PA AI Factory

• 01

  5.0/ 10

  Power delivery (250–350 MW by end-2027 → 1.2 GW)

  Speculative

  Controlling claim. Plausible, not substantiated — should rise only after Nebius publishes a project-specific utility-service path.
  5.0/ 10
• 02

  4.0/ 10

  Timeline ('light up' 250–350 MW by end-2027)

  Speculative

  Possible if the private record is mature; weak on public evidence. No site, zoning, land-development, stormwater, air-permit, water/wastewater, building-permit, or host-utility service record is public.
  4.0/ 10
• 03

  4.0/ 10

  Site / location credibility ('secured land and power')

  Preliminary

  Credible as a sponsor statement; not yet credible as a public buildability record — 'secured' cannot be distinguished from fee ownership, lease, option, exclusivity, or conditional service pathway.
  4.0/ 10
• 04

  6.0/ 10

  Demand / offtake durability (Meta + Nvidia)

  Meaningful but qualified

  Strongest part of the case. Corporate demand evidence is real, but project-level linkage to the Pennsylvania capacity is not proven — no customer prepayment, minimum-revenue, or PA-specific allocation is public.
  6.0/ 10
• 05

  5.5/ 10

  Financing / capital viability (corporate capex plan)

  Meaningful but qualified

  Credible capital capacity for early phases; unproven full-campus financing. A 1.2 GW build is large enough that transmission, substation, backup-generation, water, site-civil, equipment, and schedule risk can materially alter the capital case.
  5.5/ 10
• 06

  5.0/ 10

  Economic viability

  Preliminary

  Plausible, but project-level economics are not auditable. Even favorable cases are not enough to treat the Pennsylvania project as economically proven absent capex, power price, tariff, PJM exposure, and incentive specifics.
  5.0/ 10
• 07

  4.0/ 10

  Community and permitting durability

  Speculative

  A path exists; durability is not proven. No host community is publicly known — no elected-official map, planning calendar, school-district analysis, environmental-justice screen, or opposition forecast can be completed.
  4.0/ 10

1. 01

   Undisclosed site and utility opacity

   This is the dominant gap across every claim. Without a site, the analysis cannot test utility headroom, zoning, title, geotech, slope, wetlands, floodplain, protected species, cultural resources, water supply, wastewater capacity, road access, heavy-haul constraints, adjacent uses, or local politics. Without a host utility, the analysis cannot test service agreements, upgrade scope, tariff exposure, cost allocation, or energization dates.

2. 02

   PJM large-load timing

   The end-2027 claim depends on a utility-service path that is already mature. If material network upgrades, substation additions, transformer procurement, or transmission-owner studies remain early, the schedule likely slips. PJM and utility large-load processes increasingly focus on readiness, collateral, service agreements, and stranded-cost protection because the market is seeing more speculative or duplicative load requests.

3. 03

   Demand is stronger than project proof

   The Meta and Nvidia facts strengthen Nebius's corporate case. They do not automatically prove that the Pennsylvania site has assigned revenue, customer commitments, equipment allocation, or project-level financing. The score is not a view on Nebius as a company; it is a score on a specific Pennsylvania project claim.

4. 04

   Capital intensity versus public economics

   A 1.2 GW AI factory is a capital-intensive infrastructure program. Nebius's 2026 capex guidance of $20–25B is large enough to support seriousness, but also large enough to create execution risk. If the Pennsylvania project inherits major transmission, substation, water, backup-generation, or site-civil costs, the economics could change materially.

5. 05

   Community politics can move faster than filings

   Once the site is named, local reaction can form quickly around power prices, water use, noise, backup generation, traffic, tax incentives, school-district revenue, and sponsor identity. The best time to publish the community-impact ledger is before the first incentive vote or discretionary local hearing.

Project

Nebius Pennsylvania AI Factory

Stage tag

Site Undisclosed · Pre-permitting

Sponsor / operator

Nebius Group N.V.

Site status

Pennsylvania site has not been publicly named — county, municipality, parcel, acreage, host utility undisclosed

Scale

Up to 1.2 GW at full buildout · 250–350 MW 'lights up' by end-2027 · ~300 MW/year thereafter

Demand anchors

Meta agreement up to $27B (5-year) · Nvidia $2B equity investment and partnership

Commercial momentum

Q1 2026 revenue $399M (+684% YoY) · >3.5 GW contracted power (target >4 GW by year-end) · ~$9.3B cash · 2026 capex guidance $20–25B

Comparable site

Nebius Independence, MO — groundbreaking May 2026, ~400 acres, up-to-1.2 GW, ~1,200 construction jobs, ~125–130 permanent jobs

PA policy context

Act 25 of 2021 Computer Data Center Sales and Use Tax Exemption · PA DCED GRID Standards · active data-center tax / moratorium proposals

PJM context

Pennsylvania sits in PJM territory; 1.2 GW campus likely requires direct utility planning, not ordinary retail service

Scoring arithmetic

Claim mean 33.5/7 = 4.8 · controlling claim 5.0 (firm power + site/utility opacity) · cap 6.5 (does not bind)

Evidence as of

May 29, 2026

Companion reports — Power & Grid Path, Permitting & Litigation, Community Impact, and Schedule Forensics — are produced on request.