AI Energy Usage Simulator (Bottoms Up)

Explore how AI query demand and model efficiency affect energy consumption over time.

Query Demand

Define how AI usage (queries) grows over time

Preset Scenarios

Growth Model

Initial Demand (billion queries/day)

300.0B

Early CAGR (80%)

80%

Mid CAGR (40%)

40%

Carrying Capacity (8× initial)

8×

Select AI model and set efficiency parameters

AI Model (Predefined List)

Energy: 8.68 Wh per 1k queries

Overhead + PUE Factor (1.6×)

1.6×

Includes idle power overhead and data center PUE. Typical range: 1.2-2.5×

Include Training Energy

Training Energy Ratio (30× inference)

30×

How much more energy-intensive training is compared to inference

Monthly Retraining Rate (30%)

30%

Fraction of models retrained each month (25% = retrain every 4 months)

Include Efficiency Gains

Annual Efficiency Improvement

20%

Years to Project

AI Energy Score = N/A
PUE Factor = N/A
Current Calculated Energy/Query (Wall-plug, incl. PUE, initial efficiency): N/A

Inference Energy (Wh) = (Query Demand / 1000) × AI Energy Score × PUE Factor

(Note: AI Energy Score is in Wh per 1k queries. Query Demand is daily.)

Training Energy = Inference Energy × Training Ratio × Monthly Retraining Rate

Total Energy = Inference Energy + Training Energy

Annual efficiency gains further reduce the actual energy per query over time.

No data available. Adjust parameters to see results.

Projection starts from ChatGPT launch (November 30, 2022)