Let’s start the year by reflecting on the colossal growth in the resource consumption of AI and what it means for 2026.

At the start of 2025, the worldwide power demand of AI systems was approximately 9.4 GW. By the end of the year, this demand had likely increased to around 23 GW. This level of electricity consumption is comparable to that of Bitcoin mining and is also approaching half of total global data center electricity consumption (excluding crypto mining) in 2024.

To put this into further perspective, it is nearing the amount of electricity required by a country such as the United Kingdom.

This power demand translates into a carbon footprint of between 32.6 and 79.7 million tons of CO₂ emissions in 2025, while the associated water footprint could reach 312.5–764.6 billion liters.

Operating AI systems worldwide may therefore result in a carbon footprint comparable to that of New York City in 2025 while the water footprint could fall within the range of global annual bottled water consumption.

For 2026, the upper end of these ranges effectively marks the starting point for the current year. How this will evolve remains uncertain, with the most pressing question being whether technology companies will be able to secure sufficient power to meet their rapidly growing demand.

What does seem certain, however, is that data center operators will continue to shift the full cost of their growing negative externalities onto society.

My research shows that the sector as a whole resists accountability and withholds critical information needed to assess the true impacts of (generative) AI. The balance between private gains and public costs is profoundly skewed, and the industry has little incentive to address this imbalance openly.

Achieving sustainable growth in data centers and AI will therefore require a fundamental rebalancing—one that must begin with transparency.

Links:
Power demand of AI systems: https://doi.org/10.1016/j.joule.2025.101961
Carbon and water footprint of AI: https://doi.org/10.1016/j.patter.2025.101430

@digiconomist But the 23 GW are still only 0.0046% of the overall datacenter usage?
Is the growth rate here the issue or the amount?

https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai

(would love to read your papers, but sadly they are behind a paywall)

Energy demand from AI – Energy and AI – Analysis - IEA

Energy and AI - Analysis and key findings. A report by the International Energy Agency.

IEA
@kolaente @digiconomist
I think you're mixing up your units (watts vs watt hours). 23 GW over the course of a year is about 201.5 TWh. The article you link says global datacenter power use 415 TWh in 2024. So that makes the 23 GW figure just shy of 50%