Weekly AI infrastructure review: 2026-06-22

[github-actions]

Weekly AI infrastructure review (2026-06-22)

Automated check from scripts/weekly-ai-infrastructure-check.mjs. Triage and either:

• Add or update entries in data/ai-infrastructure-projects.json (and mirror to public/api/ai-infrastructure/projects.json), bump last_reviewed in data/page-freshness.json, then close this issue, OR
• Comment noop and close if nothing actionable surfaced.

Current state of the registry

• Last reviewed: 2026-06-05 (17 days ago) ⚠️ OVERDUE
• Cadence: weekly
• Projects tracked: 17
• By status: construction (11), operational (4), permitted (1), announced (1)
• Announced capacity total: 15.2 GW

AI-infrastructure-flavored news, last 7 days

Matched 1 articles (keyword scan; many will be tangential).

Date	Source	Title
2026-06-18	NVIDIA AI Blog	How FERC’s Large-Load Interconnection Actions Help Address Grid Stress, Improve Affordability

Sources: NVIDIA AI Blog (1)

---

What justifies a new entry

1. A gigawatt-class campus is announced, permitted, or breaks ground.
2. A nuclear PPA or restart deal lands between a hyperscaler and an operator.
3. An AI-specialized compute operator (CoreWeave / Lambda / IREN / Hut 8 / Crusoe class) ships a materially larger fleet or PPA.
4. An existing tracked project changes status (announced to permitted to construction to operational).
5. A regulatory event materially reshapes the buildout (FERC ruling on grid bypass, NRC license decision, big court ruling).

What does NOT justify a new entry

• Generic data center announcements not tied to AI workloads.
• Speculative reporting without operator or partner attribution.
• A < 100 MW deployment from a non-flagship operator.
• Routine quarterly capex disclosures without a specific named project.

When you update the registry, bump last_reviewed in data/page-freshness.json so the visible footer stamp moves forward and the OVERDUE flag clears.