About the Project

Motivation

I started thinking about this project after constantly hearing that AI is “the future.” Every week there seemed to be another headline about new data centers being built, chips selling out, or companies racing to scale larger models. I read a Business Insider article about the rapid expansion of AI infrastructure and how energy demand from data centers is rising sharply.

That made me pause and ask something more specific: what does this actually mean environmentally?

We talk about AI in terms of breakthroughs and productivity gains, but rarely in terms of its marginal environmental cost. I wanted to understand something concrete: what is the carbon impact of a single AI task? Not just large training runs, but inference. A prompt. A response. One workload running in one region at one time.

At its core, electricity-related emissions follow a simple relationship:

$$CO_2 = \text{Energy (kWh)} \times \text{Carbon Intensity (gCO}_2\text{/kWh)}$$

But carbon intensity is not constant. It fluctuates hourly and varies by region depending on the generation mix — coal, gas, wind, solar, hydro. That means identical workloads can produce different emissions depending on when and where they run.

$$CO_{2,A} \neq CO_{2,B}$$

even if the energy consumption is the same.

The more I looked into this, the more I realized that timing and geography are under-discussed climate levers. Research suggests that shifting when electricity is used could reduce roughly 3 gigatons of CO₂ annually. Optimizing where renewables are built could reduce about 5 gigatons. Improving procurement decisions could reduce around 1 gigaton. Together, these operational changes represent over 9 gigatons of potential annual reductions.

What stood out to me is that these are not futuristic inventions. They are optimization problems.

Carbon IQ is my attempt to connect those ideas directly to AI workloads. Instead of talking abstractly about energy use, the project quantifies emissions per 1,000 tokens, visualizes carbon intensity across cloud regions, and allows users to compare deployment scenarios. The goal is simple: make the environmental cost of AI measurable at the task level, and show how operational decisions can reduce it.

Built With

• html
• javascript
• python