Technology
AI Carbon Estimate
Estimate energy and CO2e for AI inference calls with your own assumptions: number of calls, energy for each call, grid carbon intensity and PUE.
Add to workspaceYour numbers
Result dey update as you type; no Calculate button dey. All calculation dey run inside your browser.
Energy for each call and grid carbon intensity here na example inputs. Nobody measure dem for any particular model or region. Replace dem with your own measured figures or figures wey provider give before you rely on the result.
Emissions wey the model estimate
—
—
How the result break down
| Di item | Value wey e get |
|---|
Extra facility energy na total energy minus inference energy. If PUE na 1.0, that extra energy na zero.
Quick look at the result
We scale the bars against the larger energy amount. Dem show the extra energy wey PUE add; dem no compare dis system with another one.
Emissions for different numbers of calls
| Case wey you wan compare | Value wey e get |
|---|
Every row use the same energy for each call, PUE and grid inputs. Na only call count change, so all the figures scale in direct proportion.
Formula
Energy (kWh) = calls × Wh for each call × PUE ÷ 1,000. Emissions (kg CO₂e) = energy × grid grams for each kWh ÷ 1,000. Emissions for each call (g CO₂e) = Wh for each call × PUE × grid grams for each kWh ÷ 1,000.
Example wey show the calculation
Calls na 100,000; energy for each call na 0.5 Wh; PUE na 1.2; grid intensity na 450 gCO₂e/kWh. Energy = 100,000 × 0.5 × 1.2 ÷ 1,000 = 60 kWh. Emissions = 60 × 450 ÷ 1,000 = 27 kg CO₂e. For each call: 0.5 × 1.2 × 450 ÷ 1,000 = 0.27 g CO₂e; for each 1,000 calls: 270 g CO₂e. All dis inputs na examples. Replace energy for each call and grid figures with your own before you rely on the result.
Where e fit work and limits
Dis na model for a scenario, no be formal carbon-accounting result. E cover only inference energy wey you describe. Training, model development, hardware manufacturing, networking, cooling wey PUE no cover and embodied emissions no dey inside. Energy for each call dey change with model, hardware, batching and prompt length. Grid intensity dey change with place and time too. Na why you must supply both inputs yourself; no single published figure fit every case. We no provide live emissions data or certified accounting.
Questions wey people dey ask often
How una calculate am?
Energy na calls times Wh for each call times PUE, divided by 1,000 to get kWh. Emissions na energy times grid grams for each kWh, divided by 1,000 to get kilograms of CO₂e.
Why I fit edit carbon factor and energy for each call?
Because dem na assumptions, no be fixed constants. Another model, GPU or region fit give different values. Na why dis page collect dem as inputs wey you must supply, and mark the defaults as examples only.
Wetin PUE mean for here?
Power usage effectiveness na total facility energy divided by IT energy. PUE of 1.2 mean 20% extra energy for cooling and other facility loads on top of inference energy. If e na 1.0, we count no extra energy.
Dis include energy to train the model?
No. E cover only inference calls for your scenario. Training, fine-tuning and hardware manufacturing dey separate. Dem fit be much larger for some cases, but dis model no count dem.
Why result dey rise directly with call count?
Every row for the volume table use the same figure for each call. If you double calls, estimate go double. Real systems fit batch or cache requests, so the relationship fit change. Na another reason to replace the example energy for each call with measured data.
I fit use dis for official emissions report?
No. Na planning estimate from your own assumptions. Formal report need measured energy data, documented grid factor for the correct region and period, plus the reporting rules wey your organisation follow.
Other tools wey relate to dis
- AI Inference Cost — cost for the same number of calls: tokens and prices for each million tokens
- Calculator for electricity cost — change a kWh figure into estimate for electricity bill
- Savings from AI Workflow — check whether time wey an AI tool save cover its monthly fee
Results na estimates; dem no replace advice from qualified professional.