Data Center Capacity & Power Tracker

Real structured US permit filings and global trade press — not live vacancy, which no free source publishes. North American colocation vacancy sits near a record low (~1%), and enterprises are securing capacity 18-24 months ahead of deployment. This tracks the two real, freely accessible leading indicators of that capacity — construction permits and trade-press coverage of operator announcements — rather than claiming access to real-time vacancy data that doesn't exist as a free feed anywhere.

Phase 1 source audit — what's actually accessible, verified live

The PRD flagged three candidate source types and asked to confirm which form the real v1 scope before writing any code. Here's what direct testing found:

Source type	Verdict	Basis
Direct operator press releases (Equinix, Digital Realty, etc.)	❌ Dropped from v1	Confirmed live: Equinix's own newsroom listing page is client-side rendered — no individual article links found in the fully-rendered DOM even after a real Playwright browser load and a dedicated search for the page's own JSON API calls (only cookie-consent-related XHR found, no press-release API). No RSS feed exists either. Not silently skipped — genuinely checked and ruled out.
Trade press (Data Center Dynamics)	✅ Included	Confirmed live. The site is Cloudflare-protected (plain HTTP 403s with a managed "Just a moment..." challenge), but a real Playwright browser loads it cleanly with no challenge shown. Its news listing already covers real operator capacity announcements in headline form (e.g. "Siemens Energy to supply B&W with 1GW of steam turbine-generators for data center power program") — effectively covering the announcement source type too, since press releases weren't directly accessible.
Permitting/planning records	✅ Included, and the strongest source found	MLQ Infrastructure (mlq.ai) aggregates real, primary-source government permit filings across 16 US states — confirmed live via direct inspection of a real permit detail page, which explicitly links "View original filing" back to the issuing county. Same Cloudflare-protected-but-Playwright-passable pattern as DCD. See below for more on this choice.

A meaningful data-quality distinction, not just a source list

MLQ permit records are real, structured, and traceable to a primary government source — clean Date / Application / Operator / Permit # / Value / Status columns, confirmed live against real pages for Equinix (82 real permits across 16 jurisdictions, $31.7M disclosed value) and Loudoun County, VA (446 real permits). Status values map to this actor's capacityStatus enum from MLQ's own real status strings (Filed→planned, Issued/Inspection Phase→under_construction, Completed→operational). MLQ's own Operator column on jurisdiction-level pages is honestly Unknown for many real permits — a genuine gap in the source data itself, passed through as-is rather than guessed at.

Trade-press records are free-text estimates, exactly as the PRD anticipated. Region, operator, MW, and vacancy-rate figures are extracted from short news headlines via keyword/regex matching — often null when a headline doesn't explicitly mention a market or figure, which is the honest, expected outcome, not a bug. Never presented with the same confidence as a permit record; sourceType distinguishes the two on every output row.

On using MLQ specifically

MLQ is a real commercial product (it has its own "Premium"/"Subscribe" tier for alerts and deeper analytics) built on top of public government records. This fetcher only reads publicly browsable pages — no login, no paywall, no CAPTCHA bypassed, just a real browser instead of a bot-blocked plain HTTP client (the same category of workaround used elsewhere in this portfolio, e.g. Poshmark in the Resale Marketplace Comps actor). It does not access or attempt to bypass MLQ's own premium tier. Every permit record traces back to a real government filing, explicitly linked on MLQ's own page.

Confirmed live, and worth knowing: operator pages cap at the first ~50 rows on initial load (Equinix's real 82 permits only showed 50) — MLQ appears to paginate/lazy-load beyond that on this page type specifically. Jurisdiction pages did NOT show the same cap in testing (Loudoun County's 446 real permits all came through). Not silently truncated and presented as complete — this is what was actually found.

Region coverage — set expectations correctly

Permit records are inherently US-only — construction permits are a US county/city government record type with no equivalent structure internationally. Non-US markets the PRD itself named as examples (Dublin, Singapore) can only ever surface through trade-press mode, best-effort, when DCD happens to cover them in a headline. The region-normalization layer (mapping "Northern Virginia"/"Ashburn"/"NoVA"/"Data Center Alley" to one canonical key, per the PRD's own example) currently covers 6 major US markets with real, individually-verified MLQ jurisdiction slugs; an unrecognized region name falls back to a direct slug guess against MLQ's real kebab-case URL convention rather than failing outright — which works for many real markets not explicitly listed, but isn't guaranteed.

Modes

Mode	Behavior
operator_lookup	Real structured MLQ permit records for each name in operatorNames
market_snapshot	MLQ jurisdiction records for each name in regions, plus a DCD trade-press news sweep
monitor	Same fetch as market_snapshot + operator_lookup combined, plus a real cross-run check (a named KV store, not a fresh one per run) flagging which records are genuinely new since the last run

Output schema

{
  "region": "string",
  "operatorName": "string",
  "announcedCapacityMw": "number | null",
  "vacancyRateEstimate": "number | null",
  "pricingBenchmarkUsdPerKwMonth": "number | null",
  "capacityStatus": "operational | under_construction | planned | precommitted | null",
  "sourceType": "press_release | trade_press | permit_record",
  "sourceUrl": "string",
  "scrapedAt": "ISO timestamp",
  "status": "success | failed",
  "errorMessage": "string | null"
}

sourceType never actually populates as press_release in this build — kept in the schema for completeness, not removed, since the reason it's unused is documented above rather than silently dropped. pricingBenchmarkUsdPerKwMonth is always null — no free source with real per-kW pricing benchmarks was found during Phase 1 research; not fabricated.

Setup note

Playwright/Chrome base image required — both real sources used here (MLQ, DCD) are Cloudflare-protected against plain HTTP.

Pricing

Event	Price
Run started	£0.05
Reading success	£0.06
Reading, seen before (monitor mode)	£0.02
Fetch failed	free

n8n integration

• Workflow A (trigger): scheduled weekly run against watched regions/operators.
• Workflow B (processing): alert when a genuinely new permit is filed in a watched region, or a tracked operator's status changes.