This project uses a quantitative model to evaluate the effectiveness of different economic policies at reducing the environmental impact of AI related data centers. Emissions are modeled using the below framework:
E = Y x C
Variables:
- E = emissions (million metric tons of CO2 per year)
- Y = electricity consumption (in TWh)
- C = Carbon intensity (grams of CO2 per kWh)
The model compares carbon taxes, clean energy subsidies, dynamic energy pricing, clean energy mandates, and the proposed policy combination from the paper.
- Calibrated to the 2028 baseline
- Computes emissions, government costs, firm costs and total cost per ton
- Creates a baseline ranking for each policy for later analysis
- Applies a +/-10% shock to both electricity demand and carbon intensity in the model
- Evaluates the stability of each policy under uncertainty
Computes:
- expected emissions
- downside emissions (95th percentile)
- distribution of cost per ton
- 90% dispersion range
- Policies are evaluated as risk-adjusted decisions
- Compares the proposed Dynamic Energy Pricing + Clean Energy Mandate combination to the single-policy approaches
- Evaluates the tradeoffs of the combination between emission reductions and cost efficiency
- Clean Energy Mandate has strong emission changes but includes highest cost volatility
- Clean Energy Subsidy produces the lowest forecasted emissions
- Carbon Taxes provide moderate emission changes at the best fiscal cost
- Dynamic Energy Pricing + Clean Energy Mandate improves emissions performance relative to other options while keeping costs to government and firms low, preserving innovation incentive
- Policy ranking effectiveness are preserved under the +/- 10% sensitivity shock
data/ Reserved for later calibration inputs
src/ Core emissions and cost metric functions
notebooks/ Structured analytical workflow
outputs/ Generated figures and tables
To reproduce results:
Install dependencies from requirements.txt
Run notebooks in order:
- 01_deterministic_model.ipynb
- 02_sensitivity_analysis.ipynb
- 03_monte_carlo_simulation.ipynb
- 04_policy_portfolio.ipynb
Finn Case