How many systems in the world, and conclusion - 6/6

Machine-translated page

This English version was produced by automatic AI translation. Only the French version is reviewed by the author, so it remains the authoritative reference. Spotted a mistake or an awkward turn of phrase? Please report it by opening a pull request — contributions are welcome.

How many systems in the world, and what share for Windows?#

One last exercise, openly speculative but illuminating. One cannot census every system deployed on Earth; one can, however, set orders of magnitude, clearly separating the sourced from the assumed.

Four sourced anchors (primary figures):

PopulationNumberKernel familySource
Active Android devices> 3 billionLinuxGoogle, 2025
Active Apple devices~2.5 billionBSD/DarwinApple, 2026
Active Windows devices~1.6 billionNTMicrosoft, 2026
Connected objects (IoT)~21 billionLinux / RTOSIoT Analytics, 2025

To this we add estimates, for lack of recent public figures. Servers, first: the fleet numbers in the tens of millions of physical machines (worldwide shipments run at around 12 to 13 million a year), and far more virtual instances; system-side, Linux is largely dominant there — a 2017 IDC report already gave, across all deployments, Linux ~68% against Windows ~32% (Windows remaining No. 1 only in the paid market). Then the detail of the desktop (Mac, ChromeOS, Linux).

The calculation — and it must be taken for what it is. Adding up a connected sensor and a computer is debatable: many objects carry only a tiny RTOS, sometimes no system at all. But even playing along, and being deliberately generous with Windows (1.6 billion client machines, plus a deliberately inflated assumption of 0.3 billion server instances — well beyond the real fleet, as we have just seen) and cautious everywhere else, Windows’s share of all deployed systems falls to:

  • about 23% if one ignores IoT completely;
  • about 6% if one counts the ~21 billion connected objects.

In every scenario, Windows is a minority — at best less than a quarter, more likely under 10%. The world’s computing runs, by count, very largely on the Unix family (Linux first, then Darwin and BSD); Windows keeps an important role there, but one confined to the client desktop.

Sources:

Conclusion#

Three findings emerge from this survey.

Windows’s retreat is real, but it is above all a failure to expand. The segments lost in the strict sense can be dated precisely, between 2012 and 2024: the enterprise firewall, television, the tablet, the smartwatch, automotive infotainment, the phone, general-purpose embedded, virtual reality. Yet the bulk of the weakening does not come from there. It comes from the fact that almost every market born in the last twenty years escaped Windows: the smartphone, the connected TV, the container, the cloud, the AI-training cluster did not exist as mass markets in the mid-2000s. Windows contested several of them and won none. The world grew around it more than it pushed it out.

The winner is not Linux, it is the Unix family. Linux reigns unchallenged over supercomputers, dominates web servers and the majority of the cloud’s compute cores. But the two systems that replaced Windows Phone rest one on a Linux kernel, the other on Darwin, whose hybrid kernel blends a Mach microkernel and a BSD layer; the firewalls that succeeded Forefront TMG run on FreeBSD; the consoles that outsell the Xbox rest on FreeBSD or on a proprietary kernel; the infotainment that replaced Windows Embedded Automotive runs on QNX, a proprietary Unix; and certified avionics never fell to Windows or Linux, but to proprietary real-time systems. Writing “Linux won” would be inaccurate in several places. Writing that the Unix family, in the broad sense, won is accurate almost everywhere.

The strongholds that remain share one thing in common, and it is not the operating system. Enterprise directory, industrial supervision, computer-aided design workstations, cash machines and interactive terminals, the office desktop: in each of these cases, what holds Windows is not Windows. It is the software catalogue above it — Active Directory and its group policies, CATIA and SolidWorks, supervision software, the office suite, line-of-business applications certified by a vendor. Windows there is less chosen than inherited by whoever needs what runs on it. The position remains solid, and it must be stated as sharply as the retreats: about 71% of the desktop, the directory of nearly 90% of large enterprises, and some of the market’s most profitable segments. But it is a position that has become derived, held by application lock-in and organisational inertia rather than by the system’s own qualities.

From these three findings emerges a fairly clear dividing line: one never really chooses an operating system, but the catalogue of software that runs on it. Where a user sees the machine and cares about that software, the incumbent ecosystem wins: on the desktop, Windows stays held by its catalogue (office suite, games, line-of-business software), exactly as the smartphone is held by the Android and iOS app stores. But wherever the end user does not know which system they are using — in a router, a television, a cash machine, a compute cluster — no consumer catalogue is at stake: the decision falls to a technician who weighs cost, licence and technical control. These are precisely the criteria that made the Unix family win.

Suggest an edit

By Yanal-Yves FARGIALLA • Updated on August 18, 2026 (AI-assisted writing, final review by the author)
Unless otherwise noted, this content is licensed under CC BY-SA 4.0. CC BY-SA 4.0