AI’s Hidden Energy Bill: What a Single ChatGPT Prompt Really Costs – Digital Information World
From writing emails to answering questions and organizing data, generative AI has quickly become part of daily digital routines. But while these tools make work faster and more convenient, they also come with an unseen cost. Every time someone types a prompt into ChatGPT, energy is used to process the request, and with it comes a carbon footprint that’s easy to overlook.
Recent Surfshark research suggests that each ChatGPT query uses about 2 watt-hours of energy. That’s roughly the same power it takes to run a 10-watt LED bulb for 12 minutes or charge a smartphone with a 5-watt adapter for 24 minutes. Put another way, sending one query consumes as much energy as running a 1000-watt microwave for about seven seconds. Heating up a typical lunch takes three minutes, so you’d spend the same energy by making 26 chatbot queries.
The impact grows quickly when scaled. If every person in the United States sent just one query, the combined energy use would reach nearly 685 megawatt-hours. That’s enough to power 63 average homes for a full year, based on national residential electricity use.
Environmental concerns don’t stop there. Each prompt is estimated to produce 4.32 grams of carbon dioxide. That figure may seem small, but when multiplied by millions of users, the emissions become significant. A single day where every American interacts once with ChatGPT would generate roughly 1,479 metric tons of CO₂, about the same as what 322 gasoline cars emit in a year, or as much carbon as 1,500 people flying roundtrip between London and New York.
As global usage continues to grow, the pressure to optimize these models increases. By 2025, the number of generative AI users worldwide is projected to reach nearly 378 million, with 65 million new users added in just one year. That jump marks the fastest growth to date.
There’s still no single answer on exactly how much energy a chatbot query requires. Some studies suggest it could be as low as 0.3 watt-hours, particularly for newer, more efficient models. Others report figures closer to 3 watt-hours, especially with older or more complex systems. These differences reflect both advances in AI infrastructure and the challenges in measuring energy use directly, as most estimates rely on modeling, not public data from tech companies.
For context, Google Search uses about 0.3 watt-hours per query, making ChatGPT nearly seven times more power-hungry by comparison.
The carbon footprint depends on where the electricity comes from. Data centers still draw heavily from fossil fuels in many regions, which explains why a single chatbot query can carry a CO₂ price tag. Depending on local grid intensity, emissions per prompt can vary widely, from under a gram to over nine.
Researchers used standardized appliance ratings to create relatable comparisons. For example, they measured a 10-watt light, a 5-watt phone charger, a 100-watt TV, and a 1000-watt microwave, then calculated how long each could run using the same energy as one AI query. They also matched the U.S. population against national energy consumption figures to estimate how many homes could be powered if that same energy were used elsewhere.
None of this means people should stop using AI altogether. But as adoption accelerates, so does the need to make these systems cleaner and more efficient. Behind every AI-generated answer, there’s an environmental tab still being calculated.
