Three key uncertainties about the energy consumption of artificial intelligence (AI) systems remain. Earlier this year, researchers, including Casey Crownhart and James O’Donnell, focused on quantifying the energy used by major AI models, such as ChatGPT and Gemini, to generate a single response. Despite multiple inquiries, companies like Google, OpenAI, and Microsoft did not disclose this information initially. However, after the publication of their findings this summer, these companies began to release the data previously sought by researchers.
This newfound transparency raises questions about whether researchers have fully addressed the energy burden of AI. Further discussions with both existing and new sources indicate that significant uncertainties still persist regarding the overall energy consumption associated with AI systems.
Additionally, there is ongoing ambiguity about the inner workings of AI models. Understanding the mechanisms behind AI’s effectiveness is crucial, particularly when deploying these systems in sensitive areas such as healthcare. Researchers at Google DeepMind are exploring a field known as mechanistic interpretability, which aims to provide insights into the underlying processes of AI operations.
As energy consumption and interpretability are critical in assessing AI’s impact, continued investigation is necessary to evaluate the broader implications of these technologies.
Source: https://www.technologyreview.com/2025/09/10/1123489/the-download-ais-energy-future/