A readout from the workshop on digital sovereignty in practice: catalogues are not the scarce thing – shared evidence is

Workshop readout · November 2026

A readout from the workshop on digital sovereignty in practice: catalogues are not the scarce thing – shared evidence is.

Author: Stefane Fermigier (Abilian)

Event: Digital sovereignty in practice: measurability and solution-finding, Berlin, 2 November 2026

Reporting: Under Chatham House rules – no view is attributed to anyone

A summary of the session for those who were there and those who were not. For context, a closing section describes the EuroStack Solution Finder, which we are developing with funding from SPRIND and which was discussed during the workshop alongside other catalogues. Annex A prints the criterion register we keep. Annex B points to the “Dear Board” letter from EuroStack, which several passages below quote.

The short version. Price has stopped being the obstacle. A buyer is held back by capability and trust, well ahead of availability or price. Catalogues already exist – what is missing is shared evidence a buyer can act on. Share the evidence, keep the verdict local.

1Context

We are reporting on the closed-door workshop of Digital sovereignty in practice: measurability and solution-finding, a side event of the Franco-German Forum for the Future at the Digital Summit in Berlin on 2 November 2026. The Global Government Technology Centre hosted it at the GovTech Germany venue.

Around forty people were on the confirmed list. Public administration accounted for much of it, mostly, if not only, German or European: the Federal Chancellery, the federal ministry for digital transformation, the Land of Berlin, ZenDiS, a German embassy, and the Commission’s open-source programme office. The French participants came from industry, from professional associations, from consulting and from investment. The remainder were research institutes, foundations and venture capital. Several of the catalogues and assessment frameworks discussed below were represented by the people who run them.

The afternoon ran in two stages. It opened with parallel breakout sessions, each posed as a proposition to agree or disagree with:

  • Screening and audit methods for the digital dependencies of businesses. “Do you agree: we need a Franco-German governance, as a preliminary step for a European one?”
  • Catalogues and pooling approaches for digital sovereign solutions. “Do you agree: we need a catalogue of catalogues?”

The groups then reconvened for an hour on corporate perspectives and ways of pooling effort, under four questions:

  1. Real-world dependencies. When have digital technology dependencies created tangible business risks, costs, or limitations to strategic choice?
  2. Screening and catalogues. Would the approaches presented today have helped in these situations? What would need to change to make them genuinely more useful?
  3. From awareness to action. What is still missing to turn identified dependencies and available alternatives into actual procurement and deployment decisions?
  4. Collective action. Where could organizations pool requirements, purchasing power or implementation efforts, and who would be willing to explore a concrete next step?

The sections below follow the substance of the afternoon and not its timetable:

Put to the session Reported in
Real-world dependencies, and from awareness to action The points of agreement; Replies to the change-management objection
The catalogue proposition, and screening and catalogues The catalogues on the table; Measuring sovereignty; The question behind the question
The governance proposition On governance of the audit method
Collective action Next steps without a new institution

Reporting is under Chatham House rules throughout, so no view is attributed to anyone.

2The points of agreement

The catalogue proposition did not draw a straight yes. The view assembled over the afternoon was that catalogues are not the scarce thing. Shared evidence is. A buyer is held back by capability and trust, well ahead of availability or price.

Participants reached that conclusion through a sequence of agreements, in the order they arrived.

Price has stopped being the obstacle. Several participants went further: European options are frequently cheaper and lose anyway, which is also the finding of Cheaper and Losing (Fermigier 2026c). The blockers named were capability, organisational risk and the difficulty of change management. One case stood in for the rest: the impossibility of bringing the identity management you already run to a hyperscaler. Identity is the layer with the highest exit friction we have measured, its largest cost component being re-integration (Fermigier 2026f).

Evidence a buyer can act on is the scarce thing. That rules out a couple of ways of building a catalogue. A directory whose main work is to show that a European alternative exists has little left to contribute, because participants treat that question as answered. A directory built on a price comparison has less still. A buyer lacks everything after the shortlist: what leaving the incumbent costs, where the interfaces are, and whether the supplier can take on a customer of that size.

A catalogue informs, and procurement verifies. A catalogue helps a buyer find options it did not know about and tells it where to look. The checks a tender requires stay with the buyer. A catalogue owes its readers a statement of what each entry passed to be listed, so a buyer knows what it still has to verify for itself.

3Replies to the change-management objection

The diagnosis stopped at “it’s about change management”, with “so it’s about regulation” as the only move past it. Both readings leave a board waiting, on the grounds that migration is hard. The difficulty is real. Replies to it exist. They arrived in the session as passing mentions and deserve more room.

The business case can be made without appealing to sentiment. This is the objection boards actually raise: a migration decision cannot rest on patriotism, and fiduciary duty appears to point the other way. The “Dear Board” letter, from EuroStack, opens on exactly those words and replies in five moves. Your supplier sees everything you do. AI is not yet a trap you are inside, so where it runs remains your decision. The business case has to include cost escalation on whatever is hardest to replace. Unbundling the estate buys negotiating leverage and a second source. And the decision has to stay revisitable.

The third of those is where the argument turns, because the cost of staying still is measurable. Cigref, the French association of large IT buyers, found the prices of cloud and software services hardest to leave rising by nearly 9% a year over three years. The VMware repricing after the Broadcom acquisition is the same mechanism in a single case. A board that treats migration as the only risky option is comparing it with a baseline that is itself escalating. The premium falls hardest on whatever cannot be replaced.

The difficulty becomes an assignment rather than a mood. Appoint a Chief Resilience Officer, reporting to the board on a measured dependency baseline, with a tested exit and a modelled exit cost expected to fall each year. Demand export rights in open formats and caps on price increases at every renewal. Keep identity, keys, device management and data under the customer’s control. None of that requires a decision to migrate. All of it preserves the option.

And the sequence is written down. “Shifting to sovereign digital supply” (Fermigier 2026i) was written for EuroStack and published by it in September 2026. It addresses the practical questions in seven parts: scope, how far AI belongs in the move, building the business case, structuring contracts, renting or building, forming a consortium, and what an investor should require of the businesses it funds. Slides accompany it.

Together these cover the ground participants identified as missing, which is the step from a listing to a decision.

4The catalogues on the table

Five catalogues were named over the afternoon. Most participants came expecting fewer.

Catalogue Scope
The catalogue of the French CSF Numérique de Confiance1 Trusted French software and digital solutions, under public-private governance
EuVAD’s Smart Register European IT providers, reviewed by an office before publication
The DIGIT OSPO catalogue Public-sector open source across nine countries
The European Champions Alliance cybersecurity mapping Over 1,300 vendors in one sector
The EuroStack Solution Finder European digital solutions, with a published sovereignty measure

Each of these was built for a particular buyer, under admission rules of its own, with a particular idea of what counts as evidence. That variety is a reasonable response to the fact that a German ministry, a French mid-cap and a cybersecurity investor are not looking for the same thing.

5Measuring sovereignty

The breakouts ran in parallel, so whether the other group assembled a comparable list of assessment frameworks is not something this readout can say. There are more of them than there are catalogues. A register we keep for our methodology work holds twenty-three, counting the ones internal to our platform. These are the nineteen with a published result:

Framework Owner Measures Standing
Cloud Sovereignty Framework, SEAL scale European Commission the supplier Award criterion since 2025
Union assurance levels, CADA Annex II European Commission the supplier Proposal in trilogue
IRN v1.1 Digital Resilience Initiative the buyer Live
Toetsingsinstrument Soevereiniteit Clouddiensten DICTU, Netherlands the supplier Official, January 2026
C3A BSI, Germany the supplier Official, April 2026
SecNumCloud 3.2, chapter 19 ANSSI, France the supplier Live since 2022
EUCS sovereignty requirements ENISA the supplier Draft, not adopted
Sovereign and Resilient Cloud Services CISPE the service Live, April 2026, 40+ services declared
Gaia-X Compliance Document 24.06 Gaia-X AISBL the service Live
Kriterien zur Bewertung Digitaler Souveränität ZenDiS, Germany the buyer Consultation to May 2026
openCode badges ZenDiS with BSI a repository Live
Digital Sovereignty Readiness Assessment Red Hat the buyer Live
Cloud Sovereignty Framework Self Assessment SUSE the supplier Live since January 2026
Sovereignty Score Posetiv the supplier Live, 37 providers pre-scored
European Sovereign Stack Standard, SML levels Schwarz Digits and STACKIT the supplier Vendor scheme, 2026
EDSO standard EDSO the supplier Private scheme
Index of Sovereignty Bechtle the buyer Commercial, from Q1 2026
EDRIX, European Digital Resilience Index Abilian Econ Lab a member state2 Live, v2.0, quarterly
FRIX, its French sector edition Abilian Econ Lab a French sector3 Live, editions since March 2026

The remaining rows are the Solution Finder’s own working tools: the pillar specification behind the index described at the end of this note, the screening gate, the submission forms and a black-box prober. The register is open. Anyone who knows of a framework missing from it is welcome to say so. Both indices above are at edrix.eu, with the method published and the archive available on request.

The column recording what each framework measures divides the field: thirteen score a supplier or a service, a handful score a buyer’s own portfolio, and the last two score neither, taking a member state or a sector as the unit. The standing column runs from a live award criterion to a private scheme whose criterion catalogue is unpublished, which sets how much weight a buyer can put on each.

A companion register goes criterion by criterion. It holds ninety-four distinct criteria drawn from eighteen of these frameworks. Coders working blind from a written rulebook recorded which decision about the service each criterion bears on and how strong a piece of evidence meeting it is. They also recorded the strongest channel that can check it, and whether it keeps its meaning once suppliers know it is scored. Annex A prints it in full.

So a catalogue of catalogues already existsThe table above is its framework half. Annex A is its criterion half. Both are published here. They grew out of necessity: specifying an index responsibly means knowing what the field already asks, and merging corpora means learning how each source admits an entry. Whether the pair should become a maintained service, kept current and versioned for other people to build on, is the question now on the table.

6The question behind the question

The question holds two proposals. One of them merges listings. The other merges scores.

Merging listings is the easy half. Entries can be pooled tomorrow. Their meaning cannot, because each catalogue admitted its entries under its own rules. In a combined list, one provider may have been office-reviewed, a second self-declared and a third editorially selected, and nothing on the page distinguishes them for the buyer.

Merging scores is harder than it looks. Two of the Commission’s own scales place the same supplier at opposite ends. In Measure Twice, Buy Once (Fermigier 2026e) we cross the SEAL scale of the Cloud Sovereignty Framework with the Union assurance levels of the Cloud and AI Development Act, question by question. A service recognised at the top of one scale can score the bottom of the other. Neither scale contains the other. Of the thirty-nine framework questions that can lower a service’s level, seventeen are reached by nothing in the Regulation at all.

There is also an arithmetic version of the problem, which needs no disagreement between drafters. An index that publishes the worst of five dimensions as its band reports a lower one when one dimension is split, then a higher one when they are merged back. That is a property of the aggregation rule and no defect in anyone’s judgement, and it leaves a federated score difficult to defend in a tender.

Three things lie underneath both proposals and could be shared instead: one way of identifying a company, one way of grading the strength of a piece of evidence, and published admission rules, so a buyer reading a merged list knows what each entry passed. Share the evidence, keep the verdict local.

That proposition is partly testable today, because the frameworks turn out to agree more than the debate suggests. A dimension-by-dimension crosswalk is set out in Scoring the Seller (Fermigier 2026g). It runs against the Commission’s framework, the Dutch DICTU instrument and the German BSI C3A criteria, all of them public-sector. It also runs against the IRN, a client-side index published by the Digital Resilience Initiative, which is private. On jurisdiction, operations, technology and data, every supplier-side framework corresponds fully. Building a Supplier-Side Sovereignty Index (Fermigier 2026k), published since the session, extends the comparison to eighteen frameworks mapped onto one basis of 94 criteria.

One dimension diverges, the one the identity example belongs to. None of them scores the economic channel as a first-class concern: what leaving costs, and who can raise the price once a customer cannot move. The Cigref figure quoted in Replies to the change-management objection measures exactly that channel, from outside any of these frameworks.

7On governance of the audit method

The other breakout proposition asked whether Franco-German governance is the preliminary step towards a European one. It ran in parallel with the catalogue group, so what follows is our reading of it and no report of that discussion. In the programme it appeared under screening and audit methods for the digital dependencies of businesses, so the thing to be governed is a method. The catalogues are only where its evidence comes from.

Pin that object down, because it differs from the one a catalogue discussion tends to assume. An audit of a business’s digital dependencies measures the buyer’s portfolio. That is the client-side object, measured by several frameworks in the table above. The IRN of the Digital Resilience Initiative is the one recommended to boards in the “Dear Board” letter. A supplier-side index such as ours measures the other object. They compose pillar by pillar. A per-supplier score feeds a buyer’s picture without standing in for it, because the buyer adds asset criticality, available alternatives and concentration across the portfolio. Governing the method therefore means governing a criterion register, an evidence grading and an aggregation rule. It does not mean governing anyone’s verdicts.

Three further observations follow, in ascending order of inconvenience.

The pairing is already where the work is. The catalogue of the French CSF Numérique de Confiance is the corpus we are currently merging into the Solution Finder. EuVAD, whose register is German and Austrian, offered during the session to share its information. Whoever wants to test catalogue interoperability in practice will find that the two live cases are French and German.

A starting vocabulary exists in Franco-German form already. The joint paper of 17 June 2026 defines digital sovereignty and sets out six dimensions to score a product or a service against. They are the capability to implement and enforce, the capability to design and deploy and use, economic value creation, protection of data, substitutability and interoperability, and infrastructure resilience. Six is the right size for a shared criterion register. It also arrives with an endorsement from both governments that no register drafted from scratch would have. We map each of the six onto a measurable object in The Weakest Link and the Final Say (Fermigier 2026j), naming the analysis that computes it, so the step from a definition to a register is shorter than it looks. The same point bears on the criterion-count question below, since six dimensions with a few criteria under each is a smaller object than a flat list of thirty-two.

The aggregation rule has to be settled before the committee is. In the same commentary we set out the design risks in the joint paper, of which the first two apply directly. It aggregates compensatorily where dependence behaves as a weakest link, so a strong score on most dimensions can mask a chokepoint on the last. It also rests on self-assessment, which is exposed to washing in exactly the way a screening problem predicts when compliance is cheap to claim (Fermigier 2026h). A joint framework that fixes the committee and leaves the arithmetic open will produce numbers that cannot be defended in either country’s procurement.

8Criterion count and discriminating power

The criteria discussion arrived at a set of thirty-two. In the same session, companies said they do not want more complexity, and that they would rather see incentives from the public side than further obligations. Those findings are hard to reconcile, and splitting the difference will not do it.

There is a measured reason to doubt that more criteria buy more discrimination. In Ask Me Something Hard (Fermigier 2026a) we tested every criterion of eighteen sovereignty frameworks for whether it separates a foreign-controlled supplier from a European one. Three fifths of them do not. A criterion that every supplier passes adds cost to the respondent, adds length to the questionnaire and moves no decision.

One way out of the tension was offered to the room. Anyone is welcome to borrow it:

  • A small set of strong criteria does the scoring. In the room we described eleven criteria over five dimensions. The published method (Fermigier 2026k) settles on fifteen criteria that every coder graded as strong evidence, read into a profile across six dimensions. A buyer sees the weak dimension directly, which an average would hide; Counting Dependencies (Fermigier 2026d) shows why.
  • A wider disclosure layer is reported beside the verdict without diluting it. Security certifications, environmental and supply-chain attributes belong here. The assessor gets completeness. The respondent does not pay for it in the score.
  • Coverage is published next to every score. A supplier that discloses nothing is still listed and still scored, with a visibly thin result. Answering is how it improves its own number.

The last of those is an incentive of the kind the companies asked for, and the public purse pays nothing for it. A catalogue cannot hand out subsidies. It can make completeness visible and let the market read it.

9The buyer segments

The catalogue discussion closed on the question of which buyer a catalogue is built for. In one direction the position is not in doubt: any European buyer with a procurement decision in front of them should be able to act on the same catalogue. The differences lie in what each kind of buyer needs from the evidence, and in how each one is reached.

The public buyer has a mandate already. Public administration was much of the attendance, so this buyer was also the best represented. A catalogue serves this buyer before the tender: sounding the market, finding candidates it had not heard of, and preparing requirements. The award criteria and the verification belong to the procurement itself, as the Commission’s framework, a live award criterion, shows. A catalogue helps most by stating what each entry passed to be listed, so the buyer knows which facts it must still check.

Large companies are willing and do not trust a small European supplier to support them at scale. Their obstacle is supplier capacity, which is measurable and which no sovereignty score addresses: references at comparable scale, support footprint, financial standing, continuity arrangements. The “Dear Board” letter addresses this buyer. Its Chief Resilience Officer proposal is the organisational form of the same gap.

In smaller companies the decision is personal. A head of IT decides in mid-size companies, and in smaller ones it is often the chief executive, who wants a simple verdict, very little effort, and a reason to read sovereignty as commercial advantage. The channel matters as much as the tool, because an SME does not arrive at a sovereignty catalogue, it arrives at its trade association. A simplified version of the index distributed that way was proposed during the session.

One correction for us. Our own buyer profiles are all IT or procurement roles, so the owner-manager who makes this call personally is missing from them and will be added.

10Next steps without a new institution

Each of these is small, each was prompted by something said during the session, and none of them needs a governance body before it can begin.

Five next steps

  1. A shared evidence layer, written as a short specification. Company identification, evidence grading, published admission rules. Open for comment, versioned, and adopted by whoever finds it useful. No existing format covers this ground. publiccode.yml, the nearest candidate, was designed to describe open-source projects released by public administrations, which is a different object from a commercial solution a buyer is evaluating.
  2. The six Franco-German dimensions as the starting criterion register. The definition is published. Both governments stand behind it. The mapping from each dimension to a measurable object is already drafted. Starting there costs less than agreeing a fresh list of thirty-two criteria, and the governance discussion acquires a concrete object.
  3. One catalogue exchange, done properly, as the worked example. EuVAD offered during the session to share its information. That offer will be taken up. Eighty reviewed German and Austrian entries set against a larger and differently distributed corpus is small enough to be tractable and different enough to test the identifier question on real data.
  4. A requirements format several buyers can sign. Pooled demand was one of the proposals made during the session, in both its top-down and bottom-up forms. The theory is in Buyers’ Clubs (Fermigier 2026b), including how to price a shared commitment and how to identify the anchor buyer who makes it credible. The undrafted piece is a document a procurement officer can put their name to.
  5. A machine interface, so interoperability gets demonstrated. A catalogue that publishes a documented API can be queried by the other catalogues and by the AI assistants buyers are starting to use. Ours publishes sixteen paths today. A working interface is the most persuasive reply available to the catalogue-of-catalogues question.

11About the EuroStack Solution Finder

A public directory of European digital solutions, designed and built by Abilian, financed for the time being by SPRIND, and endorsed by the EuroStack Foundation e.V. It runs on European infrastructure in France and Germany.

It currently holds 1,719 solutions across seven technology layers and sixty-four functional categories, and a merge with the catalogue of the French CSF Numérique de Confiance takes it towards ten thousand. It also maps 173 non-European incumbents as a first-class way in, because a large share of real buyer intent takes the shape “a European alternative to X”. Anyone can propose an entry or a correction, and review precedes publication. The API publishes sixteen documented paths.

The sovereignty measure is the Resilience & Sovereignty Index. The deployed version scores eleven criteria over five pillars and reports the lowest pillar, a weighted score and a coverage figure, with no single blessed number.

The methodology for its next version is now public, as Building a Supplier-Side Sovereignty Index (Fermigier 2026k). It reports each service as a profile of six dimensions, one for each decision a buyer needs to keep or take back. They are who can read the data, who can alter the service, who can stop it, who sets the terms, who decides what it becomes, and who can maintain or replace it. Another family records what the service contributes to European capacity, starting with the share of each euro paid that stays in Europe. Every reading is graded by the channel it came through, so a supplier’s declaration counts for less than a registry extract or a test. The index builds in no gate: a buyer who wants a minimum on one dimension sets it. The paper also states where the deployed version departs from it, and closing that gap is the work of the next release.

Two things we do not yet claim. The structural admission criteria are specified in full and applied by hand, with enforcement in the product still to come; the test applied systematically today is the jurisdictional one, that the provider is established in the EU or an associated country. And the scores are provisional, with public evidence still thin for most entries, which is why coverage is published beside every one of them.

1 The letters CSF are shared by unrelated bodies in this readout. In France they stand for Comité Stratégique de Filière, here the committee for logiciels et solutions numériques de confiance, whose first contract with the French state runs from 2025 to 2028 and covers a catalogue of solutions alongside a directory of trusted suppliers. At the Commission they stand for the Cloud Sovereignty Framework, which is an assessment framework and appears in the table of frameworks.

2 EDRIX scores the twenty-seven member states from publicly observable data alone, needing no cooperation from anybody measured: developer density, desktop and browser shares, and the hosting, mail and DNS arrangements behind the top thousand domains of each national registry and behind each government’s own sites. Germany leads the September 2026 edition at 7.59, ahead of Finland and Austria. France is twelfth at 5.91, down from seventh in June and the steepest fall of that quarter, although EDRIX notes that under June’s scoring rules it would have ranked eleventh. Almost every rank change in the edition comes from the private-sector pillar.

3 FRIX applies the hosting measurement to France sector by sector, 964 domains across seven sectors in the September 2026 edition. The spread is where the finding lies: central administration scores 92.5% and public hospitals 71.9%, while the SBF 120 large caps reach 29.9% and the French Tech Next 120 scale-ups 18.3%.

EuroStack Initiative Foundation e.V. · eurostack.eu

References

  • EuroStack (2026). Dear Board. Open note to company boards, signed “Europe’s digital industry”.
  • France and Germany (2026). Franco-German Joint Paper on Digital Sovereignty. 17 June 2026. A definition of digital sovereignty and six dimensions to assess a product or service against.
  • Fermigier, S. (2026a). Ask Me Something Hard: Why Three Fifths of Europe’s Sovereignty Score Cannot Tell Suppliers Apart. Abilian Econ Lab, 18 September 2026. econ.lab.abilian.com
  • Fermigier, S. (2026b). Buyers’ Clubs for the Open-Source Flow: Collective Provision in the FLOSS Value Network. Abilian Econ Lab, working paper v0.9. econ.lab.abilian.com
  • Fermigier, S. (2026c). Cheaper and Losing: A Layered Model of Cloud Dependence and the Limits of Subsidy. Abilian Econ Lab, working paper v0.8. econ.lab.abilian.com
  • Fermigier, S. (2026d). Counting Dependencies: Aggregation Rules and What Official Dependency Indices Identify. Abilian Econ Lab, working paper v0.1. econ.lab.abilian.com
  • Fermigier, S. (2026e). Measure Twice, Buy Once: The Cloud and AI Development Act’s Union Assurance Levels Against the Commission’s Own SEAL Scale, Criterion by Criterion. Abilian Econ Lab, 18 September 2026. econ.lab.abilian.com
  • Fermigier, S. (2026f). Measuring the Stack: Paths, Exit Costs, and the Instruments That Reach Them. Abilian Econ Lab, 18 September 2026. econ.lab.abilian.com
  • Fermigier, S. (2026g). Scoring the Seller: An Economic Foundation for a Supplier-Side Sovereignty Index. Abilian Econ Lab, working paper v0.7, 16 September 2026. Unpublished; available on request.
  • Fermigier, S. (2026h). Screening for Sovereignty: Optimal Procurement when Compliance is Cheap to Fake and Costly to Verify. Abilian Econ Lab, working paper v0.20. econ.lab.abilian.com
  • Fermigier, S. (2026i). Shifting to Sovereign Digital Supply: a “How To” Approach. Policy paper v1.2, Abilian Econ Lab, 22 September 2026; version 1.0 published by EuroStack on eurostack.eu, September 2026. econ.lab.abilian.com · Slides
  • Fermigier, S. (2026j). The Weakest Link and the Final Say: An Economic Commentary on the Franco-German Joint Paper on Digital Sovereignty. Abilian Econ Lab, working paper v0.3, 30 September 2026; also published in French. econ.lab.abilian.com
  • Fermigier, S. (2026k). Building a Supplier-Side Sovereignty Index for the EuroStack Solution Finder. Abilian Econ Lab, working paper v0.2, 9 November 2026. econ.lab.abilian.com

The Fermigier items above are Econ Lab working drafts, position papers and notes. None is peer-reviewed, and the index is at econ.lab.abilian.com. The methodology of the Solution Finder’s index is published as Building a Supplier-Side Sovereignty Index (Fermigier 2026k). Comments are welcome.

Annex A. The criterion register

This annex prints the ninety-four criteria drawn from eighteen sovereignty frameworks, coded under the rulebook of Building a Supplier-Side Sovereignty Index (Fermigier 2026k). It replaces the version circulated in September, in which one coder’s judgements and an approach the published method has since superseded decided every verdict.

Two coders on each of two AI models coded every criterion blind from the written rulebook. Each coding records four things.

  • Dimension. The decision about the service the criterion is evidence on: who can read or use the data, who can alter the service, who can stop it, who sets the terms, who decides what it becomes, who can maintain or replace it. Criteria that bear on none of these are placed under contribution to European capacity, what the buyer can see, environmental performance, programmes and labels, or the buyer’s own posture.
  • Strength. Whether meeting the criterion is strong evidence, weak evidence or no evidence about who holds that decision. The column Strong in counts the coders who graded it strong. The fifteen criteria graded strong by all four, in bold, are the core of what the index collects.
  • Evidence. The strongest channel that can check the criterion, from a public register down to the supplier’s own declaration.
  • Holds up. Whether the criterion keeps its meaning once suppliers know it is scored (durable), loses it to suppliers who learn to meet it on paper (fragile), or stops separating anyone once all comply (self-defeating).

Each criterion appears under the dimension most coders gave it. Criteria about visibility, the environment and programmes are collected and displayed and never scored. The buyer’s own posture belongs to a client-side index. Eight criteria that asked two questions at once are split into parts marked a and b. The three groups of criteria that read one fact, the ownership chain, continuity and interface openness, are counted once. The coders are AI agents, and human coders are the next test of the rulebook.

GroupCriteria
Access: who can read or use the data22
Integrity: who can alter the service7
Continuity: who can stop it12
Terms: who sets the price and conditions5
Direction: who decides what it becomes5
Capability: who can maintain or replace it7
Contribution to European capacity5
Visibility: what the buyer can see10
Environmental performance4
Programmes and labels5
The buyer’s own posture13
Total95

Access: who can read or use the data

IDCriterionStrong inEvidenceHolds up
B04Exposure to extraterritorial compelled-access law, directly or through the parent4 of 4documentfragile
B25Customer custody of encryption keys4 of 4testdurable
B73Identity-provider control plane and root of trust under customer or reference-jurisdiction control4 of 4testdurable
M-controlOwnership and the control chain (B01, B02, B70, B84)4 of 4public registerfragile
B71Administrative access technically restricted to paths inside the reference jurisdiction, with exceptions authorised and logged1 of 4documentfragile
B06Legal and organisational separation between the parent and third-country subsidiaries0 of 4documentfragile
B14Location of the infrastructure, assets and control plane that deliver the service0 of 4documentfragile
B15Location and jurisdiction of operations and support personnel0 of 4documentfragile
B16Personnel screening, citizenship and security clearance0 of 4documentfragile
B21aIdentity and jurisdiction of subcontractors and subprocessors0 of 4documentfragile
B21bAccess that subcontractors and subprocessors have to customer data and to the service0 of 4documentfragile
B24Location of customer data, metadata, telemetry and backups, with no third-country fallback0 of 4documentfragile
B26Encryption at rest and in transit0 of 4document–
B28Retention, secure deletion and proof of erasure0 of 4declarationfragile
B29Customer data not used to train or fine-tune AI operated from a third country0 of 4declarationfragile
B30bAI models and pipelines hosted in the reference jurisdiction0 of 4documentfragile
B47Security certification held, with date, body and scope0 of 4public register–
B48Security operations and incident handling located in the reference jurisdiction0 of 4documentfragile
B74Duty to contest, notify and record third-country requests for access to data0 of 4documentfragile
B78Training and certification of personnel with access to critical systems0 of 4document–
B88Customer’s exclusive rights over their data: no provider access or use without authorisation0 of 4documentfragile
B91bLinks with foreign governments0 of 4documentfragile

Integrity: who can alter the service

IDCriterionStrong inEvidenceHolds up
B72Control of the build, packaging and code-signing chain3 of 4documentfragile
B37aControls preventing remote features from tampering with or disrupting the system2 of 4documentfragile
B37bIntegrity of updates0 of 4testfragile
B43Provenance of firmware and embedded code0 of 4documentfragile
B44Where, and by whom, software is developed, packaged, distributed and updated0 of 4documentfragile
B50Vulnerability handling: disclosure policy, patch commitments, CVE transparency0 of 4public interface–
B82Ingress control on a dedicated secure network area, with known-vulnerability screening0 of 4document–

Continuity: who can stop it

IDCriterionStrong inEvidenceHolds up
B05Exposure to export-control regimes that can restrict use or transfer4 of 4documentfragile
B45Single points of dependency on vendors, facilities or technologies outside the reference jurisdiction4 of 4documentfragile
M-continuityContinuity and operational autonomy (B17, B76, B83, B86)4 of 4testdurable
B67Jurisdiction of the naming and delivery layer: DNS, registrar, top-level domain, CDN, certificate authority2 of 4observationdurable
B85Vendor terms cannot withdraw a critical asset except for non-payment or breach2 of 4documentdurable
B22Incident-response commitments and service levels0 of 4documentfragile
B35Backup and replication adequate to a major disruption0 of 4testdurable
B40Independence in high-performance computing capabilities0 of 4documentfragile
B42Origin and manufacturing location of hardware components0 of 4document–
B77Documented and tested strategy against supply-chain disruption and sanctions: alternatives, switchover, stock0 of 4documentfragile
B81Redundant, independent connectivity, including a provider in the reference jurisdiction0 of 4observationfragile
B92Long-term financial viability of the supplier0 of 4documentdurable

Terms: who sets the price and conditions

IDCriterionStrong inEvidenceHolds up
B32Exit cost, and a tested migration path4 of 4testdurable
B03Governing law and competent court of the customer contract2 of 4documentdurable
B41bEgress fees2 of 4documentself-defeating
B89Termination without penalty if the service loses its qualification or conformity1 of 4documentdurable
B41aPricing transparency and renewal terms0 of 4documentfragile

Direction: who decides what it becomes

IDCriterionStrong inEvidenceHolds up
B07bGovernance locks against enclosure of the product2 of 4documentdurable
B11Capacity of stakeholders in the reference jurisdiction to shape the roadmap1 of 4documentfragile
B07aChange-of-control risk, and governance locks against acquisition0 of 4documentfragile
B08Sources of capital, and financial independence from outside the reference jurisdiction0 of 4documentfragile
B39Management of technological obsolescence0 of 4documentfragile

Capability: who can maintain or replace it

IDCriterionStrong inEvidenceHolds up
B18Capacity to maintain and patch without the originating vendor4 of 4documentdurable
B20Documentation, source and know-how sufficient for long-term autonomy4 of 4documentdurable
B31Data portability: export formats and procedure4 of 4testself-defeating
B36Open-source availability and licence, with rights to audit, modify and redistribute4 of 4public interfacedurable
M-interfaceInterface openness (B33, B34)4 of 4testself-defeating
B19Availability of skills in the reference jurisdiction to operate the technology0 of 4documentfragile
B69Health of the open-source projects relied on: activity, maintainers, releases, adopters0 of 4public interfacefragile

Contribution to European capacity

IDCriterionStrong inEvidenceHolds up
B12aWhere the intellectual property was created4 of 4documentfragile
B30aAI models and pipelines developed and trained under control in the reference jurisdiction4 of 4documentfragile
B68Contribution to the open-source communities the service depends on3 of 4public interfacefragile
B12bWhich jurisdiction holds the intellectual property1 of 4documentfragile
B09Economic footprint in the reference jurisdiction: investment, jobs, value creation0 of 4documentfragile

Visibility: what the buyer can see

IDCriterion
B49Customer or authority control over security monitoring and logs
B23Disclosure of incidents and breaches
B27aVisibility of data flows
B27bAccess logs available to the customer
B38Architecture transparency: components, dependencies and flows
B46Supply-chain transparency: SBOM, dependency list and audit rights
B51Independent audit rights with full access
B75Transparency and explainability of the AI models the service uses
B79Advance notification of changes in control, or of anything affecting conformity
B80Disclosure to the customer of where each category of data is stored and processed

Environmental performance

IDCriterion
B52Energy efficiency of the infrastructure
B53Hardware reuse and end-of-life treatment
B54Environmental impact measurement and reporting
B55Sourcing of renewable or low-carbon energy

Programmes and labels

IDCriterion
B10Participation in, and alignment with, the reference jurisdiction’s strategic programmes
B13Compliance with the reference jurisdiction’s sectoral regulation
B87Sovereign configuration offered by default or easily activated, and clearly labelled
B90Interfaces and first-level support in the language of the reference jurisdiction
B91aRespect for fundamental rights and Union values

The buyer’s own posture

IDCriterion
B56Formalised digital governance, with sovereignty in the strategy
B57Executive sponsorship and budget for resilience and dependency management
B58Governance of self-service digital and AI tools
B59Data and AI governance
B60Concentration of critical spend on one supplier
B61Contractual negotiating power and capability to control providers
B62Mapping of data, uses and obligations
B63AI doctrine, with a map of uses and vulnerabilities
B64Continuity of critical telecommunications and connectivity
B65Energy dependency and datacentre continuity
B66Business continuity under environmental hazards
B93Procurement and architecture policy that favours open standards, open source and alternatives
B94Dependency risk management: dependencies recorded, assessed and steered

Annex B. The Dear Board letter

Reproduced in full. The letter is an open note from EuroStack to company boards, signed “Europe’s digital industry”. It is quoted in Replies to the change-management objection above and the text is its own, not ours.

Dear Board

We understand you have some sympathy for the European “sovereignty” cause but in the end “a business migration decision cannot be based on sentiment or patriotism”. “We have fiduciary duties to our shareholders, and to care for Europe’s growth is not in our statutes”. “We need a hard business case”. “Unless shifting makes us clearly better off, we won’t embark on it”. “If the economic case is not clear-cut it makes no sense”. And best of all “We are a global company we cannot afford to upset our US customers if we are vocal on European sovereignty” — while at the same time “Donald Trump will be gone in two years, all will go back to normal”.

Let us give you a “narrative” for why you should think differently, right now, that does not depend on sentiment and patriotism nor on caring specifically for Europe’s growth (important as the latter is to us as citizens, yes?).

1. Your IT supplier sees absolutely everything that you do.

Your vendor, that you have been with for years and allowed to lovingly build a proprietary ecosystem around you, processes your data and runs your processes. This means all of your strategic information (customer records, prices and margins, contracts and all correspondence) is on their systems and carried on their infrastructure. You control absolutely nothing of it. It’s a very unequal commercial relationship: they provide a service, you hand over all your strategic information but there is no secluded sandbox in which your “stuff” is protected. They can see all of it. They can use it (yes, we know, data processor vs data controller etc.). All that intelligence is there and does not vaporize in the ether. And, they serve your competitors too. Another channel through which value leaks is their government can absolutely order them to hand all of it over on request – and you have zero protection (US CLOUD Act reaches you however many Matrioshka-style local subsidiaries are built in Europe to claim they are “sovereign”). Is this okay with you? Do you think it is a tenable position for European serious businesses going forward?

2. AI is not a legacy you are yet trapped into: you decide now where your AI runs, and do not choose to be handing over the crown jewels forever.

AI is going to make things much much worse. Whoever runs the AI assistant the company adopted holds what it learns (reads every document to build answers; absorbs everything about how the company works to learn what the company does). The urgent drive to introduce AI to manage and enhance our one area of excellence (industrial domain capabilities and specialist data), which is the envy of the world, will shift by definition the entire body of know-how, IP, expertise and data to model developers (the labs) and the hyperscalers selling compute bundled with models. Integrating US models with the factory floor is the sure recipe for a one-way traffic where our industrial wealth will be expropriated and passed on. Automating a business process into an AI platform is going to be harder to move than any office suite, and the crown jewels will no longer be “ours”. The AI agents and their developer and compute suppliers will know what the company knows. This will perpetuate our position of vassalage into the future forever more. Europe lost the war on consumer facing businesses. Are we prepared to lose the war on our industrial strength as well? This is a systemic issue but also one that directly affects your future. You are not yet captive or captured by your AI supplier. This should be an easy decision. There are alternatives. Keep options open, Control is key.

3. The “business case” for migration must include cost escalation where replacement is costly, and for anything that depends on it.

What keeps you pinned in place is that you are trapped in a single ecosystem with multiple layers of dependencies. What seemed to be convenience when you allowed someone to develop for you an integrated system has turned into a giant trap. Because of integration and interdependencies, the bit that is most difficult to replace is the one where you are charged the biggest premium, and will continue to be, going forward. Moreover anything that depends on that “bit” is caught in the net. Yes it is easier to stay put. Less hassle. But you face cost escalation on it, which vendors are placing right at the level of the cost that you would incur to switch. Indeed cost increases have been fast and projected to be steep: a study by Cigref (the association of large French IT buyers), found the prices of cloud and software services that are hardest to leave rose by nearly 9% a year in the last 3 years. The saga of VMware’s virtualisation software after acquisition by Broadcom (severalfold price increase) speaks for itself. Price increases are announced for bundles including new features, whether or not one uses them (Microsoft 365). The approach must be to push the “unbundling” logic as far as possible: what can we unwind, and what are the alternatives? Some pieces are already commoditised and cheaper – for computing capacity and storage, European vendors charge a fraction of the largest US suppliers (based on published prices): here, moving should be an ordinary sourcing decision that would pay for itself.

4. “Unbundling” your “IT estate” gives you leverage in negotiations and second-sourcing benefits.

At the very least, identifying and scoping out alternatives gives you leverage on price in future negotiations with your incumbents. And which essential line of business runs with no second-sourcing alternative – as a benchmark or a fallback? For whatever reason: a failure, an adverse commercial decision, a government’s order. Yet IT is the only one function where companies are happy to run on a single supplier for office suite, identity system, cloud and now AI, with no second source and no exit plan. The classic excuse is “it doubles the cost” – but it is not true – it does not need to be a complete duplicate of the estate. There are alternatives, and “recovery plans” can be structured as a test for second sourcing. In 2001 Nick Carr wrote on the Harvard Business Review: IT Does Not Matter – so everyone forgot about it and treated it like a commodity for accountants and IT departments. Can you really be so cavalier today? Can you see it as a non-strategic decision and leave that out of your responsibility? Or decide it’s a strategically important decision but one you will only take if it demonstrably makes you a profit? In the current geopolitical circumstances?

5. Even if you do not switch now, this must be a choice you can revisit. And it must get easier, not harder, over time.

You should appoint a CRO – Chief Resilience Officer. Reporting to the Board with a concrete measure of your dependencies. There are methodologies for this (aDRI). Can do a quick assessment in a few weeks. Ask whether exit has been tested. And whether there is concrete modelling of the exit costs, which should fall every year.

You need to demand that the configuration of key functions that administer the IT estate (access directory, encryption keys, device management, data) is under your control, and such that choice of suppliers is reversible. A job for the CTO working with the CRO.

You need to walk away from “AI transformation” that pegs you with one model and one option. Open weights. The fetish of frontier models should not infect all use cases.

You need to build in every contract renewal the right to export data in open formats with no charge; and caps on price increases: the cost of compute is falling rapidly, and should translate in falling unit prices, not increases. Even if you do not leave, your posture must be you have negotiating power and can revisit.

Happy to discuss!

Europe’s digital industry

Share the Post:

Related Posts