The Better Windows Than Windows

The world it was born into

By 1985 the IBM PC and its clones ran DOS, a simple operating system Microsoft licensed to IBM and, crucially, to everyone else. DOS ran one program at a time, could use only a small amount of memory directly and offered no protection: a misbehaving program could crash the whole machine. Everyone agreed the PC needed something better. Apple had shown a graphical interface with the Macintosh in 1984. Microsoft was working on Windows, a graphical layer on top of DOS, whose first version in 1985 made little impression.

IBM and Microsoft were bound together. IBM had created the PC standard; Microsoft supplied its software. In August 1985 they signed an agreement to develop the next operating system together. It became OS/2.

The bet

OS/2 was ambitious for a personal computer: multitasking, protected memory so programs could not crash each other, and later a graphical interface called Presentation Manager. It was meant to be the professional successor to DOS, especially for the corporations that were IBM’s core customers. When IBM announced its new PS/2 computers in April 1987, OS/2 was announced alongside them.

The bet was that the combined weight of IBM and Microsoft would define the next platform, as it had defined the first.

Why it made sense

Few people in 1987 would have bet against IBM and Microsoft together. IBM was the largest computer company in the world, with unmatched relationships with corporate buyers. Microsoft controlled the operating system on most PCs. OS/2’s technical design was sound and in several respects more advanced than anything else available for PCs. Corporate customers wanted stability and multitasking.

How it lost

The first problem was the processor. OS/2 was designed for the Intel 80286, which IBM had used in the PC/AT and in some PS/2 models. The 286 had a protected mode, but it ran old DOS programs poorly while in it. The 386, which arrived in PCs in 1986, could run several DOS programs at once in protected mode. Building for the 286 meant OS/2 ran existing DOS software badly, and existing software was what users had. OS/2 1.0, released in December 1987 without a graphical interface, also needed more memory than most PCs had, and memory was expensive.

The second problem was applications. Few were written for OS/2, so few people bought it, so few were written. This is path dependence in its most familiar form.

The third problem was Windows. In May 1990 Microsoft released Windows 3.0. It ran on top of DOS, on machines people already owned, and it made good use of the 386. It sold very well. Microsoft now had a successful product it controlled outright, and its interest in sharing a platform with IBM declined. The partnership broke up over 1990–91. Microsoft’s next-generation work, originally intended as OS/2 3.0, became Windows NT, released in 1993.

IBM continued alone, and did good work. OS/2 2.0, in 1992, and OS/2 Warp, in 1994, were stable, capable systems that could run DOS and Windows programs, often better than Windows itself. IBM marketed it as “a better DOS than DOS and a better Windows than Windows”. The slogan captured the strategy’s weakness: if Windows programs ran fine on OS/2, why would anyone write an OS/2 program? When Windows 95 arrived in August 1995 with a huge launch, broad manufacturer support and a new generation of applications, the contest was over. IBM stopped marketing OS/2 in 2005 and ended support at the end of 2006.

Explaining both the same way

The usual story credits Microsoft’s strategic skill and blames IBM’s bureaucracy. A symmetrical account looks at what each side optimised for.

IBM optimised for its own business: selling PS/2 hardware and serving corporate customers who valued stability and IBM support. Those priorities shaped the 286 decision and the emphasis on corporate features. Microsoft optimised for being on every PC from every manufacturer, which meant running on hardware people already owned and keeping the platform under its own control. When the market rewarded running on existing machines with existing software, Microsoft’s priorities matched the market and IBM’s did not. Technical quality was not the deciding factor; fit with users’ installed base was.

The partnership itself is a lesson in transaction costs. Two companies with diverging interests, jointly building a product whose value each wanted to capture, faced constant negotiation. Once Windows gave Microsoft a better option, the incentive to keep negotiating disappeared.

What the loss teaches

OS/2 shows that in platform markets the deciding question is rarely “which system is better?” It is “which system runs what I already have, and what will developers write for next?” OS/2 answered the first question badly in 1987 and answered it too well in 1992, by running Windows programs so well that it strengthened Windows.

It also shows the fragility of partnerships between companies whose long-term interests diverge. The agreement made sense for both in 1985. It stopped making sense for one of them in 1990.

Sources and limitations

This page relies on published histories of IBM and Microsoft and on Gordon Letwin’s technical account of OS/2’s design. Reasons for the 286 decision and for the partnership’s collapse are described differently by participants from each company; this page describes the decisions and their consequences rather than adjudicating motives. Sales figures are not given because reliable comparable figures were not verified for this page.