Are AI and data centres worse for the climate than meat?

No, far from it. Livestock emits more than ten times as much as the world's data centres, even if data centres grow as fast as they possibly could. The figures do not measure quite the same thing, though.

The world's data centres used about 415 TWh of electricity in 2024, about 1.5 percent of global electricity use, and this is expected to more than double to about 945 TWh by 20301. AI is the biggest reason for the growth2. Even so, data centres account for less than 1 percent of global CO2 emissions in 20303. Emissions from their electricity peak at between 215 and 320 million tonnes of CO2, and at about 475 million tonnes in the scenario where they grow the most and use the most fossil power4.

Livestock emitted 6.2 billion tonnes of CO2 equivalent in 2015, about 12 percent of all human greenhouse gas emissions5. Meat accounts for 67 percent of that6.

Two caveats. The data centre figures count only CO2 from the electricity they use, not the making of chips and buildings. The livestock figure counts the whole chain, including feed and deforestation, and converts methane and nitrous oxide into CO2 equivalents. The gap is large enough that this does not close it.

A single query uses little electricity. Google has measured that a typical text prompt to its AI Gemini uses 0.24 Wh, less than nine seconds of television7. That is Google's own measurement, and its emissions are counted with Google's purchases of clean power subtracted8. In an independent experiment, AI that writes text used about 0.05 kWh per 1,000 queries. Images take far more: the least efficient image model used about half a phone charge per image910.