Peerigon · AI Hosting Overview

LLM Hosting & Sustainability (EU)

Compare hyperscaler coverage, sustainability scores, and compliance posture—scraped automatically from AWS Bedrock, Google Vertex AI, and Azure AI Foundry docs.

AWS Bedrock Google Vertex AI Azure AI Foundry Sustainability metrics (OWID)

Generated: —

https://github.com/peerigon/llm-hosting-overview

Filters

Platform

(AWS Bedrock, Google Vertex AI, Azure AI Foundry)

Model vendor

Model name

Sustainable regions only (< 200 gCO₂/kWh)

Hide cross-region inference entries

Legend Sustainable region (< 200 gCO₂/kWh) High nuclear share (≥ 25%)

Data notes

Carbon, renewables & nuclear

Carbon intensity, renewable share, and nuclear share come from Ember (2025) + Energy Institute (2025) via Our World in Data. Each dataset inherits the OWID processing pipeline and source licenses.

Carbon intensity: Ember (2025); Energy Institute – Statistical Review of World Energy (2025)
Renewables: Ember (2025); Energy Institute – Statistical Review of World Energy (2025)
Nuclear: Ember (2025); Energy Institute – Statistical Review of World Energy (2025)

Please credit all sources listed above. Data provided via Our World in Data remains subject to the original providers’ license terms.

Scraper sources

Project details

Refresh data

Run npm run scrape to pull the latest provider docs and sync docs/data/hosting-regions.json.

Commit the refreshed JSON to redeploy the dashboard on GitHub Pages.

Data coverage

Focused on EU availability for AWS Bedrock, Google Vertex AI, and Azure AI Foundry (Data Zone + serverless) deployments.

Azure coverage reflects Microsoft Learn pages that publish explicit regional availability. Catalog-only models without concrete region data are intentionally hidden from the dashboard, while Azure Direct Mistral rows are included when the Data Zone Standard tables publish EU regions.

Each entry carries region metadata — carbon intensity, clean energy share, nuclear share, and the derived sustainable / high-nuclear tags — so you can filter by operational constraints, not just location codes.