U.S. data centers are increasingly reliant on coal-fired electricity as expanding AI infrastructure increases energy needs and makes natural gas less affordable, reports The register. However, in states with nuclear power plants, coal, slate and oil are used to smooth out power peaks and dips, not to meet the needs of leading technology giants. Let’s try to understand it.
Training
A nuclear power plant is almost perfectly suited to providing mass power to an AI training data center because its constant, high-capacity output matches the data center’s continuous, predictable demand profile. Well almost. During a training run, checkpoint intervals (when models are saved to storage) cause GPU usage to drop significantly as data is written and synchronized. These fluctuations are minor compared to the total load.
Meanwhile, because the nuclear plant’s output does not vary, grid operators must compensate for sudden drops or spikes in usage, and that’s where coal and renewables come in. Without them, energy flows steadily through high-voltage lines to transformers, which are supposed to maintain stable voltage in homes and industry, but are not intended to handle loads massive.
Essentially, nuclear power provides the power to sustain multi-week workouts without interruption, but when the chips are down, coal and renewables do their job for other consumers.
Inference
Things are different for inference, a step where a trained model is used, generating outputs from new inputs rather than learning from data. AI inference consumes power in short, intense bursts because the system processes large chunks of advanced calculations at unpredictable intervals.
Each request – whether to generate text, recognize an image or pass on a recommendation – triggers thousands of matrix multiplications through the AI accelerators, causing an electrical surge for a fraction of a second. When millions of requests arrive simultaneously, the combined load can increase sharply, straining both local power distribution and cooling systems. Inference creates sharp, rapidly fluctuating energy patterns that must be satisfied by responsive grid supply or on-site energy storage. The electricity grid must be rigid enough to support all types of customers, and this is again where renewables and coal or slate come into play.
As AI data centers drive U.S. electricity demand, coal production has jumped nearly 20%, according to Jefferies as cited by The registeralthough coal’s role is primarily to stabilize the grid rather than directly powering AI clusters.
Summary
Nuclear power plants provide constant output ideal for multi-week AI training, while coal and renewable energy help balance fluctuations caused by checkpoints and other loads. During inference, when millions of user requests arrive unpredictably, the network relies on flexible sources, including fossil fuels, to handle short, abrupt spikes in energy demand that even advanced data centers cannot handle on their own.
Follow Tom’s Hardware on Google NewsOr add us as your favorite sourceto get our latest news, analysis and opinions in your feeds.
