Unit Definition
kgCO₂e stands for kilograms of carbon dioxide equivalent. It is a standardized unit of measurement for expressing the warming effect of multiple greenhouse gases (such as methane or nitrous oxide). This unit expresses how many kilograms of carbon dioxide emissions would warm the climate equally as 1 kg of all other greenhouse gases, over a certain period of time (typically over 100 years). CO₂e allows us to express the effect of different greenhouse gases as a singular climate footprint measure.
Disclaimer
This tool estimates greenhouse gas (GHG) emissions using a compute-based, usage-driven methodology aligned with the Cloud Carbon Footprint (CCF) open-source framework and the GHG Protocol for Scope 2 emissions.
Key assumptions include:
- Idle–max power modeling is used where utilization is specified: resources consume non-zero power even when idle, and power scales linearly between idle and maximum load based on average utilization. CPU energy consumption is estimated using industry-average power envelopes (idle and peak) per virtual CPU, scaled by average utilization. GPU energy use is modeled using vendor-published power ranges for modern data center accelerators, adjusted for workload intensity. Sources for average vCPU and GPU power consumption include Energy chart, Home Server Power Consumption: Cheaper 24/7 Hosting, NVIDIA Tesla T4 AI Inferencing GPU Benchmarks and Review – Page 5 of 5 – ServeTheHome, NVIDIA A10G Specs | TechPowerUp GPU Database.
- Power Usage Effectiveness (PUE) values are provider-level averages and are applied uniformly across regions where region-specific PUE data is unavailable. They have been used from CCF’s methodology.
- Grid emission factors are region-specific averages representing the carbon intensity of electricity generation and may not reflect real-time energy sourcing or contractual renewable energy claims. They have been used from CCF’s methodology.
- We use grid emission factors for AWS, Azure and GCP published by the Cloud Carbon Footprint (CCF) project, including adjustments for Carbon-Free Energy (CFE%) based on publicly available Google data. As a result, calculated emissions may differ from values reported directly by Google Cloud. Differences arise because CCF applies its own standardized, open-source methodology, including independent data snapshots, region mappings, averaging, rounding, and the application of CFE% as a proportional adjustment rather than Google’s internal hourly matching and accounting methods. Consequently, results should be interpreted as estimates suitable for comparative analysis and internal reporting, not as exact replicas of Google-reported emissions figures. Emission factors and PUE for OVH are self-reported using their methodology OVHcloud Environmental Impact Tracker Methodology.
- Energy consumption values are derived estimates based on peer-reviewed measurements of LLM inference energy, converted to kWh using standard methodologies. Model providers do not publish verified per-token or per-response energy data; actual values vary with model size, hardware, batching, response length, and deployment configuration. These figures are intended for comparative and directional analysis only. The calculations were made based on the assumption that a user would make 10 queries per hour to a chatbot.
- All LLM emission calculations were performed using the energy consumption estimates performed by EcoLogits using their methodology and assumptions.
Additional Disclaimers:
- Pricing: Reflects typical on-demand cloud rates (2024–2025). Actual costs vary by region, traffic patterns, and autoscaling.
- LLM Costs: Assumes ~10M tokens/month (~58,823 queries at 170 tokens/query). Actual usage may differ significantly.
- Emissions: Calculated using CPU (Idle: 5W/vCPU, Max: 15W/vCPU), GPU when applicable (T4: 30-70W, A10G: 20-150W), provider-specific PUE (AWS: 1.135, Azure: 1.185, GCP: 1.10), running 24/7 for 30 days. All LLM emission estimates were done for an average prompt of about 170 output tokens used for “write an email” prompt message.
- Utilization: Light = 1 instance, Medium = 2 instances, Heavy = 3 instances based on selected usage level.
- Important: Use for planning and comparison only – not exact billing or carbon reporting.
Data Sources
