France Ditches Windows for Linux: What We're All Missing

500,000 computers. That's the number France put on the table when they announced their Linux migration. Half a million government machines moving to a free operating system. When I read it, I had to read it twice — not because it seemed impossible, but because I lived firsthand what happens when someone issues that same decree without thinking about what's underneath.

And what's underneath is a complete disaster.

The France Linux Migration: The Announcement Everyone Celebrates Without Reading the Fine Print

The French National Gendarmerie started this years ago — they're the case study everyone cites. Ubuntu, LibreOffice, Firefox. It worked. And now Macron's government is pushing a massive expansion. The headlines are beautiful: digital sovereignty, independence from Microsoft, license cost savings.

All correct. All real.

But there's something the tech articles don't tell you: the Gendarmerie is a military organization. Centralized IT with actual decision-making power. And — this is the key detail — internally developed applications that they themselves control. That's not your average government agency. That's not an Argentine municipality. That's not the provincial agency where I spent six months of my life between 2022 and 2023, trying to help with a migration that ended up being one of the most instructive failures I've ever lived through.

The Frankenstein No Decree Can Patch

I got called in for a consulting gig at a provincial agency. I'm not naming names because that's not the point — the point is that this situation is representative, not exceptional. The initial brief was simple: "we want to migrate 200 workstations to Ubuntu to save on Windows licenses." Projected savings: $40,000 USD per year. Reasonable.

What I found when I started auditing the infrastructure:

Active Directory with 15 years of history. Security groups inherited from previous administrations. Group policies nobody documented. Users with permissions for things that no longer existed. The digital equivalent of a city that grew without a zoning plan.

Printers. Oh god, the printers. Three different Ricoh models with drivers that only existed for Windows XP. A 2008 Kyocera that the accounting department used for specific forms, with a driver embedded in a 32-bit DLL that wouldn't compile on anything anymore. When I asked if they could replace the printer, the answer was: "that printer is in the State inventory — to decommission it we need a formal file that takes eighteen months to process."

The ERP. This is where everything collapsed. The management system — handling everything from payroll to purchase orders — ran on a web application that required Internet Explorer 11. Not Edge in compatibility mode. Real IE11. The ERP vendor was a mid-sized company that had won the tender in 2011 and whose maintenance contract renewed automatically. When I called them to ask if they had a roadmap for modernizing the app, the answer was: "we're evaluating a migration for 2026."

The corporate antivirus. Only had a Windows agent. Two years left on the license contract.

And the list kept going. Every layer you peeled back revealed another dependency tied to Windows — not by preference, but by the historical accumulation of decisions made without considering future lock-in.

The Real Technical Problem: It's Not the OS, It's the Full Stack

Here's the thing that political announcements never capture: Windows isn't just an operating system. In any organization with more than 50 people, Windows is the infrastructure. It's the directory, the policy management, the SSO, the printer integration, the ERP client, the PDF viewer with the tax authority's digital signature baked in.

Migrating the OS without migrating the stack is like swapping an engine without touching the transmission. The car isn't going anywhere.

What you actually need for a France-scale Linux migration to work — or any government migration — looks like this:

# Real dependency audit before touching anything
# I built this script for the agency — it finds executables calling Windows-only components

#!/bin/bash
# Scan all network share access and application dependencies
# Run this BEFORE planning anything

echo "=== Auditing critical dependencies ==="

# Check what applications are registered and their requirements
wmic product get name,version > apps_inventory.txt
echo "App inventory saved to apps_inventory.txt"

# Search for IE references in shortcuts and configs
grep -r "iexplore" /c/Users --include="*.lnk" --include="*.url" 2>/dev/null \
  | tee ie_dependencies.txt
echo "IE dependencies in ie_dependencies.txt"

# Audit GPOs applied to current user
gpresult /H gpo_report.html
echo "GPO report in gpo_report.html"

# List installed printers with their drivers
Get-Printer | Select-Object Name, DriverName, PortName \
  | Export-Csv printers_audit.csv
echo "Printers in printers_audit.csv"

This looks basic. Because it is. But at the agency where I worked, nobody had run this audit before announcing the migration internally. The announcement came first. Reality came later.

The alternative I ended up recommending — and partially implementing — was a layered migration strategy:

# Phase 1: Replace applications, not the OS
# Install free equivalents on top of Windows first
# Measure adoption and problems BEFORE changing the OS

# LibreOffice instead of Microsoft Office
winget install TheDocumentFoundation.LibreOffice

# Thunderbird instead of Outlook (where there was no critical Exchange dependency)
winget install Mozilla.Thunderbird

# Firefox as the primary browser
winget install Mozilla.Firefox

# Phase 2: Migrate the directory to something Linux-compatible
# Samba AD or migration to FreeIPA — only IF you have real time
# Don't do this in parallel with the OS change

# Phase 3: Only then consider changing the OS
# And only on workstations where you've validated there are no broken dependencies

Final result after six months: we migrated 23 workstations out of 200. The 23 that had no critical legacy application dependencies. The rest stayed on Windows. The savings came out to $4,600 USD annually — not $40,000. And that's with six months of consulting work that obviously wasn't free.

Failure? Depends how you look at it. I see it as the honest result of a problem nobody wanted to audit before promising outcomes.

The Mistakes That Repeat in Every Government Migration

Mistake 1: Confusing the OS with the complete infrastructure. Already covered this. But it's worth repeating because it's the most common mistake and the most expensive one.

Mistake 2: Underestimating the cost of human change. The accounting clerk who's been using Excel for 20 years isn't a technical problem — she's a training problem, a trust problem, a workflow problem embedded in muscle memory. LibreOffice Calc is excellent software. But if that person has to Google how to do something they used to do with Ctrl+Shift+Something, that person is going to hate Linux before giving it a real chance.

Mistake 3: The ERP vendor as a single point of failure. This is structural in the Argentine state — and I'd bet it's the same in plenty of European states too. If the critical management system is controlled by a third party with a long-running contract, you cannot migrate. Full stop. No Linux distribution is going to save you from that.

Mistake 4: Ignoring the hardware. Printers, scanners, fingerprint readers, digital signature devices. Linux has excellent hardware support for modern hardware. Government hardware has a 15-year expected lifespan and a procurement process that makes replacing it nearly impossible in the short term.

Mistake 5: Announcing before auditing. This one hurts the most because it's purely political. The migration gets announced as a victory, expectations get set, and when the technical reality shows up, the project already has political inertia that makes it hard to be honest about the obstacles.

This pattern, by the way, shows up everywhere. The same thing I saw with the OS migration I've seen with AI agent projects that promise to automate everything without understanding the real context. The announcement always outpaces the implementation.

FAQ: France Linux Migration and Government Migrations in General

Can France actually migrate 500,000 computers to Linux? Technically yes, politically yes, but not overnight and not without massive investment in middleware, training, and legacy application replacement. The Gendarmerie pulled it off over 10 years with dedicated resources. A full-state migration is a completely different order of magnitude.

Why did the Gendarmerie migration work when others fail? Three reasons: centralized control over their applications (developed in-house), an organizational structure with the authority to enforce change, and time — they didn't do it in a year. When an organization depends on third-party software it doesn't control, the room to maneuver shrinks dramatically.

What's the real cost of a Linux migration in the public sector? The license savings are real but partial. The actual cost includes: audit consulting (which nobody budgets for), user training (which nobody budgets for), development or replacement of incompatible applications (which nobody budgets for), and lost productivity during the transition (which nobody budgets for). In my experience, in year one the migration costs more than it saves.

Is Linux Desktop ready for the average corporate user? Yes and no. Ubuntu LTS, Fedora, Linux Mint — these are solid systems for general use. The problem isn't the operating system, it's the ecosystem around it. If your critical applications are modern and web-based, the migration is relatively straightforward. If you depend on legacy proprietary software, the OS is the least of your problems.

Does it make sense for Argentina to follow this path? In theory, absolutely — digital sovereignty and license savings are real arguments. In practice, the Argentine state has the same inherited infrastructure Frankenstein, plus an additional layer of fragmentation because every province, every municipality, every agency runs its own stack. A coherent migration would require coordination that has historically been hard to sustain across administrations.

What needs to happen first for a migration like this to work? I'd audit the applications first, not the OS. Complete dependency inventory. Then migrate applications to modern web alternatives while keeping the OS the same. And only when the application stack is OS-agnostic, change the OS. This process takes years, not months. And it requires ERP vendors and government software providers to modernize their stacks — which sometimes means changing procurement conditions to require cross-platform compatibility right in the contract.

France Is Right About the Destination, But the Road Is Longer Than It Looks

I'm not a Linux migration skeptic. I'm a skeptic of the speed and superficiality with which it gets planned.

France has legitimate reasons and real capacity to execute this — they have a more consolidated tradition of government IT, companies like Atos that can provide support at scale, and political will that in this case seems to have continuity. If they do it right, in ten years they'll have an incredible case study.

But "doing it right" means auditing the full stack before announcing dates. It means investing in modernizing legacy applications, not just swapping the OS. It means actually training users — not firing off a YouTube tutorial link. It means having a real plan for the ERP vendor whose product only runs in IE11.

The same honesty you need to plan a post-quantum cryptography migration — which also seems distant until suddenly it isn't — is exactly what you need for a government-scale OS migration. The devil is in the dependencies, not the operating system.

I learned that the hard way with 23 migrated machines out of 200 and six months of work. France is going to learn it at the scale of half a million computers.

I hope they learn it before the announcement. Not after.

Been through something similar? Worked on public or corporate infrastructure migrations? I'd genuinely love to compare notes — especially if you have experience with Active Directory in Linux environments or legacy ERP systems that survived against all logic. Comments are open.