Skip to content

Carbon Intensity Sources

Overview

The carbon intensity of electricity (grid_gco2_per_kwh) is one of the most influential inputs in the climate estimate. Vetted Inference supports multiple sources, selected by configuration and fallback order.

Sources

Electricity Maps (preferred)

Real-time carbon intensity by grid zone. Requires an API key and network access.

  • Coverage: 160+ countries, hourly resolution.
  • Method: Marginal or average carbon intensity depending on endpoint.
  • Uncertainty: Low for average intensity; higher for marginal due to model assumptions.
from vetted_inference import PipelineConfig

config = PipelineConfig(use_electricity_maps=True)

Fallback table

A built-in table keyed by country_iso and electricity_map_zone provides annual average grid intensities from Ember Climate and the IEA.

  • Coverage: Global.
  • Resolution: Annual average.
  • Uncertainty: Higher than real-time due to temporal and geographic averaging.

Custom override

You can supply a per-request intensity directly:

request = InferenceRequest(
    ...,
    grid_gco2_per_kwh=37.0,  # Swedish grid average
)

This is useful when you have a power purchase agreement (PPA) or market-based instrument with a known emission factor.

Water intensity

Water consumption per kWh (wue_l_per_kwh) is drawn from data-center profiles. When absent, a global average of 0.5 L/kWh is used. This is a coarse proxy; operators should measure and report actual WUE where possible.

Embodied carbon

Hardware embodied impacts are sourced from:

  • BoaviztAPI: API-driven LCA for servers and components.
  • Local table: Fallback coefficients for common GPUs (H100, A100, L40S, etc.) derived from public teardown studies and ecoinvent.

Updating sources

Carbon intensity data changes as grids decarbonise. The fallback table is updated quarterly. Operators using real-time APIs receive the latest available data automatically.