Google is trying to shed some light on a key question looming over the artificial intelligence boom: Just how much energy does it take?
That issue looms large as predictions of a surge in massive, electricity-hungry AI data centers have spurred a nationwide push to create power plants to avoid blackouts. The Trump administration has also cited AI as a justification for its efforts to keep aging coal plants alive while favoring fossil fuels over renewables.
But Google’s report argues that the power needs of its Gemini AI assistant have fallen a lot thanks to improvements in efficiency. For example, the company says, running a single text prompt on Gemini has become 33 times more efficient in the past year — and now uses less electricity than watching TV for about nine seconds.
The report, which proposes a methodology for measuring the energy, water and greenhouse gas emissions associated with Gemini text searches, drew immediate pushback from skeptics. Importantly, the report covers only a median text prompt, not photos, videos or a particularly complex response. It also doesn’t delve into the energy cost of training its large language model.
