Cour des comptes on Microsoft: office costs per workstation
Cour des comptes on Microsoft: EUR 87 a workstation a year on office tools; open-source cloud tools could avoid EUR 130 million a year against Microsoft 365.
On 22 October 2026, France’s Cour des comptes published an audit of the office and collaboration tools on about 860,000 ministry workstations: an estimated EUR 100 million in 2025, more than half of it on Microsoft licences and services, and EUR 87 per workstation a year on average over 2021 to 2025. It finds free and open-source software (logiciels libres) by far the most advantageous model for the state budget and recommends converging on free cloud tools from 2027, under a migration plan of three to four years at most. The Prime Minister’s reply welcomes all four recommendations.
This post covers the figures, the two estimates, support, the government’s reply, the Swiss angle and what to do. The report, “Les outils bureautiques et collaboratifs au sein de l’État”, its synthèse and the reply are in French only, so translations are ours.
What did the Cour des comptes find on Microsoft and office costs?
The report covers the ministries’ central administrations and regional services, not state operators, social security, hospitals or local authorities. Its figures are estimates, reconstructed by the Cour because several ministries tracked these costs poorly. For 2025:
- EUR 61 million on licences, four-fifths of it to Microsoft, Windows not counted.
- EUR 26 million on services.
- EUR 13 million on staff directly assigned to the tools.
The total excludes operating systems and includes La Suite, the open-source suite of DINUM, the interministerial digital directorate; La Suite has cost about EUR 40 million since 2021. The total is under 2% of the state’s EUR 5.5 billion digital budget and rose 50% from 2021 to 2025, notably through licences: licence and service purchases rose 47%, against 16% for the state’s non-staff operating spending.
A DINUM note of 15 September 2021 ruled out Microsoft 365 for state services over data sensitivity and non-SecNumCloud hosting, with derogations from the Prime Minister for the social ministries, whose Microsoft 365 accreditation runs to the end of 2026, and for Culture, which says it awaits SecNumCloud qualification for Bleu, a Capgemini and Orange company set up in technology partnership with Microsoft (pp. 34 and 35).
What does EUR 87 per workstation measure?
EUR 87 is not a licence price and not a 2025 figure. It is the average annual cost per workstation over 2021 to 2025 of all office and collaboration spending (licences, services and assigned staff, without operating systems), taken over five years to smooth out multi-year licence purchases (p. 49).
| Perimeter | EUR per workstation a year (2021 to 2025 average) | What the Cour says |
|---|---|---|
| All ministries | 87 | Licences, services and staff |
| Interior | 20 | “Sujet à caution” (open to caution): costs may be missing. Nearly 290,000 workstations, Gendarmerie included |
| Culture | 378 | Microsoft 365 and fixed costs spread over 7,700 workstations; Microsoft 365 and associated services: 255 |
The Cour notes that the local version of LibreOffice is free, and cites DINUM’s estimate of EUR 150 to 200 per user a year for a Microsoft Office licence, depending on the ministry.
Where do the EUR 30 to 65 million and EUR 130 million come from?
The two estimates answer different questions and must not be added.
Free office software installed on each computer, at today’s usage
Licence and service spending per workstation at Economy and Finance (EUR 52) and Interior (EUR 15), staff excluded, applied to all 860,000 workstations gives EUR 45 million and EUR 13 million. Against the EUR 76 million spent in 2025 (DINUM excluded), the rounded difference is a potential saving of EUR 30 to 65 million a year (pp. 49 and 55).
An open-source cloud suite instead of Microsoft 365 or Bleu
The cloud the Cour has in mind can be run by the state itself or qualified SecNumCloud (p. 10). DINUM puts a free suite in a SecNumCloud-qualified environment at about EUR 100 per workstation a year; at similar functionality, the Cour sets that against about EUR 250 observed for Microsoft 365 without SecNumCloud hosting. The cost in Bleu “is not known, but would necessarily be higher”. If the figures are confirmed, about EUR 130 million a year would be avoided (pp. 76 and 77).
That is spending avoided against a move to Microsoft 365 or Bleu, not a saving on today’s EUR 100 million. The figures need refining, no administration has yet deployed a free cloud suite, and the transition would take time and staff support. The Cour still expects such a suite to cost less than a proprietary one once migration is done, with a wider gap than today’s between installed proprietary and free suites, because vendors can bill collaborative and AI features dearly: “the financial equation would therefore be favourable for the state” (p. 77).
Is support the real cost of moving to open source?
On the Cour’s evidence, support decides the outcome, and the method is known. Satisfaction is broadly the same whatever the model and is shaped by habit, training and support; LibreOffice scores 87% at the tax directorate, DGFiP. The press release names the point to watch: managing the tools poses no significant difficulty, but moving to collaborative tools needs constant, substantial support.
The digital directorates report similar difficulties with free software (p. 53):
- Resistance to change, including from ministerial authorities who obtain Microsoft Office by derogation.
- Advanced uses, such as macros and pivot tables, hard to reproduce identically.
- Formats and calendars that do not fully interoperate.
- Training, short of time and resources.
Education (Microsoft Office, LibreOffice and Collabora Online) shows the method: targeted training, internal “ambassadeurs”, documentation for real uses, local support, open-source integrators and in-house skills. Culture’s earlier attempt shows what fails: its free suite of the 2000s ended when the minister’s private office kept Microsoft Office and file exchanges suffered. The Gendarmerie shows how to handle the remainder: GendBuntu Linux runs on 72,000 laptops, and 2,500 workstations keep Windows and Office for interministerial exchanges, notably on budgets, and for Windows-only business software (pp. 36 and 37).
With open source, the licence is the cheap line. Support, skills and formats are the work, each with a known method: an operations problem, not a reason to stay. The Cour concludes that free software installed locally brings “significant savings, while satisfying users”, and that migration “remains a realistic prospect” (p. 76).
What does the Cour recommend, and what did the government reply?
The four recommendations, all addressed to the Prime Minister and all from 2027, are proposals, not law.
- Roadmap and governance. Involve the ministries more in La Suite’s roadmap and governance, and bring office and collaboration tools into interministerial governance led by DINUM.
- Development model. Rely more on companies with open-source expertise to develop La Suite.
- Convergence. “From 2027, begin converging on interministerial free office and collaboration tools in cloud mode, and plan a multi-year migration, over three to four years at most, guiding the digital strategies of the ministries and their operators.”
The Cour says three to four years would fit current contract end dates, and wants a detailed study in each ministry now, to review uses rather than replicate every legacy feature and to choose file formats (the RGI framework already recommends ODF and PDF).
The Prime Minister’s reply welcomes all four as serving its goal of a sovereign digital environment “relying in particular on free software and European industrial solutions”. ARIANE, to be created on 1 January 2027 in an overhaul of interministerial digital governance, is to define with the ministries a multi-year trajectory for office and collaboration tools, and a DINUM working group, “Suites collaboratives”, plans a first workshop in autumn 2026 on needs, volumes and contract renewal dates. As of 23 October 2026, ARIANE does not yet exist.
The Cour also cites (p. 77) a National Assembly inquiry commission’s proposal of 15 July 2026: compulsory open source by 1 January 2030 in public procurement by state administrations and operators, public companies and public-service concessionaires. That is not law.
Most ministries say they want to keep Microsoft Office and are considering Bleu once Microsoft software can run there under SecNumCloud qualification. Economy and Finance spent EUR 1.3 million with Bleu in 2025 to carry over existing Office licences, still run locally, at what it says is an equivalent cost; the social ministries expect an operating surcharge of about 10% a year and a EUR 3.7 million migration (pp. 34 and 48). Bleu was not SecNumCloud-qualified at the time of the report, the Cour notes, and dependence on a vendor would remain a question.
State operators sit outside the report’s scope. On 3 November 2026 one of them, the CNRS, announced that it has subscribed to PARTAGE, the collaborative messaging service of RENATER, the French education and research network, as part of its exit from Microsoft. Its press release (in French) says mail for nearly 35,000 staff moves from outsourced Exchange to free software (Zimbra) operated by RENATER, with savings of about EUR 3.5 million a year.
Do the Cour’s numbers hold outside France?
The method transfers, the figures do not: rebuild your own baseline rather than borrow EUR 87.
On 2 September 2026, the Swiss Federal Chancellery launched a programme for a digitally sovereign workplace, testing open-source software, for about 3,000 staff from the end of 2027, in parallel with Microsoft 365 (our analysis; see also EMBAG and open source).
Schaffhausen’s cantonal government gives a fair counterpoint: its answer of 1 September 2026 cites a City of Zurich study that found openDesk not cheaper to run despite free licences, and for its own evaluation points to the extra effort for operation, integration, support, security, training and specialist applications. The city’s release gives no cost figure: it finds openDesk meets many everyday office needs, but dependencies on infrastructure and specialist applications, and a licence that also bundles endpoint management, telephony and security, rule out a full switch for now.
Zurich asked whether one suite can replace the whole Microsoft 365 bundle today; the Cour measured what administrations running free office software at scale spend. Both put the hard part in operations, integration and staff support, not the licence: compare workload by workload, with a support budget.
What to do, and by when
These dates are our suggested timetable, not legal deadlines.
- By 31 December 2026: measure your cost per workstation and list contract end dates. Three to five years of licences, services and assigned staff, operating systems excluded, so it compares with EUR 87; then every enterprise agreement, add-on and hosting contract (for EU cloud, check switching charges).
- By 31 March 2027: inventory dependencies and set a format rule. Macros, Windows-bound applications, calendar exchange with partners and bundle parts (device management, telephony, security) to replace or keep; ODF and PDF by default, and a list of legacy features you will not replicate.
- By 30 June 2027: pilot one workload. Written success criteria, training, champions, documentation, local support and no exemptions for leadership (what a pilot needs to survive).
- From 2027, at each renewal: move workloads in order, reducing licences as they leave, over three to four years.
- Throughout: budget operations explicitly. Name an owner for updates, monitoring, independent backups and the exit plan.
How Pilae fits
With open source, the work is support, operations and migration, and that is the work Pilae takes on: we operate open-source and source-available apps on your premises or on dedicated machines in Pilae Cloud in Switzerland and the EU.
- One workplace, one workload at a time. The sovereign workplace (Nextcloud with Collabora Online, Mattermost or Element, Jitsi Meet, Outline and Keycloak) can start with one app beside Microsoft 365, with Keycloak federating Entra ID for one login.
- A pilot, champions and a rollback. A migration follows a fixed-price written plan: an inventory, a pilot team whose feedback adjusts the mappings, training and runbook, a parallel run, a cutover with an agreed rollback, and training for your champions.
- Operations under contract. Under managed operations, the Pilae Agent plans every update, you approve it and it is applied in your window, under a monthly availability commitment per production app with service credits.
- Backups you can restore. For every app we operate: daily, encrypted before leaving your machine and stored offsite in your chosen country. On the Enterprise plan, an engineer also restores them every month, with the result in your console.
- An exit plan in every plan. If you leave, a full export of your data, configuration and runbooks in open formats, and help moving to another operator or back in-house (exit plan).
The Cour’s figures answer the licence question; support and operations decide whether the saving holds. Mail stays with Exchange Online or your provider, as Pilae operates neither mail nor a phone system, and Collabora’s per-user subscription is priced with you before you commit (Microsoft 365 comparison). To cost your workloads, the support a move needs and the order to move them in, talk to an engineer.
This post is general information, not legal or procurement advice. For the official text, read the Cour des comptes report (PDF, in French).