AI now consumes as much electricity as Japan. Science says it's more complex than it looks.

Published on June 23, 2026

Artificial intelligence already consumes as much electricity as an entire country like Japan. Science documents precisely why — and why the picture is more nuanced than it appears.

What the research shows

Data centers already account for 2 to 3% of global electricity consumption. Peer-reviewed projections anticipate a rise to 9% by 2030. Three documented mechanisms explain this trajectory:

→ Growing computational demand: computational demand from AI models grows faster than per-operation efficiency gains.

→ The water footprint: routinely overlooked, generative AI consumes approximately 10 ml of freshwater per query, amounting to billions of gallons annually worldwide.

→ The rebound effect: every efficiency improvement is absorbed by increased usage volume — net emissions continue to rise.

The nuance most people miss

AI can also reduce emissions: power grid optimization, predictive maintenance, resource management. The scientific consensus is moderate — not because data is lacking, but because the net balance depends directly on the energy source used. A data center running on renewables changes the equation significantly. What science identifies as a critical gap: the lack of operator transparency on actual energy consumption.

Open questions for research

→ Do the climate benefits of AI applications ultimately offset the costs of their infrastructure?

→ How can the rebound effect be measured at the scale of massively deployed large language models?

→ What regulatory frameworks can enforce transparency on data center energy consumption?

Sources

• Wang et al. (2026). Frontiers in Energy Research.
• Sidorkin (2025). AI-EDU Arxiv.
• Gao & Ayob (2026). Journal of Tourism, Hospitality and Environment Management.
• Hankendi et al. (2025). iScience.

← Back to blog