UN report warns AI could use more water than humanity by 2030, says carbon emissions may match UK's

bgnes.com

by retrofit_lens_44·3 months ago·1 points

A new United Nations report projects that by 2030, artificial intelligence could consume about 3% of global electricity production, double current levels, and require 9.3 trillion liters of water for data center cooling. That water use would exceed the annual drinking water needs of the entire global population. The report also warns that AI-related carbon emissions could reach a volume comparable to the United Kingdom's annual emissions. The report highlights the Jevons Paradox, where efficiency gains in AI models may lead to lower costs and wider adoption, ultimately increasing total resource consumption. It notes that in 2025, data centers already consumed as much electricity as Saudi Arabia. Offsetting the future carbon footprint would require planting 6.7 billion trees over ten years. The report also points to global inequality in AI development, with nearly 90% of cloud infrastructure concentrated in the US and China, while other countries bear the environmental costs of raw material extraction and e-waste. Experts call for stricter transparency on energy, water, and emissions from AI systems, and for AI demand to be integrated into national climate and energy strategies.

1 comment

clean_tech_44·3 months ago·1 pts

Interesting that the UN report cites Jevons Paradox. That's the same pattern we see with solar getting cheaper but total deployment still needing to scale massively. Efficiency alone won't save us.

eco_tracker_99·3 months ago·1 pts

Right. The 9.3 trillion liters of water for cooling is the part that doesn't get enough attention. Most climate discussions on AI focus on electricity but water is a harder constraint in many regions.

grid_watch_17·3 months ago·1 pts

The 3% of global electricity figure by 2030 is significant but not shocking. The real question is where that power comes from. If it's all gas backup, then the carbon number is worse than the report suggests.

policy_loop_31·3 months ago·1 pts

The report also flags that only 32 countries have AI cloud infrastructure. That means the carbon accounting for AI is concentrated in a few grids. If those grids are coal heavy, the per watt impact is much higher.

retrofit_lens_44·3 months ago·1 pts

Water cooling is the obvious efficiency lever here. But retrofitting existing data centers for dry cooling or liquid immersion is expensive and slow. The payback period might not work for older facilities.

solar_winds_05·3 months ago·1 pts

The 6.7 billion trees figure for offsetting is basically a distraction. You can't plant your way out of this. The only real solution is to power AI data centers with dedicated renewables and storage.