Important Caveats

• AI-specific data is estimated. Data center operators do not publicly distinguish AI from non-AI workloads. The AI share figures are derived from IEA modeling and academic research, not direct measurement.
• Company reports use different methodologies. Scope boundaries, carbon accounting methods, and water measurement vary between providers. Direct comparisons should be interpreted cautiously.
• Per-query figures are from Google only. The 0.24 Wh/query figure is specific to Google Gemini on Google infrastructure. Other providers and models will differ.
• Projections are scenarios, not predictions. IEA 2030 projections represent modeled scenarios under stated policies, not forecasts.
• Model efficiency data is approximate. The Hugging Face AI Energy Score measures GPU energy only, not full-stack (networking, cooling, storage).
• Some model entries are normalized estimates. Where a provider does not publish directly comparable benchmark numbers, the dashboard treats those values as directional rather than authoritative.
• The dashboard emphasizes resource-use context. It is intended to help interpret operational scale, not to measure social value, economic benefit, or downstream productivity effects from AI.