Skip to content
New reporting from MIT Technology Review provides the most detailed examination to date of Artificial Intelligence´s accelerating energy consumption and resulting climate impact. Journalists spent over six months collaborating with top researchers to quantify both the energy required by various types of Artificial Intelligence queries and the corresponding emissions footprint, offering rare data-based insight amid widespread industry secrecy.
The analysis revealed that the energy consumed by Artificial Intelligence is highly variable and depends on factors such as query complexity and model size. For open-source models, researchers could directly measure power use, showing that complex requests, like generating detailed travel itineraries, can use up to ten times more energy than simple queries. Larger models may consume up to seventy times more energy than smaller ones for identical tasks. Moreover, generating images or videos requires exponentially more energy than text responses—sometimes hundreds of times more—underscoring the sprawling footprint of multimedia Artificial Intelligence applications.
Crucially, the climate consequences of this increased energy demand hinge on the energy sources powering the data centers. A query served from a data center running on coal-heavy grids, such as those in West Virginia, could result in nearly double the carbon emissions as one from a cleaner grid, such as California´s. This makes the location and energy sourcing of data centers a critical climate consideration as Artificial Intelligence infrastructure proliferates.
The investigation also highlights significant gaps in knowledge, primarily due to the lack of transparency from leading Artificial Intelligence companies, which declined to share specific energy and emissions data. As Artificial Intelligence becomes increasingly embedded in all aspects of technology—potentially operating behind the scenes of everyday devices and services—these unknowns become more consequential. Transparent reporting and proactive grid management will be essential to navigate the technology´s future climate footprint.
