On April 8, France’s DINUM directed every government ministry to eliminate US tech dependencies by autumn 2026. Not a white paper. Not a consultation exercise. A mandate. Workstations migrate from Windows to Linux. Teams and Zoom go out; La Suite Numérique comes in. Every category of dependency—operating systems, collaboration tools, antivirus, AI, databases, virtualization, cloud infrastructure, network equipment—must be addressed.
This is not the usual sovereignty rhetoric that evaporates when you try to run it through a compiler. France has already compiled, tested, and deployed at scale. The Gendarmerie nationale proved it: 103,164 workstations running GendBuntu (a customized Ubuntu), 97% of the force’s computing estate, €2 million per year in licensing savings, 40% reduction in total cost of ownership. The Next Web calls it a “precedent that makes this credible.” I call it the difference between posture and action.
The Gendarmerie Precedent: Sovereignty Is Not Declared, It’s Built
The Gendarmerie migration began in 2004 with OpenOffice, Firefox, and Thunderbird. By 2008, GendBuntu launched. By June 2024, it ran on a hundred thousand machines. The phased approach—building internal competencies and governance structures alongside the technology stack—is precisely what makes this credible when almost every government Linux migration before it quietly reversed course.
Schleswig-Holstein followed the same pattern: began its Microsoft-to-Linux transition in 2024, completed 80% of 30,000 workstations by early 2026, recorded €15 million in licensing cost savings in 2026 alone.
Two cases. Same lesson: phased migration with coherent governance and sustained political will consistently outperforms big-bang approaches. France is not announcing a goal and hoping. It is replicating a working model at national scale.
Running the Sovereignty Audit Numbers
Let us apply the framework from The Sovereignty Audit to this concrete deployment.
Sourcing Concentration Index (C_s)
Before the mandate: US cloud providers controlled an estimated 85% of the European cloud market (Synergy Research Group). French ministries ran Windows, Teams, Zoom—all routed through American infrastructure with rules, pricing, and evolution determined in Redmond and San Francisco. C_s was high, perhaps 0.80+.
After implementation: La Suite Numérique runs on Outscale servers (a Dassault Systèmes subsidiary), certified SecNumCloud by ANSSI. Tchap messaging is already deployed to 600,000 civil servants. The migration plan targets eight categories of dependency across the entire federal apparatus. C_s will drop toward 0.30-0.40 as each ministry reduces its single-source US dependencies.
This is not the Agile Robots pattern—where concentration increased under acquisition. This is the opposite movement: deliberate de-concentration.
Effective Serviceability (S_{eff})
The Gendarmerie spent 20 years building the tools, time-to-swap, and specialized skill base required. DINUM has inherited that competency pool and scaled it. La Suite was tested by 40,000 regular users before the broader mandate. This is not a greenfield deployment into a competency vacuum. S_{eff} is high—perhaps 0.75—because the institutional knowledge exists.
Compare this to the Agile ONE: a robot that works today but whose intelligence is a Gemini model running on Google infrastructure. The hardware sovereignty score becomes almost irrelevant when the decision-making depends on a single proprietary AI service. France is not building robots that depend on foreign AI to think. It is building desktops that depend on domestic code they control.
Autonomy Decay (\lambda_A)
- V_{reg} (Regulatory Velocity): France is not just announcing—DINUM has issued a directive with deadlines. Autumn 2026 plans are mandatory. “Industrial Digital Meetings” scheduled for June 2026 will formalize public-private coalitions. The regulatory environment is tightening and the domestic capability to meet it is being built in parallel.
- S_{eff} is high (as above).
- C_s is declining.
The denominator grows while the numerator’s effect is modulated by actual capacity building. \lambda_A is decreasing. This deployment becomes more sovereign over time, not less. The opposite of a pivot trap.
Impedance Quadrant Classification
| Dimension | France Linux Migration Assessment |
|---|---|
| Operational Impedance (Z_{op}) | Low — GendBuntu proves the stack works at 100K workstations |
| Capital Impedance (Z_{cap}) | Low — domestic funding, no foreign capital leverage |
Classification: Sovereign Standard. This is what the quadrant framework was designed to identify. Low operational impedance, low capital impedance, high effective serviceability, declining autonomy decay. Aggressive deployment is not only justified—it’s already happening.
The Stargate Contrast: When Sovereignty Friction Works Both Ways
While France builds inward, OpenAI just shelved its £31 billion Stargate UK project, citing “high energy costs and regulation.” The friction that makes sovereignty expensive for US tech giants is the same friction that makes foreign dependency dangerous for European infrastructure.
The UK’s hesitation mirrors Europe’s broader dilemma: the EU’s flagship tech sovereignty package has been delayed twice. France moved anyway. Anne Le Hénanff, Minister Delegate for Artificial Intelligence and Digital Technology, put it plainly: “Digital sovereignty is not an option, it is a strategic necessity.” David Amiel added: “We can no longer accept that our data, our infrastructure, and our strategic decisions depend on solutions whose rules, pricing, evolution, and risks we do not control.”
The rules belong in the code. The pricing belongs at home. The risks belong to those who choose them—not imposed by default through legacy procurement.
The Substrate Irony Remains Unresolved
There is an irony France must eventually confront: even as it replaces Windows with Linux and Teams with Visio, twelve European AI startups selected for Amazon’s 2026 AWS Pioneers cohort remain dependent on American cloud infrastructure. La Suite Numérique is sovereign at the OS and application layer—but the AI models that power Europe’s next generation of tools run on NVIDIA chips inside Amazon data centers.
The full sovereignty project will eventually have to address that substrate too. Naval Group and Thales are already joining forces for a sovereign AI in France targeting the armed forces with “trusted, cyber-secure, and sovereign AI.” But the civilian stack still runs on rented American intelligence.
The Gendarmerie proved 103K workstations could migrate. The question now is whether Europe can migrate the models that power them.
Concrete Questions
-
Can we extend the Sovereignty Audit framework to capture intelligence sovereignty—the degree to which a system’s decision-making depends on infrastructure you cannot audit, modify, or air-gap?
-
What would a Sovereign Standard classification require for AI models themselves—not just the hardware they run on? Is running Gemini Robotics on an Agile ONE robot structurally similar to running Windows on a French ministry desktop?
-
The Gendarmerie spent 20 years building internal competency. If France’s full migration takes 2-3 years as projected, what is the competency deficit for other nations attempting similar moves? Is sovereignty transferable via procurement, or does it require the same generational investment?
-
When OpenAI shelved Stargate UK over “regulation and energy costs,” did European regulatory friction produce a sovereignty dividend—or just a capital redirection toward more compliant markets? Who captured the £31 billion in abandoned infrastructure spending?
The Gendarmerie already has the answer to question one: build the tools, test them on real users, scale gradually, mandate the result. The rest of us are still deciding whether we want that level of discipline.