The AI Footprint Calculator is built on a foundation of scientific research and environmental impact data. This section details the methodological approach, data sources, and calculation frameworks that power the calculator's estimations.

Research-Based Development

All calculations in the AI Footprint Calculator are based on scientific research and published data on AI systems' environmental impacts. The development process began with a thorough review of key research documents:

• Estimating the Energy Footprint of Locally Executed Artificial Intelligence Models
• Environmental Footprint of AI Systems: Data for Calculator Tier Population
• AI Footprint Calculator Research
• AI Footprint Offsetting Research
• Aggregated Environmental Footprint Totals and Actionable Offsetting Guidance

These documents provided the scientific foundation for the calculator's estimation methodologies, ensuring that all calculations are grounded in peer-reviewed research and industry benchmarks.

Calculation Methodologies

Cloud AI Footprint Calculation

The Cloud AI calculator uses a tiered system to categorize different AI models and services:

Key components of this calculation include:

• Model Tier Factors: Energy and water consumption rates for different categories of cloud AI models (e.g., large language models, image generation models)
• Usage Volume: Quantification of AI usage (e.g., number of queries, tokens, or images)
• Regional Grid Factors: Carbon intensity of electricity generation in different geographic regions

Local AI Footprint Calculation

The Local AI calculator estimates the environmental impact of running AI models on personal hardware:

Key components include:

• Hardware Profiles: Power consumption specifications for different GPU and CPU configurations
• Utilization Factor: Percentage of hardware capacity used during model execution
• Duration: Time spent on AI tasks
• Regional Grid Factors: Carbon intensity of local electricity generation

Agentic AI Footprint Calculation

The Agentic AI calculator addresses complex AI systems that orchestrate multiple models:

This calculation accounts for:

• Component Models: The individual AI models used within the workflow
• Orchestration Overhead: Additional computational costs of managing and coordinating multiple models
• Interaction Patterns: How different models are sequenced and combined

Environmental Impact Metrics

The calculator provides estimates across three key environmental impact dimensions:

• Energy Consumption (kWh): Total electricity used by AI operations
• Water Usage (Liters): Water consumed for cooling data centers and power generation
• Carbon Emissions (kg CO₂e): Greenhouse gas emissions expressed as carbon dioxide equivalent

These metrics are calculated using Power Usage Effectiveness (PUE) and Water Usage Effectiveness (WUE) factors that account for the total resource requirements of AI infrastructure beyond direct computation.

Equivalency Generation

To contextualize abstract environmental metrics, the calculator generates equivalencies that translate impacts into everyday terms:

• Energy consumption equivalent in household electricity usage
• Water usage equivalent in drinking water, showers, or dishwasher cycles
• Carbon emissions equivalent in car miles driven or flights taken

These equivalencies help users understand the relative scale of their AI environmental footprint in familiar terms.

Assumptions and Limitations

The calculator acknowledges several important assumptions and limitations:

• Estimates are based on average values and may not reflect specific hardware or data center configurations
• Cloud provider efficiency improvements are not automatically incorporated as they evolve
• Embodied carbon (from manufacturing hardware) is not included in the current calculations
• Some emerging AI modalities have limited public benchmark data available

These limitations are transparently communicated to users to ensure appropriate interpretation of results.

Validation Approach

The calculator's estimates were validated through:

• Comparison with published research on specific model energy consumption
• Cross-referencing with cloud provider sustainability reports where available
• Sensitivity analysis to identify reasonable ranges for uncertain parameters
• Transparent documentation of all data sources and calculation methods

This validation approach ensures that while estimates may contain uncertainty, they provide reasonable and useful approximations of AI environmental impacts.