AI Environmental Unit

AEU
Calculator

Measure and compare the environmental cost of AI usage at organizational scale. One AEU = the cost of 1 million standard inference calls on the average US grid.

? How to use this calculator ▾

No submit button. Every change you make updates the score instantly. Work through the five sections top to bottom using the sliders and buttons.

1
Employees using AI
Slide to the band that matches how many people at the company actively use AI tools like ChatGPT, Microsoft Copilot, or similar. If you're not sure, estimate on the low side.
2
Daily queries per employee
How often does a typical employee use AI in a day? Casual users checking in occasionally: pick 1–5. Knowledge workers using it regularly for writing or research: 6–20. Analysts or engineers relying on it heavily: 20–50.
3
Model type
Most companies using standard business tools (Microsoft Copilot, ChatGPT Team, Claude.ai) are in the "mid-size standard" tier. Pick "large frontier" only if the company runs GPT-4, Claude Opus, or similar high-end models for most work. Pick "multimodal / video" if they generate images or video at scale.
4
Query length
What does a typical request look like? Short questions and quick edits: pick "short." Writing, summarizing, or Q&A: "medium." Uploading documents, long reports, or code reviews: "long." Extended research or large file analysis: "extended."
5
Training, grid, and water
Unless you know the company trains its own AI models, leave training on "inference only." For grid and water, the defaults represent an average US data center. Only change these if you know where the company's AI compute actually runs.
Try this: Run three profiles back to back — a small company (50 employees, light use), a mid-size enterprise (2,500 employees, moderate use), and a large tech company (50K+ employees, heavy use). The gap between them is the story.
Company Scale
Number of employees using AI tools 1,000
1–5051–500500–5K5K–50K50K+
Daily AI queries per employee 20
1–56–2020–5050–100100+
AI Model Profile
Primary model type
Average query length / complexity tokens consumed per call
Includes model training? internal fine-tuning or custom training runs
Infrastructure Context
Data center energy source where compute runs
Water usage context data center cooling efficiency
—
AEU / year
calculating...
Scale position
0101001K10K100K+
Annual queries
—
Compute multiplier
—
Grid multiplier
—
Water multiplier
—
Adjust the inputs above to calculate your AEU score.

Methodology

AEU formula: (Annual Queries / 1,000,000) × Model Multiplier × Query Length Multiplier × Training Multiplier × Grid Multiplier × Water Multiplier

Multipliers are calibrated to published research: GPT-3 inference ~0.001 kWh per query; GPT-4 class ~0.003–0.01 kWh; training runs can equal millions of inference calls. Grid carbon intensity ranges from ~50g CO2/kWh (hydro-heavy) to ~900g (coal). Water usage: 0.5–2L per kWh depending on cooling method and region.

This is an estimation framework, not a certified measurement. Real figures require data center audit, utility bills, and vendor disclosure. AEU is designed to enable comparison, not precision billing.