Status: v1.0 — adopted 23 July 2026. One first-pass assessment (Verne Keflavik) has completed its right-of-reply window (no response by the deadline) and stands as published. Eight further operators are under pre-review assessment with right-of-reply windows closing 4 September 2026, and a ninth — Blix Solutions, added in batch two — is published as a pre-review draft but has not yet been contacted. See the right-of-reply log for status by operator. Formal Index scoring on both axes begins once outstanding right-of-reply responses are in. See the changelog at the end of this page.
The Greenpool Index scores AI-compute facilities and providers on two independent axes. A facility can be genuinely green and jurisdictionally exposed, or fully sovereign and quietly fossil-backed — the two questions deserve separate answers.
What we include
Until now this page set out how we score facilities but not which facilities qualify to be scored. That gap produced a fair public challenge from an operator we had left out, so here is the boundary, stated openly.
The universe we select from. For Norway, the authoritative list is Nkom’s public register of commercial datacenter operators, maintained under datasenterforskriften, updated daily and downloadable as CSV. As of 25 August 2026 it holds 60 registered commercial operators and 112 registered datacenters. Internal (virksomhetsinterne) datacenters must also register above 0.5 MW, but their names are withheld for security reasons.
For the other Nordic countries we have identified no equivalent mandatory register. Coverage there is built from operator disclosures, company registries and public sources — which is a genuinely weaker basis, and we would rather say so than imply uniform completeness across the region.
The criterion. We assess facilities relevant to AI and HPC workloads. Batch one covered the largest MW-scale operators in Norway. Batch two extends to mid-scale Norwegian operators. The threshold is capacity- and workload-relevance-based rather than a fixed MW cutoff — we would rather name the reasoning batch by batch than claim a bright line we do not actually apply.
What we exclude, and why.
- Internal, non-commercial datacenters. They do not sell capacity, so there is no buyer-side decision for this Index to inform.
- Operators whose consumption is substantially cryptocurrency mining. Nkom publishes a per-operator crypto share; nine of the sixty registered operators report some share, ranging from 1% to 100%. Norway has announced a prohibition on cryptocurrency mining in datacenters, which makes this a shrinking category as well as an out-of-scope one. The operators reporting a crypto share in the register are Currency Edge AS (100%), Thermaltech AS (100%), Exanorth AS (100%), Troll Housing AS (97%), Arctic Flux AS (95%), Bluebite GmbH (55%), Tydal Data Center AS (33%), Bluefjords AS (14%) and Nordic Blocks AS (1%) — figures as published in the register, not independently verified by us.
If we have missed you. If you operate a datacenter that meets the criterion above and we have not assessed it, tell us: [email protected]. Being challenged in public on our coverage is how this section came to exist, and a correction from an operator is worth more to us than the appearance of having been complete all along.
What the Index does not measure
Every index is a choice about what to count, and those choices leave gaps. These are the ones we know about. None of them is captured by the current scores, and saying so is more useful than letting a reader assume otherwise.
The building, not the machines inside it. The Index scores ownership, jurisdiction and operational control of the facility. It does not score who owns the servers, cooling units and cabling inside it. That equipment may belong to a third party, may be containerised, and may be relocated on far shorter notice than a building can be. Two facilities with identical ownership structures can therefore differ substantially in who actually controls the compute. This is a known gap, under consideration for a future version of the methodology; it is not something the current scores capture.
There is a worked example of this inside our own dataset. Blix Solutions — a Norwegian operator whose ownership chain ends at one named private individual, the cleanest we have found — lists Vaultica OSL01 among the co-location sites it sells from. Vaultica is owned by Apollo Global Management, a US asset manager. So Blix customer equipment sits inside a US-owned building, and the two companies land at very different points on our Sovereign axis while being, in this one respect, the same physical arrangement. Neither operator’s score reflects the other’s involvement, because the Index has no dimension for it.
Contractual and regulatory portability. Ownership and physical location do not determine whether equipment can lawfully be moved, sold or switched off. Export-control regimes have historically restricted the movement of hardware across borders regardless of who owns it or where it sits, and extraterritorial legislation can reach data held under foreign-parented operators. The Index scores where things are and who owns them. It does not score what a counterparty is contractually or legally permitted to do with them.
Hardware supply chain. GPU and component vendor concentration is industry-wide and sits outside any individual facility’s control, so it is out of scope here and addressed in Greenpool’s editorial coverage instead.
Workload and customer concentration. The Index does not assess the dependency created by a single anchor tenant or offtake agreement, though several of the facilities we have assessed have exactly that.
These limitations were sharpened by public discussion following publication rather than identified solely in-house, which is the honest account of where they came from. If you can identify others, write to [email protected].
Axis 1 — Green (0–100)
| Dimension | Weight | What we look for |
|---|---|---|
| Power source verifiability | 35% | Physical PPA or direct supply beats certificates (GOs/RECs). Grid-mix honesty. Named counterparties. |
| Energy & water efficiency | 25% | Published, audited PUE beats “design target”. Measurement standard disclosed (ISO/IEC 30134-2). Disclosure of WUE (water usage effectiveness) is explicitly rewarded. |
| Heat reuse | 20% | Waste heat delivered to district heating or industry, with named offtaker and volume. |
| Grid additionality & impact | 20% | Does the facility bring new renewable generation, or consume scarce existing hydro? Grid-connection transparency. |
Heat reuse N/A rule: where no feasible heat offtaker exists within economic distance of a facility, the heat-reuse dimension is marked N/A and its weight is redistributed pro-rata across the remaining Green dimensions. The redistribution is noted on the scored page. Geography is not a behavior; the Index scores behavior.
Axis 2 — Sovereign (0–100)
| Dimension | Weight | What we look for |
|---|---|---|
| Ownership & ultimate control | 35% | Ultimate beneficial owner, jurisdiction of parent, foreign-control exposure (e.g. US CLOUD Act, Chinese parent). |
| Legal jurisdiction of operations | 25% | Operating entity’s jurisdiction; contract governing law; EU/EEA data-protection reach. |
| Operational control | 25% | Who holds root? Where are operations staffed? Remote-management exposure. |
| Data residency guarantees | 15% | Contractual, auditable residency vs. marketing statements. |
Scoring rules
- Only public or provider-confirmed evidence counts. Unverified claims score as absent, not as true.
- Sponsors never touch scores. Commercial relationships are disclosed on every scored page.
- Uncertainty is published as uncertainty. Every score carries a confidence grade (A: audited/primary, B: corroborated secondary, C: single-source).
- Right of reply. Providers see their assessment before publication and their response is published verbatim. Should a response arrive after publication — including after a deadline has passed — we will update the assessment accordingly; the right of reply does not expire in spirit, only in the sense that a page ships without waiting further. See the right-of-reply log for every operator contacted, when, and current status.
- Corrections policy. Errors are corrected in place with a dated changelog; the original wording remains visible.
What the Index is not
Not a ranking of “best datacenter” — workloads differ. Not pay-to-play. Not a substitute for your own due diligence; it is the map you start it with. For the specific things the current scores do not capture, see What the Index does not measure above.
Sources
Primary sources we rely on, in rough order of evidential weight:
- Nkom’s register of commercial datacenter operators (Norway) — mandatory under datasenterforskriften, maintained by the Norwegian Communications Authority, updated daily, downloadable as CSV. This is the authoritative universe for Norwegian coverage and is materially stronger than any directory-derived dataset.
- Brønnøysundregistrene (Norway) — the company register, for ownership chains and beneficial ownership, accessed directly and via proff.no.
- Operator disclosures — sustainability reports, facility pages and press releases, treated as primary for the operator’s own claims and graded accordingly (an operator-published PUE is primary evidence of what the operator says, not independent verification that it is correct).
- Trade press and third-party directories — used for corroboration and lead-generation, not as a sole basis for a published figure where a primary source should exist.
Changelog
Addendum — 26 August 2026. Added the “What the Index does not measure” section, documenting four things the current scores do not capture: equipment ownership inside the facility as distinct from the facility itself; contractual and regulatory portability (export controls, extraterritorial reach); hardware supply-chain concentration; and workload/customer concentration under a single anchor tenant. The hardware supply-chain note previously carried in “What the Index is not” is consolidated there rather than stated in two places, and “What we include” now sits before the axis definitions so that scope and limitations precede mechanics. These limitations were sharpened by substantive public criticism following the launch post. No axis weights, dimensions or scoring rules changed — this addendum documents what v1.0 does not do; it does not alter what it does.
Addendum — 25 August 2026. Added the “What we include” section: the inclusion threshold, the Nkom register as the stated universe for Norwegian coverage, the exclusions (internal datacenters; substantially crypto-mining operators), and a contact route for operators we have missed. Added the Sources section above. Prompted by an operator publicly raising that they had been omitted without our having published any criterion — the criticism was fair and this is the fix. Axis weights, dimensions and scoring rules are unchanged; this addendum does not alter v1.0 scoring.
v1.0 — 23 July 2026, adopted by the founder.
- Heat reuse N/A rule added: no feasible offtaker within economic distance → dimension marked N/A, weight redistributed pro-rata across remaining Green dimensions, noted on the scored page.
- Efficiency dimension renamed “Energy & water efficiency”: PUE remains the primary measure; WUE disclosure is now explicitly rewarded.
- Scope clarification added to “What the Index is not”: hardware supply-chain sovereignty is out of scope.
- Weights on both axes confirmed unchanged from v0.1 after founder review.
v0.1 — 23 July 2026. First published version.