The Growing Energy Footprint of Artificial Intelligence

Axios

AI helps spur rise in Google's emissions

1,920 × 1,080

AI helps spur rise in Google's emissions

Visit

Google is bullish about AI helping to fight global warming, but it's also candid about the energy-thirsty tech driving up the company's own emissions for now.

Why it matters: This climate yin-yang is on display in the tech giant's latest environmental report.

Google's corporate emissions rose another 13% last year and are up 48% compared to their 2019 baseline.

That's partly because data centers serving AI and other applications are using more power.

State of play: Last year's CO2 growth reflects the "challenge of reducing emissions while compute intensity increases and we grow our technical infrastructure investment to support this AI transition," the report states.

But it also highlights ways they're moving to make AI infrastructure far more efficient.

And outside their own operations, the report touts Google's AI products that cut emissions, such as tools that cities use to improve traffic.

What's next: Google faces a tough climb to reach its 2030 net-zero goal.

The Growing Energy Footprint of Artificial Intelligence

Alex de Vries Cell

October 11, 2023

AI servers could use 0.5% of the world’s electrical generation by 2027. For context, data centers currently use around 1% of global electrical generation.

In recent years, data center electricity consumption has accounted for a relatively stable 1% of global electricity use, excluding cryptocurrency mining.

There is increasing apprehension that the computational resources necessary to develop and maintain AI models and applications could cause a surge in data centers' contribution to global electricity consumption.

Alphabet’s chairman indicated in February 2023 that interacting with an LLM could “likely cost 10 times more than a standard keyword search."
a standard Google search reportedly uses 0.3 Wh of electricity, this suggests an electricity consumption of approximately 3 Wh per LLM interaction.
This figure aligns with SemiAnalysis’ assessment of ChatGPT’s operating costs in early 2023, which estimated that ChatGPT [requires] 2.9 Wh per request.