Independent, public observatory

A data center can no longer be built just anywhere.
Let's measure its risk.

Data centers are transforming our territories. Water, energy, land, local impact, transparency: ScoreMyDataCenter publicly assesses the acceptability score of every project, with an open method and sourced facts.

We are independent. We apply the industry's own standards — PUE (ISO/IEC 30134), energy, carbon — then we measure what they leave out: water, land, territory. The result is a Nutri-Score for 3,293 real sites.

  • The grade is free
  • The method is public (v0.1.0 frozen)
  • The sources are verifiable

Newly flagged projects

See the watch →

Proof you can look at

3290 sites. Seen from orbit. Graded.

Every data centre in the corpus is photographed from orbit, cropped automatically to its footprint, and re-photographed at each update. No marketing imagery: the territory as it is.

⏳ Planned site bare land or building site: the data centre does not exist yet — we already grade the territory that will host it.

Hover to pause · click for the site's ficheImagery © Esri, Maxar, Earthstar Geographics

We don't ask permission to publish our scores. We hold ourselves to the method.

A critic reviews a film without the studio's approval. A journalist publishes an investigation without the consent of those it disturbs. ScoreMyDataCenter belongs to that tradition: freedom of criticism and contribution to the public-interest debate.

Our contract in return: a public method, sourced facts, measured words, and a permanent right to correction and response. Read why →

6% publish their PUE Fleet figures · operators' declarations · n = 560
35% under high water stress Fleet figures · WRI Aqueduct · n = 1571

The hottest issue in AI infrastructure has no independent referee. Now it does.

Our mission: measure, source and make public the acceptability of every data center project — grounding the debate in facts.

Why this observatory

A 900 MW data center is announced near you. Where do you read, clearly, what it's worth? Its local impact?

Existing maps list technical specifications; conflict trackers count disputes. No one lays out, project by project, the reality of each case: a grade, sources, a method anyone can verify and recompute. That is the gap ScoreMyDataCenter fills: the first independent acceptability grade, public and contestable — factual, sourced and methodical ratings, published as a contribution to the public-interest debate. And a scale with teeth: an A is proven in operation, never promised — even excellent promises cap at B; only verified proof earns the A. Our mission: score every data center — methodically, publicly, one file at a time.

Animation: Kyndryl – FRVENISS's fiche gets graded — the five pillar sub-scores appear, the site score fills, then the letter stamps in.

Real case: Kyndryl – FRVENISS — five pillars, the site score, then the letter. Public method, sourced data.

Why this site? Its score (57.3/100) sits closest to the median of the 3290 sites in the corpus: the most ordinary site in Europe, neither showcase nor cautionary tale. The selection rule is mechanical and published.

Our method, in 4 principles

Public method Open scoring grid, versioned and citable. Anyone can re-run every grade.
Traceable data Every value links to its source. Announced claims are never merged with measured facts.
Correction and operator response A data point seems inaccurate? It can be reported with its source. After verification, the datasheet is corrected and the change dated. The operator can also send a response, published in a dedicated space.
Independence Scored entities never pay for their grade. No influence, for or against: the grade follows the method.

Why it matters

A data center is not just an IT building: it is a project that mobilises vital resources and commits a territory.

Water resource

1117 sites out of 3259 sit in a high water-stress or water-crisis basin. Water weighs heavily in the corpus grades.

See the map →

Power grid

Grid saturation slows connections and degrades scores. Siting choices with heavy consequences.

See the map →

Land & biodiversity

Artificialisation, soils, biodiversity: local impacts that matter as much as megawatts.

Understand the stakes →

Transparency

2944 projects out of 3293 do not disclose their power. We grade on public documents — opacity never pays.

Understand the method →

One method, two grades, always with confidence.

26 indicators across 5 pillars, split between what a project endures and what its operator chooses. The grid, its weights and its thresholds are public and downloadable.

  • Site The context the project endures: the electricity grid's carbon intensity, water stress, land, local fabric and the town's economy.
  • Operator The levers the operator chooses: energy efficiency, cooling, jobs, consultation, transparency.
  • Documentation confidence A grade never appears without it. Below 40% coverage we say “insufficient data”. We do not grade the unknown.
Understand the grade, step by step →

Pillar weight in the global grade

  • Energy 25%
  • Water 20%
  • Land & biodiversity 20%
  • Local impact 20%
  • Transparency & governance 15%

Public, versioned and dated methodology — verifiable anteriority.

Intelligence Library — the analysis above the scores

The site says how much. Our analyses say where it's heading, and what it means.

  • Country and Europe trends
  • Risks by theme — water, energy, land, transparency
  • Operator benchmarks
Discover the analyses →

We open the truth.

The layer that produces the judgment is open: method, data and engine, verifiable by anyone.

Open method, data and engine

AGPLThe engine code is free: anyone who improves it must share those improvements back. engine, ODbLOpen database: freely reusable, provided you credit the source and share alike. published data, CC BY-SAFree content: reusable if you credit the author and share under the same terms. methodology. The code is public on GitHub: anyone can recompute every grade, with full rigor.

We help the project, never the grade

Scored entities never pay for their public grade. An operator may have its project analysed; but the grade moves only when the real project changes — never because it paid. "Pay to raise the grade" is structurally impossible.

Public audit log

Every scoring, revision and correction is dated, versioned and justified. Grades cannot move silently.

Governed contributions

Data corrections are free and reviewed. To stop anyone from gaming a grade, our detection rules are public in principle; their numeric thresholds stay confidential — publishing them would hand out the manual for slipping through.

See the code on GitHub ↗

Our commitment: a project's grade and its justification are public and free, forever.

Who we are

An independent observatory.

ScoreMyDataCenter is the work of Franck Bardol. At the intersection of AI infrastructure, ethics and public decision-making.

  • Independent: no operator funding, no conflict of interest.
  • Public: method, sources and grades accessible to all.
  • Open: corrections and responses possible for every scored project.

AI in service of the method, not in place of the human.

Because computing a Nutri-Score for every data center in Europe is not something you improvise. An architecture of specialised agents processes tens of thousands of sources to grade each data center rigorously — under human control.

  1. Collect

    The agents locate the useful public sources: regulatory files, institutional data, operator documents, maps, local decisions and press articles.

  2. Structure

    They turn these heterogeneous sources into comparable data: power, water, land, grid, consultation, transparency, project status.

  3. Verify

    They cross-check information, flag contradictions, identify missing data and keep the source of every fact.

  4. Compute

    They apply the same method to every project, recompute the grades and check that thresholds, rules and guardrails are respected.

  5. Monitor

    They detect changes: a new authorisation, an appeal, a project modification, a published figure, an evolution of the grid or the local context.

  6. Prepare publication

    They produce a first factual synthesis and flag fragile wording. Final validation, arbitration and publication remain my responsibility.

Human responsibility

I define the strategy, the methods, the ESG weightings and the publication rules. I arbitrate complex cases, settle disagreements, orchestrate the specialised AI agents and stand behind every published grade. Agents propose, the human decides.

Expertise in service of the public debate

  • 10+ yearsin AI, data and quantitative modelling
  • Companies & institutionsAI projects for Banque de France, Thales, Airbus, General Electric, Bouygues Colas, SNCF, Saint-Gobain, STMicroelectronics, the ITER/CEA nuclear-fusion project, Vinci, GlaxoSmithKline, the French space agency CNES, SMEs…
  • Teaching & transmissioncourses, conferences, keynotes, LinkedIn Learning MOOCs (50,000+ professionals trained in AI)
  • Ethics & responsibilityMSc in AI and machine learning (Université Paris 6 – CNAM), Management (University of Geneva), statistical modelling (Université Paris Nanterre) and a Certificate in Philosophical Ethics (University of Geneva)

AI makes it possible to work at the scale of the corpus. The method, the arbitration and the responsibility remain human.

Frequently asked questions

A project has been announced near me. Where do I start?

With the calendar, because it decides everything: the same question asked during consultation can still change the project; asked after the authorisation, it changes nothing. Appeal windows are two months — from the first day of posting for a building permit, from the last formality completed for an environmental authorisation. And once the site is running, what remains is a claim to the prefect about its prescriptions (noise, discharges, water). The whole path, step by step, is in Rights and calendar (in French), with the legal texts.

The Nutri-Score of data centers. How is a data center graded?

Each project receives two A–E grades — one for the site, one for the operator — from 26 indicators across 5 pillars (energy, water, land, local impact, transparency). A numeric score is computed, then translated into a letter, exactly like the food Nutri-Score everyone knows; next to it, a confidence level says whether the grade rests on solid or fragile data.

What separates a data center graded A from one graded B?

The number of pillars under tension, and how severe they are. A: no marked weakness on the pillars published on the fiche. B: solid, with one shadow. C: two clear tensions. D and E: several pillars in the red, or one major tension on its own. As with the food Nutri-Score, the letter sums up several criteria at once — and a single very poor criterion is enough to pull it down. One thing worth knowing: the cap only lifts on third-party verified proof, and no site in Europe reaches A today.

Why don't you publish the numeric thresholds behind the grade?

So the grade cannot be gamed. We publish the grid, the weights of the 5 pillars, the sources behind every indicator and, on each fiche, the factor that weighs most on the grade — so what counts, and in which direction. We do not publish where the thresholds sit: with that, a developer could aim at the letter instead of aiming at the ground. The method is verifiable; the recipe is not optimisable.

What do the two grades, “site” and “operator”, mean?

The site grade assesses what the project endures — territory, water, grid, land — whoever carries it. The operator grade assesses what the operator chooses — efficiency, consultation, transparency, commitments. We separate the terrain from the decisions.

Can I get the project near me graded?

Yes. Suggest it through the Contact section: we prioritize the best-documented and most-debated cases. One independence note: we never grade on behalf of a party, for or against — the grade follows the method, not the requester.

Where does the data come from?

Public sources: power grid, water, land registries, census data, official records, local press. Every displayed value links to its source, and the method is open: anyone can redo the computation.

Why don't US projects have a grade (yet)?

Because depth follows the data. A grade requires thresholds calibrated country by country (grid, water, land, registries, procedures) and a working correction-and-response channel — which is the case in Europe today. In the United States we first publish the 'on watch' layer: citizen oppositions ✋🏼 and moratoria ⏸️ — sourced facts, never a grade. The first US grades will come when the data and the calibration are ready, not before: we do not grade the unknown.

Can an operator pay to improve its grade?

No, never. Scored entities do not pay for their grade. A project's grade and its justification are public and free, forever.

Why does no data center have an A (yet)?

Because an A is proven, never promised: it is reserved for values verified in operation. A fully declarative file, however excellent, caps at B — “announced ≠ measured”. An operator earns its A by bringing third-party proof (its EED filing to the regulator, an ISO/IEC 30134 audit by an accredited body — never a marketing page) through our public correction channel: the data flips from “announced” to “measured”, the cap lifts, and the re-score is traced in the public audit log. An A cannot be bought or declared: it is demonstrated. And proof lifts the project & process grade — never the site grade: the territory is endured, it cannot be proven away.

Why we publish without asking permission

A critic reviews a film without the studio's approval. A journalist publishes an investigation without the consent of those it disturbs. ScoreMyDataCenter belongs to that tradition: freedom of criticism and contribution to the public-interest debate. The grade is an exercise of the freedom of criticism; its strength rests on the transparency of the method, the quality of the sources and the restraint of the words. We do not ask permission to publish an independent assessment. In return, we hold ourselves to a public method, sourced facts, measured words, and a permanent right to correction and response.

What if a data point is wrong, or the operator disagrees?

Two distinct tools: factual correction — any data point is corrected on evidence, the change is dated, and when the calculation is affected the grade is recomputed under the same method version — and the operator's response, published in a dedicated space, clearly attributed and dated. Publishing a response never changes the grade automatically: it only moves when new or corrected facts change the result produced by the method.

Can an operator request removal of its fiche?

A wrong data point is corrected on evidence, and the operator can send a response published alongside the grade. But a sourced, methodical grade is not removed on request: pulling it because it displeases would be selling it in reverse. The grade stays; the counter-argument is added.

Is the grade a final verdict?

No: it is an acceptability-risk grade, coupled with a confidence level, and it can evolve with the project. The numeric score is published to make the computation transparent, but the A–E letter is what counts — the grade applies an expert, sourced, versioned grid; it does not claim the precision of an oracle.

Does the grade evolve with technical progress? Will a well-graded data center still be well-graded in ten years?

Yes, by design. A grade changes for two reasons: the project evolves (new data), or the yardstick evolves with the state of the art. Our thresholds anchor on standards that tighten over time (ISO, European regulation): an A always means “the best achievable today”, not “the best of 2026”. Without that, everyone would end up A as technology progresses — which is exactly why the EU energy label was reset in 2021, when nearly every appliance had become A+++. Each grade is therefore dated and tied to a methodology version: a data center excellent in 2026 may be “average” by 2036 standards — and that is precisely the score's job: reward progress, and flag what ages.

Who are the players in the data center industry?

Several families, which often meet on the same project. Developers: they find the land, obtain permits and grid connection, then sell or lease. Colocation operators: they run data centers and rent out space, power and connectivity to their customers. Hyperscalers: the cloud giants, which build and run their own campuses. AI neoclouds: specialized providers renting out computing power for artificial intelligence. Energy companies and grid operators: they produce and supply electricity and connect the site (in France, RTE for transmission, Enedis for distribution). Builders and equipment makers: they construct the building and supply cooling, backup power, servers. Local authorities and the State: in France, the municipality or inter-municipal body sets planning rules and usually grants the building permit; the préfet reviews the environmental authorization. Investors and property companies: funds and real-estate firms that finance or own the buildings.

How to tell the players apart: developer or operator?

The developer prepares the land and the building (plot, permits, grid connection), then sells or leases it; the operator runs the servers inside and sells hosting to customers. The analogy: the property developer and the hotelier — one builds the hotel, the other brings it to life. Can one company do both? Yes, and it is even the dominant model among the big players: Data4, Equinix, Digital Realty develop and operate their campuses. Pure developers tend to be newcomers from real estate: H&DC, Prologis, Goodman, Segro, Altarea — they come from logistics or property development. A site announced by a developer is therefore, by nature, a real-estate project, not a data center in service.

Who talks to the town, and who never does?

Five roles — and above all: who talks to the territory, and who never does. The developer assembles the deal — land, permits, grid connection — builds, then sells or leases. It's the developer who meets the mayor, early on; often they vanish once the permit is granted. The owner holds the walls (a property fund, a REIT): they talk to no one locally, but they collect the rent. The operator sells the service — leasing rooms, power and connectivity to clients: it's the name on the scorecard, and the territory's contact once the site is live. The facility manager runs the building day to day (cooling, power, maintenance), often an invisible subcontractor — but it's who handles the noise and the backup generators, i.e. what disturbs neighbours. The end customer leases the rooms or the compute (a bank, a hyperscaler, a neocloud reselling further on): it never talks to the town and its name is rarely public. At the big players, the first three roles are often the same company — Data4, Equinix, Digital Realty develop and operate their campuses; among real-estate newcomers they're separate. What it changes for an elected official: the one promising jobs and recovered heat is not the one who will deliver them, and the one truly consuming is almost never in the room.

What are the stages of a data center project, from announcement to operation?

In France, broadly: 1) announcement and, sometimes, a prior public consultation; 2) filing — building permit and, most often, an environmental authorization under the classified-installations regime (ICPE); 3) review: opinion of the environmental authority, then public inquiry; 4) decision: permit granted by the mayor (or by the préfet in some cases), environmental authorization by prefectoral order; 5) appeal periods; 6) construction, often in phases; 7) commissioning, then inspections during operation. Stages can overlap, and a project can stop at any of them. The public's rights at each stage: Rights and timeline.

What is a “moratorium” on data centers? Does it stop a project?

The word covers very different things. A political motion or council resolution calling for a pause has, on its own, no legal effect on an authorization already granted. There is, however, a legal tool: the sursis à statuer (deferral of decision). When a local urban plan is being drawn up or revised, the municipality can postpone its answer to a permit application, for two years at most (French planning code, art. L.424-1). A general nationwide pause would require a law or regulation. On this site, a moratorium is reported as a sourced fact, never as a score.

Who can stop a construction site? What can the administrative court decide?

Several routes, with different effects. The administrative court: an appeal against a permit or authorization does not, by itself, stop the works. To suspend them quickly, a référé-suspension (interim suspension) is needed, granted only when there is urgency and serious doubt about legality (administrative justice code, art. L.521-1). On the merits, the judge can annul the authorization, or give the project owner time to correct a defect that can be fixed (planning code, art. L.600-5-1). These rulings can in principle be appealed, then taken to the Conseil d'État. The mayor or the préfet: they can order works carried out without authorization, or not in compliance, to be halted (planning code, art. L.480-2); for a classified installation, the préfet can serve formal notice on the operator, then suspend the activity (environment code, art. L.171-8). The project owner, finally, can suspend or abandon for financial, technical or commercial reasons.

Can a suspended project, or one under moratorium, restart?

Yes, in several cases: the deferral of decision ends (at its term, the municipality must decide within two months if the applicant confirms the application); a suspension is lifted by the judge; the authorization is regularized; an annulment is overturned on appeal; or the owner files a new application. Conversely, a building permit expires: as a general rule if works do not start within three years, or are interrupted for more than a year; this period can be extended twice by one year (planning code, art. R.424-17 and R.424-21). A stalled project is therefore not necessarily an abandoned one — which is why we never remove it without a source.

What does “announced” mean on a page?

That the data center is not in service yet: a project made public, under permitting or under construction. Until it is in service, it gets no final score but a provisional range, computed on what the territory bears.

Who are you, and how do you stay independent?

An independent observatory founded by Franck Bardol. The method is open, scored entities do not pay, and how we operate is detailed in the Open model section.

Suggest a project, or reach the observatory.

Contact