Regional Grid Carbon¶

Location-based vs market-based accounting for AI emissions.

Overview¶

Grid carbon intensity varies significantly by region:

Region	gCO₂/kWh	Notes
France	60	Nuclear-heavy
California	220	Renewables + gas
UK	250	Wind + gas
Germany	350	Mixed
US Average	380	Varied
World Average	400	Default
Poland	650	Coal-heavy
China (coal regions)	800+	Coal-dominant

Using accurate regional data can significantly improve estimate accuracy.

Current Implementation¶

Default: World Average¶

OpenClaw uses 400 gCO₂/kWh by default:

{
  "green": {
    "defaultGridCarbon": 400
  }
}

This is conservative (higher than hyperscaler averages).

Manual Configuration¶

Set regional value if known:

{
  "green": {
    "defaultGridCarbon": 220
  }
}

Accounting Methods¶

Location-Based¶

Uses grid average where electricity is consumed:

CO₂ = Energy × Regional_Grid_Average

Pros: - Reflects actual grid mix - Simple to understand - Required by some standards

Cons: - Doesn't reflect provider choices - Provider location often unknown

Market-Based¶

Uses contractual instruments (RECs, PPAs):

CO₂ = Energy × Contractual_Factor

Pros: - Reflects provider renewable purchases - Incentivizes clean energy procurement

Cons: - Requires provider disclosure - RECs may not reflect actual consumption

Provider Renewable Claims¶

Major cloud providers claim significant renewable energy:

Provider	Claim	Verification
Google Cloud	100% matched	Annual reports
Azure	100% by 2025	Commitments
AWS	100% by 2025	Commitments

Important: These are aggregate claims. Actual inference may occur in any region.

Why Default is Conservative¶

We use 400 gCO₂/kWh because:

1. Unknown location — Can't verify where inference runs
2. Time variance — Grid carbon changes hourly
3. Renewable matching — Annual matching ≠ 24/7 clean
4. Transmission losses — Often excluded from claims

Using world average ensures we don't underestimate.

Future: Real-Time Grid Data¶

Electricity Maps API¶

Electricity Maps provides real-time grid carbon:

// Future implementation
async function getGridCarbon(region: string): Promise<number> {
  const response = await fetch(
    `https://api.electricitymap.org/v3/carbon-intensity/latest?zone=${region}`,
    { headers: { "auth-token": apiKey } }
  );
  return response.json().carbonIntensity;
}

Status: Stub implemented, requires paid API access.

EPA eGRID¶

EPA eGRID provides US regional data:

Region	gCO₂/kWh
CAMX (California)	225
ERCT (Texas)	380
NYUP (NY Upstate)	115
RFCW (Midwest)	450

Status: Static data available, mapping to cloud regions needed.

IEA Emission Factors¶

IEA provides country-level data:

Country	gCO₂/kWh
France	56
UK	231
Germany	350
Japan	457
Australia	517

Status: Paid data source, annual updates.

Cloud Region Mapping¶

Map cloud regions to grid regions:

const REGION_GRID_MAP: Record<string, number> = {
  // AWS
  "us-east-1": 380,      // Virginia
  "us-west-2": 220,      // Oregon (hydro)
  "eu-west-1": 350,      // Ireland
  "eu-central-1": 350,   // Frankfurt

  // Google Cloud
  "us-central1": 400,    // Iowa
  "us-west1": 220,       // Oregon
  "europe-west1": 250,   // Belgium

  // Azure
  "eastus": 380,
  "westus2": 220,
  "westeurope": 350,
};

Challenge: Providers don't disclose which region serves each request.

Configuration Examples¶

US West Coast Deployment¶

{
  "green": {
    "defaultGridCarbon": 220
  }
}

European Deployment¶

{
  "green": {
    "defaultGridCarbon": 300
  }
}

Conservative (Unknown)¶

{
  "green": {
    "defaultGridCarbon": 400
  }
}

Compliance Implications¶

GHG Protocol¶

Both methods allowed: - Location-based for Scope 2 - Market-based with contractual proof - Scope 3: Supplier's method if available

SBTi¶

Prefers market-based for renewable claims: - Must have contractual instruments - Annual matching minimum - 24/7 matching encouraged

TCFD¶

Report both if material: - Location-based as primary - Market-based as supplementary

Recommendations¶

Today¶

1. Use default (400) if location unknown
2. Configure regional value if you know the data center
3. Document your assumption

When Data Available¶

1. Enable real-time API
2. Use location-based for accuracy
3. Apply market-based if provider discloses

For Reporting¶

1. State which method used
2. Document grid carbon assumption
3. Note limitations