Sovereignty as a Strategy, Not a Cost

The compliance case for sovereignty is well rehearsed and slightly dull. The commercial case is newer: a market being created by procurement, and a supplier position that can now be priced.

The instrument. €180m · Awarded. The Commission's own sovereign cloud framework, April 2026, with a minimum sovereignty level required of bidders. 48 criteria. Eight weighted categories. A measurable property becomes a procurable one.
In production. 1 · Mistral. Three billion euro raised in September 2026, at a twenty-one billion euro valuation. 2 · LaSuite. More than five hundred thousand French civil servants each month. 3 · Italy's PSN. More than five hundred and seventy administrations, on contracts running to 2035.
Not yet law. 4 levels. The sovereignty scale proposed in the Cloud and AI Development Act, with external audit from the second upward.

Procurement is making the market

In April 2026 the European Commission awarded a sovereign cloud framework worth up to one hundred and eighty million euro over six years, to four consortia led by European providers. Bidders had to reach a minimum sovereignty level on a published scale, and most winners reached the level above it [1]. The lasting part of that award is the framework, not the contract. The Commission's framework scores providers against forty-eight criteria in eight weighted categories, and the Commission's own remark on publishing it was that before the framework existed, digital sovereignty could not be measured at all [1].

A measurable property becomes a procurable one, and a procurable one becomes a market. Analysts expect worldwide sovereign cloud infrastructure spending to grow by around a third in 2026, with European spending growing by more than eighty per cent again in 2027 and overtaking North America in the same year; governments lead, followed by regulated industries and critical infrastructure [2]. A board can dislike the politics of this and still notice the procurement signal underneath it.

The alternatives are already in production

The European alternatives are no longer a slide of logos. Mistral raised three billion euro in September 2026 at a twenty-one billion euro valuation, the largest equity round the company says a European technology firm has completed, and earlier in the year raised eight hundred and thirty million euro of debt for a forty-four megawatt site outside Paris housing nearly fourteen thousand accelerators [3]. OVHcloud reported quarterly revenue of two hundred and ninety million euro with public cloud growing at twenty per cent [3].

The public sector deployments are larger than most boards assume. France's shared public administration suite is used by more than five hundred thousand civil servants each month, with its messaging service at six hundred thousand [4]. Italy's national strategic pole has onboarded more than five hundred and seventy administrations under contracts running to 2035 [4]. Germany's Deutschlandplattform runtime was awarded to a domestic provider and a German software firm for up to thirty million euro, hosted in Germany and Austria [4]. A European messaging stack certified for restricted German government material runs at well over a hundred thousand users in the armed forces alone [4].

Policy is turning into procurement rules

The Commission proposed a Cloud and AI Development Act in June 2026. It sets a four-level sovereignty framework, from European establishment and data location at the first level to a prohibition on third-country control of any component at the fourth, requires independent external audit from the second level upward, and creates a common procurement framework so that public buyers can act together. It is a structured preference rather than a blanket exclusion, with exemptions where no service is available or the cost would be disproportionate [5]. It is also not yet law: trilogues are not expected before 2027, and the obligations would apply a year after adoption [5].

The direction of travel does not wait for that. The European Parliament adopted a technological sovereignty report by 471 votes to 68 in January 2026 [6]. France issued a binding interministerial directive in April 2026 covering operating systems, collaborative tools, cloud and AI, requiring every ministry to submit a dependency reduction plan [6]. Switzerland committed nine million francs to move three thousand federal workstations to an open-source suite through 2027, while stating openly that the suite cannot yet fully replace what it is replacing [6]. That candour is worth more to a board than any of the enthusiasm around it.

Public investment. €20bn · Mobilised. For AI gigafactories, with seventy-seven expressions of interest across sixteen member states. 19 · AI factories. Deployed on European supercomputers, with thirteen further antennas.
One migration. 44,000 mailboxes. Moved by Schleswig-Holstein, with fifteen million euro of licensing savings reported after switching costs.
The gap. 15% · Share. European providers' share of their own market, flat since 2022, against seventy per cent for the three largest.

Public money is arriving at scale

The European public investment is now large enough to change what is available to buy. Nineteen AI factories are deployed on European supercomputers with thirteen further antennas, and a separate gigafactory programme has mobilised twenty billion euro, drawing seventy-seven expressions of interest across sixteen member states and sixty sites, with a call for tender issued in July 2026 and first construction expected in 2027 [7]. An open model programme funded under the Digital Europe Programme is building a family of models covering every official European language, with the data, models and training recipes published openly [7].

Alongside the money sits a procurement shift. The Commission's open source strategy, presented in June 2026, commits to procurement guidance that evaluates open source bids fairly, a maintenance instrument for the components everyone depends on, and openness as a design principle in digital investment decisions [8]. Public procurement is around a seventh of European gross domestic product, so a preference expressed there is not a gesture [8]. For a board, the practical question is whether the organisation is positioned to sell into that, buy from it, or neither.

The organisations that moved published their numbers

Schleswig-Holstein moved more than forty-four thousand mailboxes and a hundred and ten million items off its incumbent suite, reports fifteen million euro of licensing savings after switching costs, and has put nine million of that back into the open-source ecosystem it now depends on [9]. Aarhus is moving half of its twenty-four thousand staff, has migrated sixty systems to a European host, and published one workload whose annual cost fell from eight hundred thousand to two hundred and twenty-five thousand Danish kroner [9]. Denmark's digitalisation ministry moved its office suite and said plainly that it would revert if the complexity proved unmanageable [9].

The instructive counterexample is the arrangement marketed as a national AI capability for Germany, which delivers a United States model to public sector users through a domestic subsidiary running on the technology of a United States platform [10]. Sovereign at the label is not sovereign at the layer, and a board buying either should know which one it has bought.

The gap is real and worth stating

European providers held fifteen per cent of their own regional market in the first half of 2025, flat since 2022 and down from twenty-nine per cent in 2017, while the three largest United States providers held seventy per cent. The analyst view is blunt: cloud is a game of scale, and no European company has yet made bets of that size [11]. The Commission's own impact assessment puts the European data centre capacity gap at roughly three gigawatts today, widening towards nineteen by 2036, and Europe attracts around a twentieth of global venture capital [11].

The gap in open models is sharper still. Since the start of 2026 the leading open-weight models have trailed the closed frontier by an average of about four months, and the leading open-weight models are Chinese rather than European or American [12]. European developers remain inside a United States accelerator software ecosystem that is expensive to leave [12]. And in June 2026 a United States export-control order caused a frontier provider to disable two models for customers worldwide: no ownership rule, sovereignty certificate or citizenship requirement would have preserved that access [13].

Below the application layer the dependency runs deeper again, with hyperscalers now carrying about three quarters of international bandwidth and participating in two thirds of planned subsea cable projects [13].

4 months. The average lag of the leading open-weight models behind the closed frontier since January 2026.
Demand. 61% of CIOs. Western European technology leaders intending to increase their use of local providers.

The return is not in the cost line

The cost case is the weakest part of the argument and should be treated that way. There is no credible independent comparison of total cost between hyperscalers and European providers; the figures circulating are trade assertions or list price comparisons of unlike things. The one area with a real and documented gap is specialist accelerator capacity, where dedicated providers price well below hyperscaler list rates for comparable hardware [14].

The return sits in four less glamorous places. Eligibility: sovereignty levels are entering tenders, and a supplier that cannot evidence one is being scored down before the commercial conversation starts. Negotiation: a priced, credible second supplier changes a renewal conversation whether or not it is ever used. Continuity: the ability to keep operating through a suspension, a sanction or a legal order is worth what the outage would have cost. Demand: in a survey of two hundred and forty-one Western European technology leaders, sixty-one per cent wanted to increase their use of local providers and fifty-three per cent expected geopolitics to restrict their future use of global ones [14]. Those customers are already writing the questions into their own procurement.

The first step. 1 workload. Every organisation that has moved started with one, at production scale.

Ninety days of useful work

1. Score the three most material suppliers against the Commission's published criteria, and record the level each would reach. It costs nothing and it produces a number.
2. Price one credible European alternative for one real workload, and keep the quotation on file with its date.
3. Run one proof of concept at production scale, not a pilot on a spare laptop. The organisations that have moved all started with one workload.
4. Put a sovereignty question into the next tender you issue, and see what your own supply chain can evidence.
5. Separate the layers in your own description of what you buy: who runs the service, whose technology it uses, and who ultimately controls both.
6. Decide which workloads are strategic enough to justify a second supplier, and say so in the minutes rather than in a strategy deck.
7. Check whether the public programmes now funding European capacity create a customer, a partner or a competitor for this organisation, and say which.

This is a commercial decision, not a political one

Sovereignty as a virtue is a political argument, and boards are right to be wary of it. Sovereignty as an option has a price, a counterparty and a payoff, which makes it a commercial decision like any other. Europe will not build a full alternative stack this decade, and no serious person suggests otherwise. What is available now is narrower and worth having: a measurable standard to score suppliers against, a set of providers credible enough to price, and customers who have started asking the question first. A board that can answer it has an advantage that does not depend on any regulation passing.

Twelve questions, and the answers that should worry you

Notes and sources

1. European Commission, sovereign cloud framework award, 17 April 2026; Cloud Sovereignty Framework v1.2.1, October 2025, and Sovereign cloud framework explained, 1 June 2026.
2. Gartner, forecast of worldwide sovereign cloud infrastructure spending, 9 February 2026.
3. TechCrunch, 8 September 2026, and Data Center Dynamics, 30 March 2026, on Mistral's funding and Paris site; OVHcloud third quarter results, 25 June 2026.
4. La Suite numérique service pages, September 2026; Il Sole 24 Ore, 17 December 2025, on the Polo Strategico Nazionale; Schwarz Digits, 10 June 2026; Element case studies on the Bundeswehr messenger.
5. European Commission, proposal for a Cloud and AI Development Act, 3 June 2026; Freshfields analysis of the framework and timetable, 2 September 2026.
6. European Parliament report on technological sovereignty, adopted 23 January 2026; French interministerial directive of 8 April 2026; Cybernews, 4 September 2026, on the Swiss federal decision.
7. European Commission, AI Continent Action Plan milestones, 9 April 2026; AI gigafactories programme, July 2026; OpenEuroLLM under the Digital Europe Programme.
8. European Commission, EU Open Source Strategy, 3 June 2026; Chatham House, European preference signals wider change in EU doctrine, 5 March 2026.
9. European Commission Open Source Observatory on Schleswig-Holstein, and on Aarhus, 18 June 2025; The Record, 13 June 2025, on the Danish ministry. Savings figures are the public bodies' own.
10. OpenAI, OpenAI for Germany, 24 September 2025.
11. Synergy Research Group, European cloud provider market share, 24 July 2025; International Center for Law and Economics on the Cloud and AI Development Act impact assessment, 2026.
12. Epoch AI, open and closed model capability gap, 29 May 2026; Bruegel, Europe needs a strategy to close the artificial intelligence compute gap, 19 May 2026.
13. Nextgov/FCW, June 2026, on models suspended after a United States export-control order; TeleGeography data on hyperscaler bandwidth and subsea cables, fourth quarter 2025.
14. InfoWorld, 14 April 2026, on accelerator pricing; Gartner survey of 241 Western European technology leaders, May to July 2025, reported 13 November 2025.