The Hidden Cost of Smarter Models: AI Energy Appetite
As AI models grow more capable, their computational demands raise uncomfortable questions about sustainability, resource allocation, and who can actually afford frontier AI.
The Compute Explosion
Training a frontier AI model in 2026 requires more electricity than some small countries consume in a year. Data centres are being built at unprecedented scale.
This is not just a corporate expense -- it is a claim on shared resources: electricity, water, land, and rare minerals.
The Equity Dimension
Only a handful of companies can afford frontier models. This concentration means a few corporations determine what AI can do, what languages it speaks, and whose values it reflects.
Countries in the Global South have the least access to AI capabilities that could help them adapt to climate change.
Democratic Questions
Should AI compute be critical infrastructure subject to democratic oversight? Should companies disclose full environmental costs? Should international frameworks ensure equitable access?