FOR THE DISTRICT OF COLUMBIA

____________________________________

)

UNITED STATES OF AMERICA, )

)

Plaintiff, )

)

v. ) Civil Action No. 98-1232 (TPJ)

)

MICROSOFT CORPORATION, )

)

Defendant. )

)

____________________________________)

)

STATE OF NEW YORK, ex rel. )

Attorney General ELIOT SPITZER, )

et al., )

)

Plaintiffs and )

Counterclaim-Defendants, )

)

v. ) Civil Action No. 98-1233 (TPJ)

)

MICROSOFT CORPORATION, )

)

Defendant and )

Counterclaim-Plaintiff. )

_______________________)

These consolidated civil antitrust actions alleging violations of the Sherman Act, §§ 1 and 2, and various state statutes by the defendant Microsoft Corporation, were tried to the Court, sitting without a jury, between October 19, 1998, and June 24, 1999. The Court has considered the record evidence submitted by the parties, made determinations as to its relevancy and materiality, assessed the credibility of the testimony of the witnesses, both written and oral, and ascertained for its purposes the probative significance of the documentary and visual evidence presented. Upon the record before the Court as of July 28, 1999, at the close of the admission of evidence, pursuant to Fed. R. Civ. P. 52(a), the Court finds the following facts to have been proved by a preponderance of the evidence. The Court shall state the conclusions of law to be drawn therefrom in a separate Memorandum and Order to be filed in due course.

1. A "personal computer" ("PC") is a digital information processing device designed for use by one person at a time. A typical PC consists of central processing components (e.g., a microprocessor and main memory) and mass data storage (such as a hard disk). A typical PC system consists of a PC, certain peripheral input/output devices (including a monitor, a keyboard, a mouse, and a printer), and an operating system. PC systems, which include desktop and laptop models, can be distinguished from more powerful, more expensive computer systems known as "servers," which are designed to provide data, services, and functionality through a digital network to multiple users.

2. An "operating system" is a software program that controls the allocation and use of computer resources (such as central processing unit time, main memory space, disk space, and input/output channels). The operating system also supports the functions of software programs, called "applications," that perform specific user-oriented tasks. The operating system supports the functions of applications by exposing interfaces, called "application programming interfaces," or "APIs." These are synapses at which the developer of an application can connect to invoke pre-fabricated blocks of code in the operating system. These blocks of code in turn

perform crucial tasks, such as displaying text on the computer screen. Because it supports applications while interacting more closely with the PC system’s hardware, the operating system is said to serve as a "platform."

3. An Intel-compatible PC is one designed to function with Intel’s 80x86/Pentium families of microprocessors or with compatible microprocessors manufactured by Intel or by other firms.

4. An operating system designed to run on an Intel-compatible PC will not function on a non-Intel-compatible PC, nor will an operating system designed for a non-Intel-compatible PC function on an Intel-compatible one. Similarly, an application that relies on APIs specific to one operating system will not, generally speaking, function on another operating system unless it is first adapted, or "ported," to the APIs of the other operating system.

5. Defendant Microsoft Corporation is organized under the laws of the State of Washington, and its headquarters are situated in Redmond, Washington. Since its inception, Microsoft has focused primarily on developing software and licensing it to various purchasers.

6. In 1981, Microsoft released the first version of its Microsoft Disk Operating System, commonly known as "MS-DOS." The system had a character-based user interface that required the user to type specific instructions at a command prompt in order to perform tasks such as launching applications and copying files. When the International Business Machines Corporation ("IBM") selected MS-DOS for pre-installation on its first generation of PCs, Microsoft’s product became the predominant operating system sold for Intel-compatible PCs.

7. In 1985, Microsoft began shipping a software package called Windows. The product included a graphical user interface, which enabled users to perform tasks by selecting icons and words on the screen using a mouse. Although originally just a user-interface, or "shell," sitting on top of MS-DOS, Windows took on more operating-system functionality over time.

8. In 1995, Microsoft introduced a software package called Windows 95, which announced itself as the first operating system for Intel-compatible PCs that exhibited the same sort of integrated features as the Mac OS running PCs manufactured by Apple Computer, Inc. ("Apple"). Windows 95 enjoyed unprecedented popularity with consumers, and in June 1998, Microsoft released its successor, Windows 98.

9. Microsoft is the leading supplier of operating systems for PCs. The company transacts business in all fifty of the United States and in most countries around the world.

10. Microsoft licenses copies of its software programs directly to consumers. The largest part of its MS-DOS and Windows sales, however, consists of licensing the products to manufacturers of PCs (known as "original equipment manufacturers" or "OEMs"), such as the IBM PC Company and the Compaq Computer Corporation ("Compaq"). An OEM typically installs a copy of Windows onto one of its PCs before selling the package to a consumer under a single price.

11. The Internet is a global electronic network, consisting of smaller, interconnected networks, which allows millions of computers to exchange information over telephone wires, dedicated data cables, and wireless links. The Internet links PCs by means of servers, which run specialized operating systems and applications designed for servicing a network environment.

12. The World Wide Web ("the Web") is a massive collection of digital information resources stored on servers throughout the Internet. These resources are typically provided in the form of hypertext documents, commonly referred to as "Web pages," that may incorporate any combination of text, graphics, audio and video content, software programs, and other data. A user of a computer connected to the Internet can publish a page on the Web simply by copying it into a specially designated, publicly accessible directory on a Web server. Some Web resources are in the form of applications that provide functionality through a user’s PC system but actually execute on a server.

13. Internet content providers ("ICPs") are the individuals and organizations that have established a presence, or "site," on the Web by publishing a collection of Web pages. Most Web pages are in the form of "hypertext"; that is, they contain annotated references, or "hyperlinks," to other Web pages. Hyperlinks can be used as cross-references within a single document, between documents on the same site, or between documents on different sites.

14. Typically, one page on each Web site is the "home page," or the first access point to the site. The home page is usually a hypertext document that presents an overview of the site and hyperlinks to the other pages comprising the site.

15. PCs typically connect to the Internet through the services of Internet access providers ("IAPs"), which generally charge subscription fees to their customers in the United States. There are two types of IAPs. Online services ("OLSs") such as America Online ("AOL"), Prodigy, and the Microsoft Network ("MSN") offer, in addition to Internet access, various services and an array of proprietary content. Internet service providers ("ISPs") such as MindSpring and Netcom, on the other hand, offer few services apart from Internet access and relatively little of their own content.

16. A "Web client" is software that, when running on a computer connected to the Internet, sends information to and receives information from Web servers throughout the Internet. Web clients and servers transfer data using a standard known as the Hypertext Transfer Protocol ("HTTP"). A "Web browser" is a type of Web client that enables a user to select, retrieve, and perceive resources on the Web. In particular, Web browsers provide a way for a user to view hypertext documents and follow the hyperlinks that connect them, typically by moving the cursor over a link and depressing the mouse button.

17. Although certain Web browsers provided graphical user interfaces as far back as 1993, the first widely-popular graphical browser distributed for profit, called Navigator, was brought to market by the Netscape Communications Corporation in December 1994. Microsoft introduced its browser, called Internet Explorer, in July 1995.

18. Currently there are no products, nor are there likely to be any in the near future, that a significant percentage of consumers world-wide could substitute for Intel-compatible PC operating systems without incurring substantial costs. Furthermore, no firm that does not currently market Intel-compatible PC operating systems could start doing so in a way that would, within a reasonably short period of time, present a significant percentage of consumers with a viable alternative to existing Intel-compatible PC operating systems. It follows that, if one firm controlled the licensing of all Intel-compatible PC operating systems world-wide, it could set the price of a license substantially above that which would be charged in a competitive market and leave the price there for a significant period of time without losing so many customers as to make the action unprofitable. Therefore, in determining the level of Microsoft’s market power, the relevant market is the licensing of all Intel-compatible PC operating systems world-wide.

19. Consumers could not turn from Intel-compatible PC operating systems to Intel-compatible server operating systems without incurring substantial costs, since the latter type of system is sold at a significantly higher price than the former. A consumer intent on acquiring a server operating system would also have to buy a computer of substantially greater power and price than an Intel-compatible PC, because server operating systems generally cannot function properly on PC hardware. The price of an Intel-compatible PC operating system accounts for only a very small percentage of the price of an Intel-compatible PC system. Thus, even a substantial increase in the price of an Intel-compatible PC operating system above the competitive level would result in only a trivial increase in the price of an Intel-compatible PC system. Very few consumers would purchase expensive servers in response to a trivial increase in the price of an Intel-compatible PC system. Furthermore, a consumer would not obtain a satisfactory substitute for an Intel-compatible PC operating system even if he purchased a server, since server operating systems lack the features — and support for the breadth of applications — that induce users to purchase Intel-compatible PC operating systems.

20. Since only Intel-compatible PC operating systems will work with Intel-compatible PCs, a consumer cannot opt for a non-Intel-compatible PC operating system without obtaining a non-Intel-compatible PC. Thus, for consumers who already own an Intel-compatible PC system, the cost of switching to a non-Intel compatible PC operating system includes the price of not only a new operating system, but also a new PC and new peripheral devices. It also includes the effort of learning to use the new system, the cost of acquiring a new set of compatible applications, and the work of replacing files and documents that were associated with the old applications. Very few consumers would incur these costs in response to the trivial increase in the price of an Intel-compatible PC system that would result from even a substantial increase in the price of an Intel-compatible PC operating system. For example, users of Intel-compatible PC operating systems would not switch in large numbers to the Mac OS in response to even a substantial, sustained increase in the price of an Intel-compatible PC operating system.

21. The response to a price increase would be somewhat greater among consumers buying their first PC system, because they would not have already invested time and money in an Intel-compatible PC system and a set of compatible applications. Apple does not license the Mac OS separately from its PC hardware, however, and the package of hardware and software comprising an Apple PC system is priced substantially higher than the average price of an Intel-compatible PC system. Furthermore, consumer demand for Apple PC systems suffers on account of the relative dearth of applications written to run on the Mac OS. It is unlikely, then, that a firm controlling the licensing of all Intel-compatible PC operating systems would lose so many new PC users to Apple as the result of a substantial, enduring price increase as to make the action unprofitable. It is therefore proper to define a relevant market that excludes the Mac OS. In any event, as Section III of these findings demonstrates, including the Mac OS in the relevant market would not alter the Court’s conclusion as to the level of Microsoft’s market power.

22. No operating system designed for a hand-held computer, a "smart" wireless telephone, a television set-top box, or a game console is capable of performing as an adequate operating system for an Intel-compatible PC. Therefore, in order to adopt a substitute for the Intel-compatible PC operating system from the realm of "information appliances," a consumer must acquire one or more of these devices in lieu of an Intel-compatible PC system.

23. It is possible that, within the next few years, those consumers who otherwise would use an Intel-compatible PC system solely for storing addresses and schedules, for sending and receiving E-mail, for browsing the Web, and for playing video games might be able to choose a complementary set of information appliances over an Intel-compatible PC system without incurring substantial costs. To the extent this substitution occurs, though, it will be the result of innovation by the producers of information appliances, and it will occur even if Intel-compatible PC operating systems are priced at the same level that they would be in a competitive market. More importantly, while some consumers may decide to make do with one or more information appliances in place of an Intel-compatible PC system, the number of these consumers will, for the foreseeable future, remain small in comparison to the number of consumers deciding that they still need an Intel-compatible PC system. One reason for this is the fact that no single type of information appliance, nor even all types in the aggregate, provides all of the features that most consumers have come to rely on in their PC systems and in the applications that run on them. Thus, most of those who buy information appliances will do so in addition to, rather than instead of, buying an Intel-compatible PC system. Not surprisingly, then, sales of PC systems are not expected to suffer on account of the growing consumer interest in information appliances. It follows that, for the foreseeable future, a firm controlling the licensing of all Intel-compatible PC operating systems could set prices substantially above competitive levels without losing an unacceptable amount of business to information appliances.

24. A network computer system (sometimes called a "thin client") typically contains central processing components with basic capabilities, certain key peripheral devices (such as a monitor, a keyboard, and a mouse), an operating system, and a browser. The system contains no mass storage, however, and it processes little if any data locally. Instead, the system receives processed data and software as needed from a server across a network. A network computer system lacks the hardware resources to support an Intel-compatible PC operating system. It follows that software applications written to run on a specific Intel-compatible PC operating system will not run on a network computer. Network computers can run applications residing on a designated server, however. Moreover, a network computer system typically can run applications residing on other servers, so long as those applications are accessible through Web sites. The ability to run server-based applications is not exclusive to network computer systems, however. Generally speaking, any PC system equipped with a browser and an Internet connection is capable of accessing applications hosted through Web sites.

25. Since the network computing model relies heavily on the processing power and memory of servers, the requirements for the user’s hardware (and thus the price of that hardware) are low relative to those of an Intel-compatible PC system. Still, a user who already owns a relatively expensive Intel-compatible PC system is not likely to abandon the investment and acquire less powerful hardware just because one of the least expensive components of his PC system — the operating system — is substantially more expensive than it would be under competitive conditions. Just as does the Mac OS, the network computing model presents a somewhat more attractive alternative to the first-time computer buyer. But as in the case where a prospective purchaser is considering acquiring the Apple alternative, a new buyer considering the network computing model must choose between types of computer systems. If the consumer opts for the less expensive hardware of the network computer, that hardware will not support an Intel-compatible PC operating system; and if the new buyer opts for the more expensive hardware of an Intel-compatible PC, an Intel-compatible PC operating system will almost certainly come pre-installed (and in any event represent very little additional cost relative to the price of the hardware).

26. Only a few firms currently market network computer systems, and the systems have yet to attract substantial consumer demand. In part, this is because PC systems, which can store and process data locally as well as communicate with a server, have decreased so much in price as to call into question the value proposition of buying a network computer system. This fact would not change if the price of an Intel-compatible PC operating system rose significantly, because the resulting change in the price of an Intel-compatible PC system would be very minor. Another reason for the limited demand for network computer systems is the fact that few consumers are in a position to turn from PC systems to network computer systems without making substantial sacrifices; for the network computing option exhibits significant shortcomings for current PC owners and first-time buyers alike. The problems of latency, congestion, asynchrony, and insecurity across a communications network, and contention for limited processing and memory resources at the remote server, can all result in a substantial derogation of computing performance. Moreover, the owner of a network computer is required to enter into long-term dependency upon the owner of a remote server in order to obtain functionality that would reside within his control if he owned a PC system. If network computing becomes a viable alternative to PC-based computing, it will be because innovation by the proponents of the network computing model overcomes these problems, and it will happen even if Intel-compatible PC operating systems are priced at competitive levels. In any case, that day has not arrived, nor does it appear imminent.

27. As the bandwidth available to the average user increases, "portal" Web sites, which aggregate Web content and provide services such as search engines, E-mail, and travel reservation systems, could begin to host full lines of the server-based, personal-productivity applications that have begun to appear in small numbers on the Web. If so, increasing numbers of computer users equipped with Web browsers and IAP connections could begin to conduct a significant portion of their computing through these portals. To the extent they might do so, users probably would not regard the Mac OS’s limited stock of compatible applications as the major drawback to using an Apple PC system that it is today, and they might be increasingly drawn to network computer systems and information appliances. The variety and ease of use of server-based applications accessible through browsers would have to increase a great deal from today’s levels, however, before the total costs of dispensing with an Intel-compatible PC operating system would decline sufficiently to impose a significant constraint on the pricing of those systems. Again, that day is not imminent; for at least the next few years, the overwhelming majority of consumers accessing server-based applications will do so using an Intel-compatible PC system and a browser.

28. Operating systems are not the only software programs that expose APIs to application developers. The Netscape Web browser and Sun Microsystems, Inc.’s Java class libraries are examples of non-operating system software that do likewise. Such software is often called "middleware" because it relies on the interfaces provided by the underlying operating system while simultaneously exposing its own APIs to developers. Currently no middleware product exposes enough APIs to allow independent software vendors ("ISVs") profitably to write full-featured personal productivity applications that rely solely on those APIs.

29. Even if middleware deployed enough APIs to support full-featured applications, it would not function on a computer without an operating system to perform tasks such as managing hardware resources and controlling peripheral devices. But to the extent the array of applications relying solely on middleware comes to satisfy all of a user’s needs, the user will not care whether there exists a large number of other applications that are directly compatible with the underlying operating system. Thus, the growth of middleware-based applications could lower the costs to users of choosing a non-Intel-compatible PC operating system like the Mac OS. It remains to be seen, though, whether there will ever be a sustained stream of full-featured applications written solely to middleware APIs. In any event, it would take several years for middlware and the applications it supports to evolve from the status quo to a point at which the cost to the average consumer of choosing a non-Intel compatible PC operating system over an Intel-compatible one falls so low as to constrain the pricing of the latter systems.

30. Firms that do not currently produce Intel-compatible PC operating systems could do so. What is more, once a firm had written the necessary software code, it could produce millions of copies of its operating system at relatively low cost. The ability to meet a large demand is useless, however, if the demand for the product is small, and signs do not indicate large demand for a new Intel-compatible PC operating system. To the contrary, they indicate that the demand for a new Intel-compatible PC operating system would be severely constrained by an intractable "chicken-and-egg" problem: The overwhelming majority of consumers will only use a PC operating system for which there already exists a large and varied set of high-quality, full-featured applications, and for which it seems relatively certain that new types of applications and new versions of existing applications will continue to be marketed at pace with those written for other operating systems. Unfortunately for firms whose products do not fit that bill, the porting of applications from one operating system to another is a costly process. Consequently, software developers generally write applications first, and often exclusively, for the operating system that is already used by a dominant share of all PC users. Users do not want to invest in an operating system until it is clear that the system will support generations of applications that will meet their needs, and developers do not want to invest in writing or quickly porting applications for an operating system until it is clear that there will be a sizeable and stable market for it. What is more, consumers who already use one Intel-compatible PC operating system are even less likely than first-time buyers to choose a newcomer to the field, for switching to a new system would require these users to scrap the investment they have made in applications, training, and certain hardware.

31. The chicken-and-egg problem notwithstanding, a firm might reasonably expect to make a profit by introducing an Intel-compatible PC operating system designed to support a type of application that satisfies the special interests of a particular subset of users. For example, Be, Inc. (‘Be") markets an Intel-compatible PC operating system called BeOS that offers superior support for multimedia applications, and the operating system enjoys a certain amount of success with the segment of the consumer population that has a special interest in creating and playing multimedia content with a PC system. Still, while a niche operating system might turn a profit, the chicken-and-egg problem (hereinafter referred to as the "applications barrier to entry") would make it prohibitively expensive for a new Intel-compatible operating system to attract enough developers and consumers to become a viable alternative to a dominant incumbent in less than a few years.

32. To the extent that developers begin writing attractive applications that rely solely on servers or middleware instead of PC operating systems, the applications barrier to entry could erode. As the Court finds above, however, it remains to be seen whether server- or middleware-based development will flourish at all. Even if such development were already flourishing, it would be several years before the applications barrier eroded enough to clear the way for the relatively rapid emergence of a viable alternative to incumbent Intel-compatible PC operating systems. It is highly unlikely, then, that a firm not already marketing an Intel-compatible PC operating system could begin marketing one that would, in less than a few years, present a significant percentage of consumers with a viable alternative to incumbents.

33. Microsoft enjoys so much power in the market for Intel-compatible PC operating systems that if it wished to exercise this power solely in terms of price, it could charge a price for Windows substantially above that which could be charged in a competitive market. Moreover, it could do so for a significant period of time without losing an unacceptable amount of business to competitors. In other words, Microsoft enjoys monopoly power in the relevant market.

34. Viewed together, three main facts indicate that Microsoft enjoys monopoly power. First, Microsoft’s share of the market for Intel-compatible PC operating systems is extremely large and stable. Second, Microsoft’s dominant market share is protected by a high barrier to entry. Third, and largely as a result of that barrier, Microsoft’s customers lack a commercially viable alternative to Windows.

35. Microsoft possesses a dominant, persistent, and increasing share of the world-wide market for Intel-compatible PC operating systems. Every year for the last decade, Microsoft’s share of the market for Intel-compatible PC operating systems has stood above ninety percent. For the last couple of years the figure has been at least ninety-five percent, and analysts project that the share will climb even higher over the next few years. Even if Apple’s Mac OS were included in the relevant market, Microsoft’s share would still stand well above eighty percent.

1. Description of the Applications Barrier to Entry

36. Microsoft’s dominant market share is protected by the same barrier that helps define the market for Intel-compatible PC operating systems. As explained above, the applications barrier would prevent an aspiring entrant into the relevant market from drawing a significant number of customers away from a dominant incumbent even if the incumbent priced its products substantially above competitive levels for a significant period of time. Because Microsoft’s market share is so dominant, the barrier has a similar effect within the market: It prevents Intel-compatible PC operating systems other than Windows from attracting significant consumer demand, and it would continue to do so even if Microsoft held its prices substantially above the competitive level.

37. Consumer interest in a PC operating system derives primarily from the ability of that system to run applications. The consumer wants an operating system that runs not only types of applications that he knows he will want to use, but also those types in which he might develop an interest later. Also, the consumer knows that if he chooses an operating system with enough demand to support multiple applications in each product category, he will be less likely to find himself straitened later by having to use an application whose features disappoint him. Finally, the average user knows that, generally speaking, applications improve through successive versions. He thus wants an operating system for which successive generations of his favorite applications will be released — promptly at that. The fact that a vastly larger number of applications are written for Windows than for other PC operating systems attracts consumers to Windows, because it reassures them that their interests will be met as long as they use Microsoft’s product.

38. Software development is characterized by substantial economies of scale. The fixed costs of producing software, including applications, is very high. By contrast, marginal costs are very low. Moreover, the costs of developing software are "sunk" — once expended to develop software, resources so devoted cannot be used for another purpose. The result of economies of scale and sunk costs is that application developers seek to sell as many copies of their applications as possible. An application that is written for one PC operating system will operate on another PC operating system only if it is ported to that system, and porting applications is both time-consuming and expensive. Therefore, application developers tend to write first to the operating system with the most users — Windows. Developers might then port their applications to other operating systems, but only to the extent that the marginal added sales justify the cost of porting. In order to recover that cost, ISVs that do go to the effort of porting frequently set the price of ported applications considerably higher than that of the original versions written for Windows.

39. Consumer demand for Windows enjoys positive network effects. A positive network effect is a phenomenon by which the attractiveness of a product increases with the number of people using it. The fact that there is a multitude of people using Windows makes the product more attractive to consumers. The large installed base attracts corporate customers who want to use an operating system that new employees are already likely to know how to use, and it attracts academic consumers who want to use software that will allow them to share files easily with colleagues at other institutions. The main reason that demand for Windows experiences positive network effects, however, is that the size of Windows’ installed base impels ISVs to write applications first and foremost to Windows, thereby ensuring a large body of applications from which consumers can choose. The large body of applications thus reinforces demand for Windows, augmenting Microsoft’s dominant position and thereby perpetuating ISV incentives to write applications principally for Windows. This self-reinforcing cycle is often referred to as a "positive feedback loop."

40. What for Microsoft is a positive feedback loop is for would-be competitors a vicious cycle. For just as Microsoft’s large market share creates incentives for ISVs to develop applications first and foremost for Windows, the small or non-existent market share of an aspiring competitor makes it prohibitively expensive for the aspirant to develop its PC operating system into an acceptable substitute for Windows. To provide a viable substitute for Windows, another PC operating system would need a large and varied enough base of compatible applications to reassure consumers that their interests in variety, choice, and currency would be met to more-or-less the same extent as if they chose Windows. Even if the contender attracted several thousand compatible applications, it would still look like a gamble from the consumer’s perspective next to Windows, which supports over 70,000 applications. The amount it would cost an operating system vendor to create that many applications is prohibitively large. Therefore, in order to ensure the availability of a set of applications comparable to that available for Windows, a potential rival would need to induce a very large number of ISVs to write to its operating system.

41. In deciding whether to develop an application for a new operating system, an ISV’s first consideration is the number of users it expects the operating system to attract. Out of this focus arises a collective-action problem: Each ISV realizes that the new operating system could attract a significant number of users if enough ISVs developed applications for it; but few ISVs want to sink resources into developing for the system until it becomes established. Since everyone is waiting for everyone else to bear the risk of early adoption, the new operating system has difficulty attracting enough applications to generate a positive feedback loop. The vendor of a new operating system cannot effectively solve this problem by paying the necessary number of ISVs to write for its operating system, because the cost of doing so would dwarf the expected return.

42. Counteracting the collective-action phenomenon is another known as the "first-mover incentive." For an ISV interested in attracting users, there may be an advantage to offering the first and, for a while, only application in its category that runs on a new PC operating system. The user base of the new system may be small, but every user of that system who wants such an application will be compelled to use the ISV’s offering. Moreover, if demand for the new operating system suddenly explodes, the first mover will reap large sales before any competitors arrive. An ISV thus might be drawn to a new PC operating system as a "protected harbor." Once first-movers stake claims to the major categories of applications, however, there is a strong chance that the new operating system could stall; it would not support the most familiar applications, nor the variety and number of applications, that attract large numbers of consumers, and there would no longer exist a first-mover incentive to attract additional ISVs to the important application categories. Although the upstart operating system might find itself with enough applications support to hold a fraction of the market, the collective-action phenomenon would still prevent the system from gaining the kind of positive feedback momentum that can turn a fringe entrant into a rival that would put competitive pressure on Windows.

43. The cost to a would-be entrant of inducing ISVs to write applications for its operating system exceeds the cost that Microsoft itself has faced in inducing ISVs to write applications for its operating system products, for Microsoft never confronted a highly penetrated market dominated by a single competitor. Of course, the fact that it is extremely difficult for an efficient would-be rival to accumulate enough applications support to compete with Windows does not mean that sustaining its own applications support is effortless for Microsoft. In fact, if Microsoft stopped investing the hundreds of millions of dollars it spends each year inducing ISVs to write applications for Windows, it might become easier than it currently is for a competitor to develop its own positive feedback loop. But given that Windows today enjoys overwhelmingly more applications support than any other PC operating system, it would still take that competitor years to develop the necessary momentum. Plus, while Microsoft may spend more on platform "evangelization," even in relative terms, than any other PC operating-system vendor, it is not difficult to understand why it is worthwhile for the principal beneficiary of the applications barrier to devote more resources to augmenting it than aspiring rivals are willing to expend in speculative efforts to erode it.

44. Microsoft continually releases "new and improved" versions of its PC operating system. Each time it does, Microsoft must convince ISVs to write applications that take advantage of new APIs, so that existing Windows users will have incentive to buy an upgrade. Since ISVs are usually still earning substantial revenue from applications written for the last version of Windows, Microsoft must convince them to write for the new version. Even if ISVs are slow to take advantage of the new APIs, though, no applications barrier stands in the way of consumers adopting the new system, for Microsoft ensures that successive versions of Windows retain the ability to run applications developed for earlier versions. In fact, since ISVs know that consumers do not feel locked into their old versions of Windows and that new versions have historically attracted substantial consumer demand, ISVs will generally write to new APIs as long as the interfaces enable attractive, innovative features. Microsoft supplements developers’ incentives by extending various ‘seals of approval’ — visible to consumers, investors, and industry analysts — to those ISVs that promptly develop new versions of their applications adapted to the newest version of Windows. In addition, Microsoft works closely with ISVs to help them adapt their applications to the newest version of the operating system — a process that is in any event far easier than porting an application from one vendor’s PC operating system to another’s. In sum, despite the substantial resources Microsoft expends inducing ISVs to develop applications for new versions of Windows, the company does not face any obstacles nearly as imposing as the barrier to entry that vendors and would-be vendors of other PC operating systems must overcome.

2. Empirical Evidence of the Applications Barrier to Entry

45. The experiences of IBM and Apple, Microsoft’s most significant operating system rivals in the mid- and late 1990s, confirm the strength of the applications barrier to entry.

46. IBM’s inability to gain widespread developer support for its OS/2 Warp operating system illustrates how the massive Windows installed base makes it prohibitively costly for a rival operating system to attract enough developer support to challenge Windows. In late 1994, IBM introduced its Intel-compatible OS/2 Warp operating system and spent tens of millions of dollars in an effort to attract ISVs to develop applications for OS/2 and in an attempt to reverse-engineer, or "clone," part of the Windows API set. Despite these efforts, IBM could obtain neither significant market share nor ISV support for OS/2 Warp. Thus, although at its peak OS/2 ran approximately 2,500 applications and had 10% of the market for Intel-compatible PC operating systems, IBM ultimately determined that the applications barrier prevented effective competition against Windows 95. For that reason, in 1996 IBM stopped trying to convince ISVs to write for OS/2 Warp. IBM now targets the product at a market niche, namely enterprise customers (mainly banks) that are interested in particular types of application that run on OS/2 Warp. The fact that IBM no longer tries to compete with Windows is evidenced by the fact that it prices OS/2 Warp at about two-and-one-half times the price of Windows 98.

47. The inability of Apple to compete effectively with Windows provides another example of the applications barrier to entry in operation. Although Apple’s Mac OS supports more than 12,000 applications, even an inventory of that magnitude is not sufficient to enable Apple to present a significant percentage of users with a viable substitute for Windows. The absence of a large installed base, in turn, reinforces the disparity between the applications made available for the Mac OS and those made available for Windows, further inhibiting Apple’s sales. The applications barrier thus prevents the Mac OS from hindering Microsoft’s ability to control price, regardless of whether the Mac OS is regarded as being in the relevant market or not.

48. The applications barrier to entry does not prevent non-Microsoft, Intel-compatible PC operating systems from attracting enough consumer demand and ISV support to survive. It does not even prevent vendors of those products from making a profit. The barrier does, however, prevent the products from drawing a significant percentage of consumers away from Windows.

49. As discussed above, Be markets an Intel-compatible PC operating system, called BeOS, that is specially suited to support multimedia functions. The operating system survives on a relatively minuscule number of applications (approximately 1,000) and a user base which, at around 750,000, is trivial compared to the number of Windows users. One of the reasons the BeOS can even attract that many users despite its small base of applications is that it advertises itself as a complement to, rather than as a substitute for, Windows. Although the BeOS could run an Intel-compatible PC system without Windows, it is almost always loaded on a system along with Windows. What is more, when these dual-loaded PC systems are turned on, Windows automatically boots; the user must then take affirmative steps to invoke the BeOS. While this scheme allows the BeOS to occupy a niche in the market, it does not place the product on a trajectory to replace Windows on a significant number of PCs. The special multimedia support provided by the BeOS may, for a small number of users, outweigh the disadvantages of maintaining two large, complex operating systems on one PC. Of that group, however, it is likely that only a tiny number of users will find that support so attractive that they would be willing to forego Windows, and its huge base of compatible applications, altogether.

50. The experience of the Linux operating system, a version of which runs on Intel-compatible PCs, similarly fails to refute the existence of an applications barrier to entry. Linux is an "open source" operating system that was created, and is continuously updated, by a global network of software developers who contribute their labor for free. Although Linux has between ten and fifteen million users, the majority of them use the operating system to run servers, not PCs. Several ISVs have announced their development of (or plans to develop) Linux versions of their applications. To date, though, legions of ISVs have not followed the lead of these first movers. Similarly, consumers have by and large shown little inclination to abandon Windows, with its reliable developer support, in favor of an operating system whose future in the PC realm is unclear. By itself, Linux’s open-source development model shows no signs of liberating that operating system from the cycle of consumer preferences and developer incentives that, when fueled by Windows’ enormous reservoir of applications, prevents non-Microsoft operating systems from competing.

51. Since application developers working under an open-source model are not looking to recoup their investment and make a profit by selling copies of their finished products, they are free from the imperative that compels proprietary developers to concentrate their efforts on Windows. In theory, then, open-source developers are at least as likely to develop applications for a non-Microsoft operating system as they are to write Windows-compatible applications. In fact, they may be disposed ideologically to focus their efforts on open-source platforms like Linux. Fortunately for Microsoft, however, there are only so many developers in the world willing to devote their talents to writing, testing, and debugging software pro bono publico. A small corps may be willing to concentrate its efforts on popular applications, such as browsers and office productivity applications, that are of value to most users. It is unlikely, though, that a sufficient number of open-source developers will commit to developing and continually updating the large variety of applications that an operating system would need to attract in order to present a significant number of users with a viable alternative to Windows. In practice, then, the open-source model of applications development may increase the base of applications that run on non-Microsoft PC operating systems, but it cannot dissolve the barrier that prevents such operating systems from challenging Windows.

52. Theoretically, the developer of a non-Microsoft, Intel-compatible PC operating system could circumvent the applications barrier to entry by cloning the APIs exposed by the 32-bit versions of Windows (Windows 9x and Windows NT). Applications written for Windows would then also run on the rival system, and consumers could use the rival system confident in that knowledge. Translating this theory into practice is virtually impossible, however. First of all, cloning the thousands of APIs already exposed by Windows would be an enormously expensive undertaking. More daunting is the fact that Microsoft continually adds APIs to Windows through updates and new versions. By the time a rival finished cloning the APIs currently in existence, Windows would have exposed a multitude of new ones. Since the rival would never catch up, it would never be able to assure consumers that its operating system would run all of the applications written for Windows. IBM discovered this to its dismay in the mid-1990s when it failed, despite a massive investment, to clone a sufficiently large part of the 32-bit Windows APIs. In short, attempting to clone the 32-bit Windows APIs is such an expensive, uncertain undertaking that it fails to present a practical option for a would-be competitor to Windows.

53. That Microsoft’s market share and the applications barrier to entry together endow the company with monopoly power in the market for Intel-compatible PC operating systems is directly evidenced by the sustained absence of realistic commercial alternatives to Microsoft’s PC operating-system products.

54. OEMs are the most important direct customers for operating systems for Intel-compatible PCs. Because competition among OEMs is intense, they pay particularly close attention to consumer demand. OEMs are thus not only important customers in their own right, they are also surrogates for consumers in identifying reasonably-available commercial alternatives to Windows. Without significant exception, all OEMs pre-install Windows on the vast majority of PCs that they sell, and they uniformly are of a mind that there exists no commercially viable alternative to which they could switch in response to a substantial and sustained price increase or its equivalent by Microsoft. For example, in 1995, at a time when IBM still placed hope in OS/2's ability to rival Windows, the firm nevertheless calculated that its PC company would lose between seventy and ninety percent of its sales volume if failed to load Windows 95 on its PCs. Although a few OEMs have announced their intention to pre-install Linux on some of the computers they ship, none of them plan to install Linux in lieu of Windows on any appreciable number of PC (as opposed to server) systems. For its part, Be is not even attempting to persuade OEMs to install the BeOS on PCs to the exclusion of Windows.

55. OEMs believe that the likelihood of a viable alternative to Windows emerging any time in the next few years is too low to constrain Microsoft from raising prices or imposing other burdens on customers and users. The accuracy of this belief is highlighted by the fact that the other vendors of Intel-compatible PC operating systems do not view their own offerings as viable alternatives to Windows. Microsoft knows that OEMs have no choice but to load Windows, both because it has a good understanding of the market in which it operates and because OEMs have told Microsoft as much. Indicative of Microsoft’s assessment of the situation is the fact that, in a 1996 presentation to the firm’s executive committee, the Microsoft executive in charge of OEM licensing reported that piracy continued to be the main competition to the company’s operating system products. Secure in this knowledge, Microsoft did not consider the prices of other Intel-compatible PC operating systems when it set the price of Windows 98.

56. As the Court found above, the growth of server- and middleware-based applications development might eventually weaken the applications barrier to entry. This would not only make it easier for outside firms to enter the market, it could also make it easier for non-Microsoft firms already in the market to present a viable alternative to Windows. But as the Court also found above, it is not clear whether ISVs will ever develop a large, diverse body of full-featured applications that rely solely on APIs exposed by servers and middleware. Furthermore, even assuming that such a movement has already begun in earnest, it will take several years for the applications barrier to erode enough to enable a non-Microsoft, Intel-compatible PC operating system to develop into a viable alternative to Windows.

D. Price Restraint Posed by Microsoft’s Installed Base

57. Software never expires, so consumers who already have a version of Windows with which they are content and who are not shopping for a new PC system are somewhat reluctant to incur the cost of upgrading to a new version of Windows. Fortunately for Microsoft, the pace of innovation in PC hardware is rapid, and the price of that hardware has declined steadily in recent years. As a result, existing PC users buy new PC systems relatively frequently, and OEMs still attract at a healthy rate buyers who have never owned a computer. The license for one of Microsoft’s operating system products prohibits the user from transferring the operating system to another machine, so there is no legal secondary market in Microsoft operating systems. This means that any consumer who buys a new Intel-compatible PC and wants Windows must buy a new copy of the operating system. Microsoft takes pains to ensure that the versions of its operating system that OEMs pre-install on new PC systems are the most current. It does this, in part, by increasing the price to OEMs of older versions of Windows when the newer versions are released. Since Microsoft can sell so many copies of each new operating system through the sales of new PC systems, the average price it sets for those systems is little affected by the fact that older versions of Windows never wear out.

58. Although there is no legal secondary market for Microsoft’s PC operating systems, there is a thriving illegal one. Software pirates illegally copy software products such as Windows, selling each copy for a fraction of the vendor’s usual price. One of the ways Microsoft combats piracy is by advising OEMs that they will be charged a higher price for Windows unless they drastically limit the number of PCs that they sell without an operating system pre-installed. In 1998, all major OEMs agreed to this restriction. Naturally, it is hard to sell a pirated copy of Windows to a consumer who has already received a legal copy included in the price of his new PC system. Thus, Microsoft is able to effectively contain, if not extinguish, the illegal secondary market for its operating-system products. So even though Microsoft is more concerned about piracy than it is about other firms’ operating system products, the company’s pricing is not substantially constrained by the need to reduce the incentives for consumers to acquire their copies of Windows illegally.

59. The software industry in general is characterized by dynamic, vigorous competition. In many cases, one of the early entrants into a new software category quickly captures a lion’s share of the sales, while other products in the category are either driven out altogether or relegated to niche positions. What eventually displaces the leader is often not competition from another product within the same software category, but rather a technological advance that renders the boundaries defining the category obsolete. These events, in which categories are redefined and leaders are superseded in the process, are spoken of as "inflection points."

60. The exponential growth of the Internet represents an inflection point born of complementary technological advances in the computer and telecommunications industries. The rise of the Internet in turn has fueled the growth of server-based computing, middleware, and open-source software development. Working together, these nascent paradigms could oust the PC operating system from its position as the primary platform for applications development and the main interface between users and their computers. Microsoft recognizes that new paradigms could arise to depreciate the value of selling PC operating systems; however, the fact that these new paradigms already exist in embryonic or primitive form does not prevent Microsoft from enjoying monopoly power today. For while consumers might one day turn to network computers, or Linux, or a combination of middleware and some other operating system, as an alternative to Windows, the fact remains that they are not doing so today. Nor are consumers likely to do so in appreciable numbers any time in the next few years. Unless and until that day arrives, no significant percentage of consumers will be able to abandon Windows without incurring substantial costs. Microsoft can therefore set the price of Windows substantially higher than that which would be charged in a competitive market — or impose other burdens on consumers — without losing so much business as to make the action unprofitable. If Microsoft exerted its power solely to raise price, the day when users could turn away from Windows without incurring substantial costs would still be several years distant. Moreover, Microsoft could keep its prices high for a significant period of time and still lower them in time to meet the threat of a new paradigm. Alternatively, Microsoft could delay the arrival of a new paradigm on the scene by expending surplus monopoly power in ways other than the maintenance of high prices.

61. The fact that Microsoft invests heavily in research and development does not evidence a lack of monopoly power. Indeed, Microsoft has incentives to innovate aggressively despite its monopoly power. First, if there are innovations that will make Intel-compatible PC systems attractive to more consumers, and those consumers less sensitive to the price of Windows, the innovations will translate into increased profits for Microsoft. Second, although Microsoft could significantly restrict its investment in innovation and still not face a viable alternative to Windows for several years, it can push the emergence of competition even farther into the future by continuing to innovate aggressively. While Microsoft may not be able to stave off all potential paradigm shifts through innovation, it can thwart some and delay others by improving its own products to the greater satisfaction of consumers.

62. Microsoft’s actual pricing behavior is consistent with the proposition that the firm enjoys monopoly power in the market for Intel-compatible PC operating systems. The company’s decision not to consider the prices of other vendors’ Intel-compatible PC operating systems when setting the price of Windows 98, for example, is probative of monopoly power. One would expect a firm in a competitive market to pay much closer attention to the prices charged by other firms in the market. Another indication of monopoly power is the fact that Microsoft raised the price that it charged OEMs for Windows 95, with trivial exceptions, to the same level as the price it charged for Windows 98 just prior to releasing the newer product. In a competitive market, one would expect the price of an older operating system to stay the same or decrease upon the release of a newer, more attractive version. Microsoft, however, was only concerned with inducing OEMs to ship Windows 98 in favor of the older version. It is unlikely that Microsoft would have imposed this price increase if it were genuinely concerned that OEMs might shift their business to another vendor of operating systems or hasten the development of viable alternatives to Windows.

63. Finally, it is indicative of monopoly power that Microsoft felt that it had substantial discretion in setting the price of its Windows 98 upgrade product (the operating system product it sells to existing users of Windows 95). A Microsoft study from November 1997 reveals that the company could have charged $49 for an upgrade to Windows 98 — there is no reason to believe that the $49 price would have been unprofitable — but the study identifies $89 as the revenue-maximizing price. Microsoft thus opted for the higher price.

64. An aspect of Microsoft’s pricing behavior that, while not tending to prove monopoly power, is consistent with it is the fact that the firm charges different OEMs different prices for Windows, depending on the degree to which the individual OEMs comply with Microsoft’s wishes. Among the five largest OEMs, Gateway and IBM, which in various ways have resisted Microsoft’s efforts to enlist them in its efforts to preserve the applications barrier to entry, pay higher prices than Compaq, Dell, and Hewlett-Packard, which have pursued less contentious relationships with Microsoft.

65. It is not possible with the available data to determine with any level of confidence whether the price that a profit-maximizing firm with monopoly power would charge for Windows 98 comports with the price that Microsoft actually charges. Even if it could be determined that Microsoft charges less than the profit-maximizing monopoly price, though, that would not be probative of a lack of monopoly power, for Microsoft could be charging what seems like a low short-term price in order to maximize its profits in the future for reasons unrelated to underselling any incipient competitors. For instance, Microsoft could be stimulating the growth of the market for Intel-compatible PC operating systems by keeping the price of Windows low today. Given the size and stability of its market share, Microsoft stands to reap almost all of the future rewards if there are yet more consumers of Intel-compatible PC operating systems. By pricing low relative to the short-run profit-maximizing price, thereby focusing on attracting new users to the Windows platform, Microsoft would also intensify the positive network effects that add to the impenetrability of the applications barrier to entry.

66. Furthermore, Microsoft expends a significant portion of its monopoly power, which could otherwise be spent maximizing price, on imposing burdensome restrictions on its customers — and in inducing them to behave in ways — that augment and prolong that monopoly power. For example, Microsoft attaches to a Windows license conditions that restrict the ability of OEMs to promote software that Microsoft believes could weaken the applications barrier to entry. Microsoft also charges a lower price to OEMs who agree to ensure that all of their Windows machines are powerful enough to run Windows NT for Workstations. To the extent this provision induces OEMs to concentrate their efforts on the development of relatively powerful, expensive PCs, it makes OEMs less likely to pursue simultaneously the opposite path of developing "thin client" systems, which could threaten demand for Microsoft’s Intel-compatible PC operating system products. In addition, Microsoft charges a lower price to OEMs who agree to ship all but a minute fraction of their machines with an operating system pre-installed. While this helps combat piracy, it also makes it less likely that consumers will detect increases in the price of Windows and renders operating systems not pre-installed by OEMs in large numbers even less attractive to consumers. After all, a consumer’s interest in a non-Windows operating system might not outweigh the burdens on system memory and performance associated with supporting two operating systems on a single PC. Other such restrictions and incentives are described below.

67. Microsoft’s monopoly power is also evidenced by the fact that, over the course of several years, Microsoft took actions that could only have been advantageous if they operated to reinforce monopoly power. These actions are described below.

68. Middleware technologies, as previously noted, have the potential to weaken the applications barrier to entry. Microsoft was apprehensive that the APIs exposed by middleware technologies would attract so much developer interest, and would become so numerous and varied, that there would arise a substantial and growing number of full-featured applications that relied largely, or even wholly, on middleware APIs. The applications relying largely on middleware APIs would potentially be relatively easy to port from one operating system to another. The applications relying exclusively on middleware APIs would run, as written, on any operating system hosting the requisite middleware. So the more popular middleware became and the more APIs it exposed, the more the positive feedback loop that sustains the applications barrier to entry would dissipate. Microsoft was concerned with middleware as a category of software; each type of middleware contributed to the threat posed by the entire category. At the same time, Microsoft focused its antipathy on two incarnations of middleware that, working together, had the potential to weaken the applications barrier severely without the assistance of any other middleware. These were Netscape’s Web browser and Sun’s implementation of the Java technologies.

69. Netscape Navigator possesses three key middleware attributes that endow it with the potential to diminish the applications barrier to entry. First, in contrast to non-Microsoft, Intel-compatible PC operating systems, which few users would want to use on the same PC systems that carry their copies of Windows, a browser can gain widespread use based on its value as a complement to Windows. Second, because Navigator exposes a set (albeit a limited one) of APIs, it can serve as a platform for other software used by consumers. A browser product is particularly well positioned to serve as a platform for network-centric applications that run in association with Web pages. Finally, Navigator has been ported to more than fifteen different operating systems. Thus, if a developer writes an application that relies solely on the APIs exposed by Navigator, that application will, without any porting, run on many different operating systems.

70. Adding to Navigator’s potential to weaken the applications barrier to entry is the fact that the Internet has become both a major inducement for consumers to buy PCs for the first time and a major occupier of the time and attention of current PCs users. For any firm looking to turn its browser product into an applications platform such to rival Windows, the intense consumer interest in all things Internet-related is a great boon.

71. Microsoft knew in the fall of 1994 that Netscape was developing versions of a Web browser to run on different operating systems. It did not yet know, however, that Netscape would employ Navigator to generate revenue directly, much less that the product would evolve in such a way as to threaten Microsoft. In fact, in late December 1994, Netscape’s chairman and chief executive officer ("CEO"), Jim Clark, told a Microsoft executive that the focus of Netscape’s business would be applications running on servers and that Netscape did not intend to succeed at Microsoft’s expense.

72. As soon as Netscape released Navigator on December 15, 1994, the product began to enjoy dramatic acceptance by the public; shortly after its release, consumers were already using Navigator far more than any other browser product. This alarmed Microsoft, which feared that Navigator’s enthusiastic reception could embolden Netscape to develop Navigator into an alternative platform for applications development. In late May 1995, Bill Gates, the chairman and CEO of Microsoft, sent a memorandum entitled "The Internet Tidal Wave" to Microsoft’s executives describing Netscape as a "new competitor ‘born’ on the Internet." He warned his colleagues within Microsoft that Netscape was "pursuing a multi-platform strategy where they move the key API into the client to commoditize the underlying operating system." By the late spring of 1995, the executives responsible for setting Microsoft’s corporate strategy were deeply concerned that Netscape was moving its business in a direction that could diminish the applications barrier to entry.

73. The term "Java" refers to four interlocking elements. First, there is a Java programming language with which developers can write applications. Second, there is a set of programs written in Java that expose APIs on which developers writing in Java can rely. These programs are called the "Java class libraries." The third element is the Java compiler, which translates the code written by the developer into Java "bytecode." Finally, there are programs called "Java virtual machines," or "JVMs," which translate Java bytecode into instructions comprehensible to the underlying operating system. If the Java class libraries and a JVM are present on a PC system, the system is said to carry a "Java runtime environment."

74. The inventors of Java at Sun Microsystems intended the technology to enable applications written in the Java language to run on a variety of platforms with minimal porting. A program written in Java and relying only on APIs exposed by the Java class libraries will run on any PC system containing a JVM that has itself been ported to the resident operating system. Therefore, Java developers need to port their applications only to the extent that those applications rely directly on the APIs exposed by a particular operating system. The more an application written in Java relies on APIs exposed by the Java class libraries, the less work its developer will need to do to port the application to different operating systems. The easier it is for developers to port their applications to different operating systems, the more applications will be written for operating systems other than Windows. To date, the Java class libraries do not expose enough APIs to support the development of full-featured applications that will run well on multiple operating systems without the need for porting; however, they do allow relatively simple, network-centric applications to be written cross-platform. It is Sun’s ultimate ambition to expand the class libraries to such an extent that many full-featured, end-user-oriented applications will be written cross-platform. The closer Sun gets to this goal of "write once, run anywhere," the more the applications barrier to entry will erode.

75. Sun announced in May 1995 that it had developed the Java programming language. Mid-level executives at Microsoft began to express concern about Sun’s Java vision in the fall of that year, and by late spring of 1996, senior Microsoft executives were deeply worried about the potential of Sun’s Java technologies to diminish the applications barrier to entry.

76. Sun’s strategy could only succeed if a Java runtime environment that complied with Sun’s standards found its way onto PC systems running Windows. Sun could not count on Microsoft to ship with Windows an implementation of the Java runtime environment that threatened the applications barrier to entry. Fortunately for Sun, Netscape agreed in May 1995 to include a copy of Sun’s Java runtime environment with every copy of Navigator, and Navigator quickly became the principal vehicle by which Sun placed copies of its Java runtime environment on the PC systems of Windows users.

77. The combined efforts of Netscape and Sun threatened to hasten the demise of the applications barrier to entry, opening the way for non-Microsoft operating systems to emerge as acceptable substitutes for Windows. By stimulating the development of network-centric Java applications accessible to users through browser products, the collaboration of Netscape and Sun also heralded the day when vendors of information appliances and network computers could present users with viable alternatives to PCs themselves. Nevertheless, these middleware technologies have a long way to go before they might imperil the applications barrier to entry. Windows 98 exposes nearly ten thousand APIs, whereas the combined APIs of Navigator and the Java class libraries, together representing the greatest hope for proponents of middleware, total less than a thousand. Decision-makers at Microsoft are apprehensive of potential as well as present threats, though, and in 1995 the implications of the symbiosis between Navigator and Sun’s Java implementation were not lost on executives at Microsoft, who viewed Netscape’s cooperation with Sun as a further reason to dread the increasing use of Navigator.

78. Although they have been the most prominent, Netscape’s Navigator and Sun’s Java implementation are not the only manifestations of middleware that Microsoft has perceived as having the potential to weaken the applications barrier to entry. Starting in 1994, Microsoft exhibited considerable concern over the software product Notes, distributed first by Lotus and then by IBM. Microsoft worried about Notes for several reasons: It presented a graphical interface that was common across multiple operating systems; it also exposed a set of APIs to developers; and, like Navigator, it served as a distribution vehicle for Sun’s Java runtime environment. Then in 1995, Microsoft reacted with alarm to Intel’s Native Signal Processing software, which interacted with the microprocessor independently of the operating system and exposed APIs directly to developers of multimedia content. Finally, in 1997 Microsoft noted the dangers of Apple’s and RealNetworks’ multimedia playback technologies, which ran on several platforms (including the Mac OS and Windows) and similarly exposed APIs to content developers. Microsoft feared all of these technologies because they facilitated the development of user-oriented software that would be indifferent to the identity of the underlying operating system.

A. Microsoft’s Attempt to Dissuade Netscape from Developing Navigator as a Platform

79. Microsoft’s first response to the threat posed by Navigator was an effort to persuade Netscape to structure its business such that the company would not distribute platform-level browsing software for Windows. Netscape’s assent would have ensured that, for the foreseeable future, Microsoft would produce the only platform-level browsing software distributed to run on Windows. This would have eliminated the prospect that non-Microsoft browsing software could weaken the applications barrier to entry.

80. Executives at Microsoft received confirmation in early May 1995 that Netscape was developing a version of Navigator to run on Windows 95, which was due to be released in a couple of months. Microsoft’s senior executives understood that if they could prevent this version of Navigator from presenting alternatives to the Internet-related APIs in Windows 95, the technologies branded as Navigator would cease to present an alternative platform to developers. Even if non-Windows versions of Navigator exposed Internet-related APIs, applications written to those APIs would not run on the platform Microsoft executives expected to enjoy the largest installed base, i.e., Windows 95. So, as long as the version of Navigator written for Windows 95 relied on Microsoft’s Internet-related APIs instead of exposing its own, developing for Navigator would not mean developing cross-platform. Developers of network-centric applications thus would not be drawn to Navigator’s APIs in substantial numbers. Therefore, with the encouragement and support of Gates, a group of Microsoft executives commenced a campaign in the summer of 1995 to convince Netscape to halt its development of platform-level browsing technologies for Windows 95.

81. In a meeting held at Microsoft’s headquarters on June 2, 1995, Microsoft executives suggested to Jim Clark’s replacement as CEO at Netscape, James Barksdale, that the version of Navigator written for Windows 95 be designed to rely upon the Internet-related APIs in Windows 95 and distinguish itself with "value-added" software components. The Microsoft executives left unsaid the fact that value-added software, by definition, does not present a significant platform for applications development. For his part, Barksdale informed the Microsoft representatives that the browser represented an important part of Netscape’s business strategy and that Windows 3.1 and Windows 95 were expected to be the primary platforms for which Navigator would be distributed.

82. At the conclusion of the June 2 meeting, Microsoft still did not know whether or not Netscape intended to preserve Navigator’s own platform capabilities and expand the set of APIs that it exposed to developers. In the hope that Netscape could still be persuaded to forswear any platform ambitions and instead rely on the Internet technologies in Windows 95, Microsoft accepted Barksdale’s invitation to send a group of representatives to Netscape’s headquarters for a technology "brainstorming session" on June 21. Netscape’s senior executives saw the meeting as an opportunity to ask Microsoft for access to crucial technical information, including certain APIs, that Netscape needed in order to ensure that Navigator would work well on systems running Windows 95.

83. Early in the June 21 meeting, Microsoft representatives told Barksdale and the other Netscape executives present that they wanted to explore the possibility of building a broader and closer relationship between the two companies. To this end, the Microsoft representatives wanted to know whether Netscape intended to adopt and build on top of the Internet-related platform that Microsoft planned to include in Windows 95, or rather to expose its own Internet-related APIs, which would compete with Microsoft’s. If Netscape was not committed to providing an alternative platform for network-centric applications, Microsoft would assist Netscape in developing server- and (to a limited extent) PC-based software applications that relied on Microsoft’s Internet technologies. For one thing, the representatives explained, Microsoft would be content to leave the development of browser products for the Mac OS, UNIX, and Microsoft’s 16-bit operating system products to Netscape. Alternatively, Netscape could license to Microsoft the underlying code for a Microsoft-branded browser to run on those platforms. The Microsoft representatives made it clear, however, that Microsoft would be marketing its own browser for Windows 95, and that this product would rely on Microsoft’s platform-level Internet technologies. If Netscape marketed browsing software for Windows 95 based on different technologies, then Microsoft would view Netscape as a competitor, not a partner.

84. When Barksdale brought the discussion back to the particular Windows 95 APIs that Netscape actually wanted to rely on and needed from Microsoft, the representatives from Microsoft explained that if Netscape entered a "special relationship" with Microsoft, the company would treat Netscape as a "preferred ISV." This meant that Netscape would enjoy preferential access to technical information, including APIs. They intimated that Microsoft’s internal developers had already created the APIs that Netscape was seeking, and that Microsoft had not yet decided either which ISVs would be privileged to receive them or when access would be granted. The Microsoft representatives made clear that the alacrity with which Netscape would receive the desired Windows 95 APIs and other technical information would depend on whether Netscape entered this "special relationship" with Microsoft.

85. After listening to Microsoft’s proposal, Barksdale had two main questions: First, where would the line between platform (Microsoft’s exclusive domain) and applications (where Netscape could continue to function) be situated? Second, who would get to decide where the line would lie? After all, the attractiveness of a special relationship with Microsoft depended a great deal on how much room would remain for Netscape to innovate and seek profit. The Microsoft representatives replied that Microsoft would incorporate most of the functionality of the current Netscape browser into the Windows 95 platform, perhaps leaving room for Netscape to distribute a user-interface shell. Where Netscape would have the most scope to innovate would be in the development of software "solutions," which are applications (mainly server-based) focused on meeting the needs of specific types of commercial users. Since such applications are already minutely calibrated to the needs of their users, they do not present platforms for the development of more specific applications. Although the representatives from Microsoft assured Barksdale that the line between platform and solutions was fixed by a collaborative decision-making process between Microsoft and its ISV partners, those representatives had already indicated that the space Netscape would be allowed to occupy between the user and Microsoft’s platform domain was a very narrow one. Simply put, if Navigator exposed APIs that competed for developer attention with the Internet-related APIs Microsoft was planning to build into its platform, Microsoft would regard Netscape as a trespasser on its territory.

86. The Microsoft representatives did not insist at the June 21 meeting that Netscape executives accept their proposal on the spot. For his part, Barksdale said only that he would like more information regarding where Microsoft proposed to place the line between its platform and Netscape’s applications. In the ensuing, more technical discussions, the Netscape executives agreed to adopt one component of Microsoft’s platform-level Internet technology called Internet Shortcuts. The meeting ended cordially, with both sides promising to keep the lines of communication open.

87. The executive who led Microsoft’s contingent on June 21, Daniel Rosen, emerged from the meeting optimistic that Netscape would abandon its platform ambitions in exchange for special help from Microsoft in developing solutions. His sentiments were not shared by another Microsoft participant, Thomas Reardon, who had not failed to notice the Netscape executives grow tense when the Microsoft representatives referred to incorporating Navigator’s functionality into Windows. Reardon predicted that Netscape would compete with almost all of Microsoft’s platform-level Internet technologies. Once he heard both viewpoints, Gates concluded that Rosen was being a bit naive and that Reardon had assessed the situation more accurately. In the middle of July 1995, Rosen’s superiors instructed him to drop the effort to reach a strategic concord with Netscape.

88. Had Netscape accepted Microsoft’s proposal, it would have forfeited any prospect of presenting a comprehensive platform for the development of network-centric applications. Even if the versions of Navigator written for the Mac OS, UNIX, and 16-bit Windows had continued to expose APIs controlled by Netscape, the fact that Netscape would not have marketed any platform software for Windows 95, the operating system that was destined to become dominant, would have ensured that, for the foreseeable future, too few developers would rely on Navigator’s APIs to create a threat to the applications barrier to entry. In fact, although the discussions ended before Microsoft was compelled to demarcate precisely where the boundary between its platform and Netscape’s applications would lie, it is unclear whether Netscape’s acceptance of Microsoft’s proposal would have left the firm with even the ability to survive as an independent business.

89. At the time Microsoft presented its proposal, Navigator was the only browser product with a significant share of the market and thus the only one with the potential to weaken the applications barrier to entry. Thus, had it convinced Netscape to accept its offer of a "special relationship," Microsoft quickly would have gained such control over the extensions and standards that network-centric applications (including Web sites) employ as to make it all but impossible for any future browser rival to lure appreciable developer interest away from Microsoft’s platform.

90. Microsoft knew that Netscape needed certain critical technical information and assistance in order to complete its Windows 95 version of Navigator in time for the retail release of Windows 95. Indeed, Netscape executives had made a point of requesting this information, especially the so-called Remote Network Access ("RNA") API, at the June 21 meeting. As was discussed above, the Microsoft representatives at the meeting had responded that the haste with which Netscape received the desired technical information would depend on whether Netscape entered the so-called "special relationship" with Microsoft. Specifically, Microsoft representative J. Allard had told Barksdale that the way in which the two companies concluded the meeting would determine whether Netscape received the RNA API immediately or in three months.

91. Although Netscape declined the special relationship with Microsoft, its executives continued, over the weeks following the June 21 meeting, to plead for the RNA API. Despite Netscape’s persistence, Microsoft did not release the API to Netscape until late October, i.e., as Allard had warned, more than three months later. The delay in turn forced Netscape to postpone the release of its Windows 95 browser until substantially after the release of Windows 95 (and Internet Explorer) in August 1995. As a result, Netscape was excluded from most of the holiday selling season.

92. Microsoft similarly withheld a scripting tool that Netscape needed to make its browser compatible with certain dial-up ISPs. Microsoft had licensed the tool freely to ISPs that wanted it, and in fact had cooperated with Netscape in drafting a license agreement that, by mid-July 1996, needed only to be signed by an authorized Microsoft executive to go into effect. There the process halted, however. In mid-August, a Microsoft representative informed Netscape that senior executives at Microsoft had decided to link the grant of the license to the resolution of all open issues between the companies. Netscape never received a license to the scripting tool, and as a result, was unable to do business with certain ISPs for a time.

93. Other firms in the computer industry have had encounters with Microsoft similar to the experiences of Netscape described above. These interactions demonstrate that it is Microsoft’s corporate practice to pressure other firms to halt software development that either shows the potential to weaken the applications barrier to entry or competes directly with Microsoft’s most cherished software products.

94. At the same time that Microsoft was trying to convince Netscape to stop developing cross-platform APIs, it was trying to convince Intel to halt the development of software that presented developers with a set of operating-system-independent interfaces.

95. Although Intel is engaged principally in the design and manufacture of microprocessors, it also develops some software. Intel’s software development efforts, which take place at the Intel Architecture Labs ("IAL"), are directed primarily at finding useful ways to consume more microprocessor cycles, thereby stimulating demand for advanced Intel microprocessors. By early 1995, IAL was in the advanced stages of developing software that would enable Intel 80x86 microprocessors to carry out tasks usually performed by separate chips known as "digital signal processors." By enabling this migration, the software, called Native Signal Processing ("NSP") software, would endow Intel microprocessors with substantially enhanced video and graphics performance.

96. Intel was eager for software developers and hardware manufacturers to write software and build peripheral devices that would implement the enhanced capabilities that its microprocessors and its NSP software together offered. Intel did not believe, however, that the set of APIs and device driver interfaces ("DDIs") in Windows had kept pace with the growing ability of Intel’s microprocessors to deliver audio/visual content. Consequently, IAL designed its NSP software to expose Intel’s own APIs and DDIs that, when invoked by developers and hardware manufacturers, would demonstrate the multimedia capabilities of an Intel microprocessor utilizing NSP.

97. Microsoft reacted to Intel’s NSP software with alarm. First of all, the software threatened to offer ISVs and device manufacturers an alternative to waiting for Windows to provide system-level support for products that would take advantage of advances in hardware technology. More troubling was the fact that Intel was developing versions of its NSP software for non-Microsoft operating systems. The different versions of the NSP software exposed the same set of software interfaces to developers, so the more an application took advantage of interfaces exposed by NSP software, the easier it would be to port that application to non-Microsoft operating systems. In short, Intel’s NSP software bore the potential to weaken the barrier protecting Microsoft’s monopoly power.

98. Over time, Microsoft developed additional qualms about Intel’s NSP software. For instance, Intel initially designed the NSP software to be compatible with only Windows 3.1. At the time, Microsoft was preparing to release Windows 95, and the company did not want anything rekindling the interest of ISVs, equipment manufacturers, and consumers in the soon-to-be obsolescent version of Windows. More acute was Microsoft’s concern that users who received NSP software on their Windows 3.1 systems would have difficulty upgrading those systems to Windows 95. By June 1995, Intel had completed a pre-release, or "beta," version of its NSP software for Windows 95, but Microsoft worried that a commercial version would not be ready by the time OEMs began loading Windows 95.

99. Along with its concerns about contemporaneous compatibility, Microsoft also complained that Intel had not subjected its software to sufficient quality-assurance testing. Microsoft was quick to point out that if Windows users detected problems with the software that came pre-installed on their PC systems, they would blame Microsoft or the OEMs, even if fault lay with Intel. Microsoft’s concerns with compatibility and quality were genuine. Both pre-dating and over-shadowing these transient and remediable concerns, however, was a more abiding fear at Microsoft that the NSP software would render ISVs, device manufacturers, and (ultimately) consumers less dependent on Windows. Without this fear, Microsoft would not have subjected Intel to the level of pressure that it brought to bear in the summer of 1995.

100. Microsoft began complaining to Intel about its NSP software in inter-company communications sent in the spring of 1995. In May, Microsoft raised the profile of its complaints by sending some of its senior executives to Intel to discuss the latter’s incursion into Microsoft’s platform territory. Returning from the May meeting, one Microsoft employee urged his superiors to refuse to allow Intel to offer platform-level software, even if it meant that Intel could not innovate as quickly as it would like. If Intel wished to enable a new function, the employee wrote, its only "winning path" would be to convince Microsoft to support the effort in its platform software. At any rate, "[s]ometimes Intel would have to accept the outcome that the time isn’t right for [Microsoft]." In the first week of July, Gates himself met with Intel’s CEO, Andrew Grove, to discuss, among other things, NSP. In a subsequent memorandum to senior Microsoft executives, Gates reported that he had tried to convince Grove "to basically not ship NSP" and more generally to reduce the number of people working on software at Intel.

101. The development of an alternative platform to challenge Windows was not the primary objective of Intel’s NSP efforts. In fact, Intel was interested in providing APIs and DDIs only to the extent the effort was necessary to ensure the development of applications and devices that would spark demand for Intel’s most advanced microprocessors. Understanding Intel’s limited ambitions, Microsoft hastened to assure Intel that if it would stop promoting NSP’s interfaces, Microsoft would accelerate its own work to incorporate the functions of the NSP software into Windows, thereby stimulating the development of applications and devices that relied on the new capabilities of Intel’s microprocessors. At the same time, Microsoft pressured the major OEMs to not install NSP software on their PCs until the software ceased to expose APIs. NSP software could not find its way onto PCs without the cooperation of the OEMs, so Intel realized that it had no choice but to surrender the pace of software innovation to Microsoft. By the end of July 1995, Intel had agreed to stop promoting its NSP software. Microsoft subsequently incorporated some of NSP’s components into its operating-system products. Even as late as the end of 1998, though, Microsoft still had not implemented key capabilities that Intel had been poised to offer consumers in 1995.

102. Microsoft was not content to merely quash Intel’s NSP software. At a second meeting at Intel’s headquarters on August 2, 1995, Gates told Grove that he had a fundamental problem with Intel using revenues from its microprocessor business to fund the development and distribution of free platform-level software. In fact, Gates said, Intel could not count on Microsoft to support Intel’s next generation of microprocessors as long as Intel was developing platform-level software that competed with Windows. Intel’s senior executives knew full well that Intel would have difficultly selling PC microprocessors if Microsoft stopped cooperating in making them compatible with Windows and if Microsoft stated to OEMs that it did not support Intel’s chips. Faced with Gates’ threat, Intel agreed to stop developing platform-level interfaces that might draw support away from interfaces exposed by Windows.

103. OEMs represent the primary customers for Intel’s microprocessors. Since OEMs are dependent on Microsoft for Windows, Microsoft enjoys continuing leverage over Intel. To illustrate, Gates was able to report to other senior Microsoft executives in October 1995 that "Intel feels we have all the OEMs on hold with our NSP chill." He added:

This is good news because it means OEMs are listening to us. Andy [Grove] believes Intel is living up to its part of the NSP bargain and that we should let OEMs know that some of the new software work Intel is doing is OK. If Intel is not sticking totally to its part of the deal let me know.