(NB: I am moving some years-old writings into Due Diligence for future reference. This piece originally appeared in early 1997 on Howard Rheingold's old Electric Minds site. Little or no effort has been expended on cleaning up links broken in the intervening decade.)
In The Beginning
The dominant computing platform of the mid-seventies was IBM mainframe architecture. For those allergic to big iron, there were minicomputers from companies like DEC, Data General, HP, and, of course, IBM. Each had its own proprietary operating system. Programs often came from the platform vendors as well, or were custom-written in COBOL and FORTRAN, typically using features peculiar to the machine. The entire stack, from hardware to end user, was vertically integrated and highly specific.
The first hobbyist microcomputers appeared in 1975. They were regarded as toys by the mainstream computer industry. With 4 kilobytes of memory, a processor a little more powerful than a pocket calculator, and data storage on audiocassettes, they were clearly no threat to the mainframe or minicomputer companies. They were mostly programmed in raw machine code and there was no operating system at all. Each machine type was an incompatible oddball, an island, and they usually ran games.
Fire in the Valley
By 1977, a pattern had emerged. Data storage moved to 8" floppy disks, which had to be organized into files. File storage, starting up programs, and handling other input and output was managed by CP/M, an operating system from Digital Research. CP/M had a downward interface as well, called a BIOS (basic input/output system). Porting the BIOS onto a new machine would let CP/M and programs that used it run there.
Most machines were built around processors evolved from an Intel design: the 8008, Z80 and 8080. An open standard plug-in bus, the S-100, allowed the user to extend the platform by adding cards, to talk to a teletype or paper tape, or to generate graphics on a TV display. Memory was up to 16KB, and a lot of programming was now being done in BASIC, which had to be bought (or pirated) and loaded onto the machine. The most popular BASIC was from a small company called Microsoft.
This was the first de facto microcomputer platform: a combination of a proprietary disk system (CP/M), a commoditized hardware interface (S-100), and an add-on language (BASIC). It was a "hollowed-out" platform, and startup companies like IMSAI, NorthStar and Godbout competed in hardware to support the platform. Some of them were turning in million-dollar monthly sales.
A few general-purpose programs for word processing and simple accounting were popping up. The big computer companies weren't interested yet, but some consumer electronics outfits thought they saw an opportunity to build a whole new market for computers in homes and small offices.
Radio Shack, Texas Instruments, Sinclair, Commodore, and Atari entered the market. Each went down a similar path: each observed that the hollowed-out CP/M platform was too complex to set up and operate for nonhobbyists. Further, it was inefficient: S-100 connectors were large and expensive, and the BIOS and CP/M took up precious memory space. Each of these competitors built a cost-reduced, simplified, vertically integrated design, with a slim, proprietary operating system, closed packaging, and idiosyncratic programming environment. Some did quite well for a while -- most notably the Commodore 64. In the end, they all succumbed to the same problem: their machines could not be extended to accommodate growing functionality, all the programming was specific to that machine, and they failed to establish themselves as platforms.
One company got it right. At the beginning of its history, Apple Computer changed the conversation. Foregoing the Intel architecture, Apple chose a competing chip called the 6502. It passed up CP/M and wrote its own operating system. It licensed the most popular BASIC from Microsoft and soon burned it into ROM as a standard. Color graphics became part of the platform, not an add-on. The Apple II had a sleek, consumerized package, but it left room for cost-reduced plug-in slots for hardware and RAM expansion cards. 5 1/4" floppy drives were still add-ons, but they quickly became the standard. The result was a vertically integrated platform that greatly expanded the scope of standardization while leaving room for further growth.
Apple took off like a rocket. It passed $100 million annual sales in early 1980 and it went public later that year. The first spreadsheet, Visicalc, appeared on the Apple. This was something that even the mainframes and minis weren't doing. A substantial business software market began to evolve on the Apple platform, led by Visicorp, Peachtree, and Software Publishing. The mainframe big boys finally took notice.
The Empire Strikes Back
IBM was falling badly behind, and its business customers were noticing. Needing a time-to-market move, it sequestered a small, entrepreneurial team in Florida. Working almost entirely outside the corporate hierarchy, they created an Apple killer. For speed in design, it was built from off-the-shelf chips. The central processor, an 8088 sourced from Intel, was several times more powerful than the Apple's 6502. However, the design that became the IBM PC was architecturally a near replica of the Apple II: Microsoft BASIC in ROM, integrated graphics system, new operating system (also from Microsoft), and a set of slots for add-on hardware. It also integrated the floppy disc drives into the system case.
The 1981 launch of the IBM PC against Apple was a clean case of head-to-head competition, and IBM's brand strength and distribution power made it a winner. The PC racked up a billion dollars of revenue in its first year in the market. It destroyed the remaining CP/M market and drove the Apple II to a minority share, freezing its penetration into business. Business software providers scrambled to get on the PC platform, and those that lagged were overthrown by new entrants -- for example, Visicalc was virtually eliminated by Lotus 1-2-3.
But IBM's control over its own platform was dangerously weak. The operating system and programming languages were owned by Microsoft. IBM had published the code and interfaces for its graphic system and BIOS, and the specification for the PC's hardware slots. Its architecture of common chips was easily analyzed by potential competitors. This would return to haunt Big Blue.
But in the short run, the network effects of IBM's platform win were crushing competitors. DEC and AT&T, for instance, built "almost PCs" that attempted to maintain some differentiation of software and hardware. As they soon found out, if a machine couldn't run the stock versions of PC-DOS and 1-2-3 (and Flight Simulator), the market didn't want it.
Meanwhile, Apple had misfired with its expansion of Apple II architecture into the Apple III, and with its first attempt at a new generation, the
Lisa. Then Apple also isolated an entrepreneurial team and produced the Macintosh, its second attempt to change the conversation. Released in 1984, the Mac did establish a new genre, with operating system, disk storage, and graphics fully merged into a point-and-click user interface. Monitor, Motorola 68000 processor, and disk drives were all integrated into one case. The Mac expanded the range of possible applications, and brought users who could not deal with the complexity of PC-DOS into the market. Its architectural advantage in integration and ease of use was so great that it eventually took the PC market over seven years for to catch up, though Microsoft, IBM, and Digital Research began work on imitations immediately.
But the Mac did not take off. The first versions were underpowered, and the closed package was nearly impossible to extend. Application software was limited at first release, and the novel architecture complicated and slowed development. By 1987, many unique applications had arrived, and expandable Macs with more powerful processors were on the market, stabilizing Apple's market share as the Apple II dwindled. In some markets where it pioneered, such as graphics and desktop publishing, the Mac entrenched itself with a majority share.
In 1985, IBM held 80 percent of the business microcomputer market, with the remainder split among Apple and a few cloners. One of these, Compaq, had beaten IBM to market with the first "luggable" portable PC. Some other cracks had appeared in the armor: the PCjr was flopping in the home market, and the first versions of the PC AT had flawed 80286 chips and hard drives that were prone to crash.
Now IBM made two disastrous decisions. First, it negotiated a reduced royalty on Microsoft's PC-DOS by giving Microsoft all royalties from the sale of the operating system to PC cloners: IBM lost its last point of proprietary control. Bill Gates now had the incentive, and soon the cash flow, to move independently of IBM. Second, IBM decided it needed to define a new PC standard, over which it had firmer control. The next-generation PC would contain a proprietary operating system and hardware, which cloners would have to license from IBM. In effect, it planned to change the conversation on its own platform. IBM counted on the power of its marketing, distribution, and reputation with customers to force these changes through.
IBM lost its bet, badly. Sluggish execution let the cloners run away with the market. Compaq shipped an 80386-based PC in September 1986. It got its chips from Intel, the operating system from Microsoft, and built its design around the old PC AT architecture. IBM took another year to get MicroChannel, its proprietary hardware architecture, into the market in a '386-based PS/2. By then, no one cared; when IBM discontinued the PC AT, the customers switched to clones instead of PS/2s. IBM's market share was down to 40 percent and falling fast. Compaq, Intel, and the other cloners had seized control of the hardware portion of the platform; in late 1988 IBM threw in the towel and began shipping PS/2s without MicroChannel.
Worse: OS/2, IBM's new operating system, was even further behind. Coordinating specifications and development between a contract project at Microsoft and multiple "human wave" efforts at IBM stretched the schedule and bloated features and size. The first versions of OS/2 shipped in 1988; cloners, developers, and the market largely ignored it. They stuck with Microsoft's DOS, and some began to pay attention to Windows. By the end of the decade, IBM's market share was under 20 percent.
In 1988, Apple was riding high. The Macintosh hardware had been redesigned for more power and extensibility. For graphics-heavy applications, it was clearly differentiated from the PC platform. Apple was getting high margins in its sales to the markets that needed the Mac's capabilities; its financials looked great. Nonetheless, it was still trapped in a niche, which would become a dangerous position if and when the PC platform acquired a decent graphical interface.
Apple had a number of potential moves available:
- Accretion: Apple added incremental functionality through the various releases of the Macintosh, but aggressive accretion was limited by caution about "competing with developers" who had to be kept loyal to a minority platform.
- Hollowing out: Apple could have licensed out its hardware platform design to cloners while retaining control of the software. Apple found itself in a paradox. While its differentiation was in the software, its profitable business model seemed to rely on selling boxes. It took another six years before this option was pursued.
- Apple also considered an offensive hollowing-out strategy, taking its software onto the Intel hardware platform. In doing the analysis for this project, it became clear that over seven years of development had taken their toll on the Mac system software. It was a spiderweb of intercalling functions, many of them highly dependent on Apple's peculiar hardware. The Mac system would need an overhaul before much more functionality could be added, let alone making the move to Intel. [The later release of A/UX with a re-entrant Mac system showed it was possible.]
Apple had to pick: do the dirty work of cleaning up its existing system, or start over with a second system with new capabilities. It took the second choice. Like IBM, Apple had decided to change the conversationon its own platform, and it too lost badly.
Designers and management sorted the system functions for the future into two sets. The differentiating features went into the new system software project, called Pink, after the color of index cards for
its features. The Pink project was eventually spun off into the Taligent joint venture with IBM. It never shipped. After hundreds of millions of dollars in investment, Taligent was quietly disbanded, and its assets absorbed into an IBM development group.
The incremental features for the Mac OS were on a second set of index cards, blue ones. Blue became Mac System 7.0 and its subsequent revisions. When it was evident that Pink would not reach market, Apple moved to add some of its differentiating features back into the old architecture, in a project called Copland. It also failed. By the time Copland was cancelled in 1996, the Macintosh had lost nearly half its peak market share, the PC-based platform had almost erased its competitive advantage, and Apple was bleeding cash.
The Ascent of Windows
A project called Interface Manager began quietly at Microsoft in late 1981. An attempt to create a unified user interface for DOS, it got more support when Mac prototypes reached Microsoft. By late 1983, it had been renamed Windows and Gates announced it at Comdex, ahead of the Macintosh release. Microsoft promised delivery in mid-1984; Windows would include both a graphical interface and the ability to multitask, something the Mac couldn't do.
Windows 1.0 didn't actually ship until mid-1985, delayed in part by a struggle with bugs in the PC AT's 80286 chip. Its best successes were as an interface manager for developers of graphical applications trying to address the PC's installed base. Windows was poorly integrated with the DOS and PC hardware platform, and suffered from sluggish performance and a more confusing interface than the Mac. It still had a long way to go.
Windows was good enough, however, to exacerbate conflict in Microsoft's relationship with IBM, which backed first Topview and then OS/2, and was getting the larger share of Microsoft's resources. A small Windows team kept plugging away, and version 2.0 came out in October, 1987. Although Microsoft claimed up to a million units had shipped by the year's end, the number paid for and actually used was a lot smaller. Although Apple was becoming concerned enough to file a suit (which it eventually lost) for copyright infringement against Microsoft, support for Windows inside Microsoft was weakened; at one point the team consisted of a single programmer.
Microsoft's relationship with IBM was getting worse. Disagreements around OS/2 goals were rampant, and IBM was experimenting with UNIX and NeXT. When the Windows team figured out how to beat a memory size limitation, Microsoft staffed up the project again. An industry conference in October, 1989, supposed to publicly reconcile the two companies' differences, instead made it obvious that Microsoft was now fully backing Windows. When Version 3.0 shipped in mid-1990, the improvement in PC processor speeds had made its performance acceptable, and Macintosh applications began to migrate to Windows in large numbers.
When Windows 3.1 came out in April, 1992, Microsoft backed it with television advertising and sold a million units in 50 days. Two months later, the IBM-Microsoft relationship dissolved. OS/2 had been
relegated to a small minority market share. Apple had been trapped in its niche. The market share of PC-platform machines now outweighed remaining problems in Windows usability, and Microsoft was working on an overhaul to erase the remaining Mac advantages. "Wintel" was now the dominant platform.