Forge is preparing the requested surface and verifying the live route.
Forge is preparing the requested surface and verifying the live route.
AI inference consumes real electricity. Forge's compression, caching, and routing layers reduce that consumption automatically — with transparent, citation-backed methodology.
Aggregated across all Forge users. Updated every 5 minutes.
12,847 kWh
Energy saved
and counting
4,959 kg
CO2 avoided
conservative estimate
8.4M+
Total API calls
optimized through Forge
2100.0M+
Tokens saved
via ICE compression
12,275
Car miles not driven
86,194
Tree-days of CO2 absorption
1.3M+
Smartphone charges equivalent
Infinite Context Engine compresses prompts by 40-60%, meaning fewer tokens processed and less energy per inference call.
Identical or near-identical queries are served from cache — zero inference energy. Cache hits avoid the GPU entirely.
Simple queries route to smaller, more efficient models. A 7B model uses ~70% less energy than a 175B model for the same task.
Our calculations use conservative lower-bound estimates from peer-reviewed research. We cite every source. We never inflate numbers. When in doubt, we round down.
Patterson et al. (2021) — Carbon Emissions and Large Neural Network Training
Luccioni et al. (2022) — Estimating the Carbon Footprint of BLOOM
US EPA (2024) eGRID — 386 g CO2e/kWh US grid average
Strubell et al. (2019) — Energy and Policy Considerations for Deep Learning
Every Forge project includes energy impact tracking. Export compliance-ready reports for your ESG disclosures.
GET /v1/console/projects/:id/energy/export?format=json
{
"reportPeriod": "2026-03-01 to 2026-03-31",
"energySavedKwh": 6.231,
"co2AvoidedKg": 2.405,
"co2AvoidedLbs": 5.3,
"certificationNote": "Suitable for Scope 3..."
}
Suitable for Scope 3 Category 1 reporting under the GHG Protocol. CSV and JSON formats available.
Every API call through Forge automatically saves energy. No config needed. Just use Forge.