT alone could consume nearly 10 times as much electricity as Google searches. That figure puts a familiar interface into a larger infrastructure context: each query depends on a chain of computing resources that must be powered somewhere.
For grid operators, the concern is stability. Electricity systems have to match supply and demand continuously. When demand rises from large, power-hungry facilities, operators may become more cautious about retiring existing power plants, even plants tied to an older energy source.
Coal is down, but not out of the picture
The pressure to keep coal capacity available comes after a long decline in coal's role in U.S. electricity. According to the US Energy Information Administration, coal's share of U.S. electricity fell from 40% in 2014 to 16% in 2023.
That drop shows the broader shift away from coal is real. The current issue is whether the pace of that shift can continue as AI and other digital services increase electricity use.
The situation creates a practical tension:
- Technology companies need more computing power to build and operate generative AI systems.
- Data centers need more electricity as AI, cloud services and crypto-mining expand.
- Grid operators are focused on reliability and may prefer to keep more capacity available.
- Coal retirements may still happen, but some shutdown schedules are being delayed.
This does not overturn the decline of coal by itself. But it does show how AI can affect energy planning beyond the technology sector. The growth of digital infrastructure can change decisions about power plants, grid capacity and the speed of retirement plans.
Water and energy are part of the AI cost
The energy demands of AI are not limited to electricity generation. Data centers for generative AI systems also need large amounts of fresh water, according to Kate Crawford, a researcher at USC Annenberg and Microsoft Research. The water is used to cool processors and to make electricity.
By 2027, global water use for AI could equal half of the UK's use. That estimate adds another constraint to the AI build-out: the systems require physical resources, not just chips and software.
This matters because the public conversation around AI often focuses on model capability, product features and competition between companies. The energy and water requirements are less visible, but they shape what can be built, where it can be built and what other infrastructure must support it.
OpenAI CEO Sam Altman has acknowledged the scale of the energy challenge. He expects more capable AI models to require huge amounts of energy and a breakthrough in energy production. He sees fusion energy, specifically nuclear fusion, as a possible answer.
The next AI bottleneck may be infrastructure
The source article points to a clear conclusion: AI growth is not only a software race. It is also an electricity demand problem, a grid stability problem and, in some cases, a water use problem.
Some coal plants may stay available longer because operators are trying to manage rising demand from AI, crypto-mining and cloud services. At the same time, coal's share of U.S. electricity has already fallen significantly, and a delayed shutdown date does not guarantee heavy use.
The central question is whether the US power system can expand and adapt quickly enough while still retiring older coal capacity. For now, the AI boom is making that transition more complicated.