👣 From FLOPs to Footprints: The Resource Cost of AI

To understand both the computational demands and environmental costs of AI development, calculate the GPU requirements and material resources needed to train your AI model. Our tool provides two modes:

Training Estimation: Input your model parameters to estimate compute requirements, assuming sequential training to provide a baseline for cumulative compute requirements
Direct Hardware Analysis: Enter a GPU count (x) to instantly see the material footprint of x NVIDIA A100s.

Learn more about the methodology in our research paper.

🧮 Calculation Mode

Input Method

Choose how to specify your calculation

Model Configuration Direct GPU Count

Model Architecture

Select the type of model you want to analyze

Dense Transformer Mixture-of-Experts (MoE)

⚙️ Input Parameters

N - Model Parameters (Total / Active)

Dense: Total parameters | MoE: Active parameters during forward pass

D - Dataset Size (Tokens)

Use T (Trillion), B (Billion), M (Million) or direct number

GPU Lifespan (years)

Expected GPU operational lifetime

1 10

MFU - Model FLOPs Utilization

Efficiency factor (0.20 - 0.60)

0.2 0.6

Visualization

👣 Material Footprint per Element

📝 Formulas (Dense Transformer):

Annual Computational Throughput = (312 × 10¹²) × (365 × 24 × 60 × 60)
Compute Budget = 6 × N × D
Required GPUs = Compute Budget / (Annual Throughput × Lifespan)
Scaling Factor = 1 / MFU
GPU Adjusted = Required GPUs × Scaling Factor
Material Footprint = kg_per_gpu × GPU Adjusted

Note: For Mixture-of-Experts (MoE) models, N in formulas represents active parameters during the forward pass.

Direct GPU Count Mode: Material Footprint = kg_per_gpu × Number of GPUs

Cite as (BibTeX)

@misc{falk2025flopsfootprintsresourcecost,
  title={From FLOPs to Footprints: The Resource Cost of Artificial Intelligence},
  author={Sophia Falk and Nicholas Kluge Corrêa and Sasha Luccioni and Lisa Biber-Freudenberger and Aimee van Wynsberghe},
  year={2025},
  eprint={2512.04142},
  archivePrefix={arXiv},
  primaryClass={cs.CY},
  url={https://arxiv.org/abs/2512.04142}
}