# Microsoft Azure — `azure-regions`, `microsoft-news`

## azure-regions (cloud regions)

| | |
|---|---|
| Source | `https://datacenters.microsoft.com/globe/data/geo/regions.json` (public JSON behind the Azure globe) |
| Kind / mode | `cloud_provider` / `json`, implementation `cloud_regions`, parser `cloud_azure_regions_v1` |
| Fetch | L1 direct, 0 credits (robots.txt: allow all, crawl-delay 10 → rpm 10) |
| Schedule | weekly |
| License / attribution | Publicly published provider documentation; facts only — "Microsoft Azure — datacenters.microsoft.com" |
| Verified live | 2026-09-11: 57 regions, 57 valid, 0 rejected (37 operational, 15 announced, 5 restricted) |

**Fields.** `id` → code, `displayName` → name, `location` → city/state, `geographyId` → country fallback, `isOpen` → status
(`announced` when false), `availableTo` mentioning "business presence in China" → `restricted` (the 21Vianet-operated
China regions), `yearOpen`/`monthOpen` → `launchedOn` (year, or `YYYY-MM` when a month is published), `availabilityZonesId`
(`available | soon | nearest | nearestSoon`) → `externalIds.availability_zones_state`. Microsoft does not publish the
number of zones, so `availabilityZones` is null. `announcementLink`, `dataResidency`, `typeId` kept in `externalIds`.

**Geo.** Curated city table when `location` is a city (43/57, precision `city`; "N. Virginia"-style metros → `metro`).
When Microsoft only names a US state or "Hebei", their own region marker is used with precision `approximate`
(`geo.source = provider-marker:azure`). When they only name a country (North Europe = "Ireland", West Europe =
"Netherlands", Israel Central = "Israel") no coordinates are emitted — their country markers sit offshore or across
a border (the West Europe marker is in Germany). Spot checks: `eastus` → Northern Virginia US (metro), `southeastasia`
→ Singapore SG, `japaneast` ("Tokyo, Saitama") → Tokyo JP, `swedencentral` ("Gävle and Sandviken") → Gävle SE.

**Quirks.** `geographies.json` only has continent markers (not used). Announced regions have no `yearOpen`.

## microsoft-news (newsroom)

Implementation `cloud_newsroom` over four feeds: `blogs.microsoft.com/blog/tag/datacenters/feed/` (the precise one:
Pecos, Fairwater, water intensity, Maia, Fungible/Lumenisity acquisitions), `blogs.microsoft.com/feed/`,
`blogs.microsoft.com/on-the-issues/feed/`, `news.microsoft.com/source/feed/`. Items are filtered on title+summary
**before fetching** (`params.include` / `params.exclude`), then run through the generic article classifier; a project
candidate is produced when MW + status keywords appear (verified: "new datacenter in Pecos" → project, 2000 MW, announced).
Live 2026-09-11: 7/36 feed items matched, 6 entities, 0 rejected, 0 credits. Daily.