I. Introduction

The U.S. software industry is characterized by astonishing levels of growth, innovative activity, and competition.1 Some argue that innovation in software and related industries has driven much of the innovation in other industries in recent decades. Federal government statistics suggest that it is one of the few information technology sectors that consistently shows a large trade surplus,3 and as the pressures of globalization dilute the comparative advantage of American employees in many sectors, it is worth noting the remarkable level of employment growth in the software industry over the last decade, from 854,000 jobs in 1992 to more than 2.1 million jobs in 2000 (a 12% annual growth rate).4

Academics, however, generally see a different picture. They see an industry burdened by an intellectual property (IP) system which grants so many software patents5 that small companies cannot effectively innovate.6 That perspective interests me for several reasons. First, unless it is merely a broader attack on the entire IP system,7 it assumes that innovation in software is so different from innovation in other areas that traditional IP protections are inappropriate. It also is at odds with my general skepticism about the deterministic effect of legal institutions. My intuition is to doubt that legal rules granting patent protection have a sufficiently substantial effect to alter the course of innovation in either direction.

The existing literature on the subject8 focuses on the nature and effects of software patents. Most writers proceed by identifying patents that fall within the PTO or IPC classes that correspond most closely to software innovation; they then examine data about the performance and behavior of the firms that hold those patents. That approach-although useful in examining the nature of software patents and the work of the PTO-has two major disadvantages for the broader agenda of evaluating the effects of software patents. First, the quality of the research depends entirely on the propriety of the definition of a "software" patent. Because software is a recently devised technology, it does not fall naturally within any particular class or classes. Thus, any definition that relies on patent classes is to some degree arbitrary.9 Moreover, large manufacturing firms (Ford, GM, and the like) outside the software industry hold the overwhelming majority of the patents that such papers analyze.10 Thus, it is unclear whether the empirical results reflect the effects of the software patents or whether similar results would be obtained for firms that receive substantial revenues from the sale of software products or services.

This Article rejects that approach, opting instead to analyze innovation in the software industry itself.11 My approach identifies firms that develop software and then studies the effects of IP on the behavior of those firms. Thus, I focus on firms like IBM, Microsoft, and their smaller competitors that often are ignored in the existing research. My project also examines the smaller firms in the industry, rather than looking exclusively at large, publicly traded companies.12 Given the importance of small companies to software innovation-a major theme of this Article-that extension is a substantial analytical advance.

My methodology is empirical and analytical. I rely on a set of about 60 interviews13 with a variety of professionals14 knowledgeable about the software industry: software developers, venture capitalists, angel investors, banks that lend to software startups, large software and hardware firms, and knowledgeable attorneys. Those interviews are designed to provide qualitative information about the motivations and practices that form the institutional environment within which software firms operate.

Analytically, I connect the interviews to well-developed literature relevant to questions the project raises. Initially, I account for a substantial body of doctrinal scholarship examining the question of how to accommodate existing IP law to the nature of innovation in the software industry.15 Because my goal is to understand the relation between IP and innovation, I also engage the rich and varied economic analysis of innovation. That literature includes formal and informal analyses of how best to allocate the profits from ideas among the various actors in a sequential scheme of innovation,16 historical analyses about the effects patents have had over time,17 and empirical analyses (relying primarily on questionnaires) regarding the value of patents in reaping the profits of innovation in various industries.18 Finally, because I am interested in the ability of patents to facilitate financing of software firms, I examine empirical studies of entrepreneurial innovation, which consider the nature and effects of venture capital investing.19

Any effort to examine the relation between patents and innovation must proceed with modest goals. It is not plausible to think that researchers can obtain the evidence necessary to determine whether patents cause innovation in an industry to proceed at an optimal rate in optimal directions.20 Thus, my work here is consciously imprecise. My goal is to provide a richer understanding of the possible effects that patents have in the software industry. Using this methodology, I can only exclude explanations that are inconsistent with events "on the ground." I cannot hope to provide a comprehensive or definitive account of the effects of patents on innovation.

With that in mind, I set the stage in Part II with an overview of the software industry's structure and the debate about patenting in that industry. I follow with three substantive Parts that discuss in turn the potential benefits of patents in the industry, the potential costs of patents in the industry, and the role of copyright, trade secret, and other alternative schemes that a firm might use to protect its software-related innovations. Although the analysis is often contextual, it does have an overriding theme: The effects of patents are much more likely to benefit small firms and contribute to industry fragmentation than to benefit large firms and contribute to industry concentration.21 Among other things, I find the common thesis that large firms use a patent "thicket" to fence out potential competitors most implausible. I close with a brief and skeptical discussion of several possible responses to the imperfections of the existing system.

II. The Software Industry

It is important to begin with an understanding of the industry's structure. The industry is young. It generally is regarded as originating in the mid-1960s.22 The concept of the software product-a product designed by firm A and sold to firm B for use on firm B's computer-first originated because of the increasing complexity of software23 and a shortage of the labor needed for each hardware firm to make its own software.24 The most crucial event was IBM's decision in late 1968 to "unbundle" its software from its hardware.25 Sales of software products grew rapidly throughout the 1970s. By the 1980s, the United States had a large and well-developed corporate software products industry with more than 1,800 firms.26

The industry was not, however, fated to retain the unitary status it had when it first evolved out of the IBM-dominated days of the 1960s. On the contrary, the last quarter-century has seen a succession of events that have repeatedly broadened the industry's traditional focus on software for use by business enterprises. The number of developments makes any list of key events arbitrary, but for my purposes the first salient landmark in the fragmentation of the industry was the introduction of the personal computer in the mid-1970s. That development rapidly led to a largely separate set of companies producing software for personal computers.27 The popularization of the graphical user interface in the early 1990s brought with it an increasingly large role for Microsoft, but to this day dozens of competitors continue to provide significant products for personal computers.28 Another sector of the industry that arose by the late 1990s is the massive sector producing games and other entertainment software.29 With the maturation of the software industry, the last decade saw the emergence of yet another large sector that specializes in software specifically designed to facilitate software development and design.

The relations among these sectors are complicated by the fact that the line between software products and services is a shifting one, with substantially different business models for firms that specialize in the sale of products and the sale of services.30 Complicating that point, many firms operate in multiple sectors, developing complex strategies to sell hardware or one type of software product at below-market prices to foster profitable businesses in other sectors. So, for example, Microsoft markets the Xbox at a price below its cost, hoping in the future to profit from sales of games.31 Adobe markets software to consumers for free, hoping to profit from sales of enterprise software.32 Finally, the rise of the Internet brought first a tremendous influx of capital into the industry and then a subsequent crash and weeding out when companies were not able to produce results that justified the elevated equity valuations of 2000 and 2001. Indeed, that weeding out continues even now, as consolidation continues to be a prominent trend in the industry.33

As I write, many believe that a major fissure is developing between proprietary and open source models of software development.34 The traditional model of development has been a proprietary one in which a firm develops a product and then profits through sales of that product. Recently, some firms have rejected that model, at least in part, to engage in open source development.35 Open source development generally proceeds on the premise that software products developed under that model are not subject to the proprietary control of any individual or firm.36 I discuss the relation between open source development and patent protection in subpart IV(B), but two points are important to the present discussion. First, the demarcation between the two models is not as complete as open source proponents might suggest. As suggested above, proprietary firms often can take advantage of open source development of one product to foster a profitable market position in a related product.37 Second, to the extent there is in fact an important demarcation between proprietary and open source development, this project is focused on firms that expect to receive revenues from the sale of software products or services.

A remarkable feature of the industry as it has matured is its lack of concentration-a facet that has considerable implications for the competitive structure of the industry and its openness to innovation.38 Although press reports (and much of the academic writing) are preoccupied with concerns about the dominance of Microsoft, the industry is populated with an unusually large number of significant commercial players. Census Bureau statistics report more than 40,000 firms in the industry as of 2000.39 Nearly 500 firms in the industry had $1 million or more in sales in 2003, even after contractions in the industry at the turn of the millennium.40 In 2002, 156 firms received their first round of venture capital financing, receiving a combined total of $691 million-an average of more than $4 million for each firm-during a markedly down year.41 Moreover, despite the existence of some prominent firms, the number of large firms is very small; there are only three software firms in the current Fortune 500.42 Indeed, the top ten firms in revenue had less than thirty percent of the revenues of the industry as a whole.43

Another important aspect of the software industry is the ebb and flow of IP software protection for much of the industry's history. Although the form of the protection has changed over time, I share Rob Merges's view that "[t]he United States has traditionally embraced strong protection for computer software."44 In the early days, it was generally believed that it was "trivially easy to replicate" the software program of a competitor.45 When initial efforts by major industry players to obtain patents on their products were unsuccessful,46 firms and Congress turned to copyright as an alternative.47 The Copyright Office formally decided to permit registration of programs in the mid-1960s.48 Initially, this was a promising arrangement based on analogizing literary expression to the lines of code that compose a software program49 and until the late 1980s, copyright provided relatively strong protection for software.50

Over the years, however, as courts became familiar with software cases, they narrowed copyright so that it ceased to provide robust protection.51 The problem that courts increasingly confronted was that "there is nothing in the statute nor in the legislative history to indicate that Congress intended for copyright to protect the results (that is, behavior) brought about by the execution of program instructions."52 Thus, in Computer Associates v. Altai in 1992, the second Circuit adopted a "hard-look" framework that made it difficult to obtain copyright protection for the broader structural features of programs.53 The court limited protection to specific pieces of the program that constituted "expression."54 Two years later, the Ninth Circuit refused to protect Apple's graphical user interface from appropriation by Microsoft.55 The faade of pervasive copyright protection came crashing to a definitive ruin with the celebrated decision of the First Circuit three years later in Lotus Development Corp. v. Borland International, Inc.56

Yet long before those decisions limited the overarching importance of copyright protection, major firms in the industry had begun to turn to patent protection. Direct protection of software patents was difficult in the wake of the Supreme Court's 1972 decision in Gottschalk v. Benson.57 Still, several of my interviews suggest that software patents were easy to obtain. Because much of the software through the 1980s was produced by hardware firms, patents easily could be obtained on an object (a microprocessor) programmed to accomplish the relevant function.58 To be sure, that artifice was not effective in the 1990s when pure software firms like Microsoft started to play major roles because those firms could appropriate the innovation of those patents in a software program without infringement. By that time, however, patent doctrine had changed so that patent protection was available for software, at least to those firms that were sufficiently familiar with the process to exploit it.59

In sum, despite the contrary mythology of a golden age of IP freedom, it is not clear that there was any time when software was not protected by IP.60 When copyright protection seemed likely to provide adequate protection, many who were active in the industry thought that patent protection would be counterproductive.61 Nevertheless, as it became increasingly clear that copyright protection was inadequate, supporters of patent protection in the industry gained force, and many of the leading firms now have large numbers of patents.62 This Article considers the role patents have come to play as the industry has matured.

III. Do Patents Induce Commercialization in the Software Industry?

As suggested above, it is difficult to develop any concrete understanding about the effects patents might have on innovation in a particular industry. Even if we assume that all increases in innovative activity are positive,63 it is difficult to separate economic effects related to the patent protection from the economic effects of the innovation for which the patent is granted.64 Recognizing that difficulty, this Part of the Article examines the potential positive effects of patents in the software industry.65

To understand the effect of patents on software development, I focus on small firms, which typically are venture backed. Several related considerations support that choice. First, and most obviously, many of the most important innovations in the software industry come from relatively small firms.66 Yet the scholarship to date focuses exclusively on large publicly traded firms.67 Second, the complex capital arrangements of public firms make it harder to analyze the relation between patent portfolios and the flow of capital into and out of those firms. Thus, a study of the simpler arrangements of venture-backed firms is preferable. The final point relates to the nature of qualitative interviews. Generally, it is easier to obtain reliable interviews from smaller firms and their venture capital investors than from large firms. It is less common at a large firm to find a person with complete hands-on responsibility for both the financing arrangements and the policies with respect to IP development and protection. Executives at larger firms also are much more likely to articulate views constrained by the legal positions underlying the broader interests of the firm and are thus relatively unlikely to engage in the kinds of wide-ranging conversations likely to provide useful information to the quasi-anthropological research that I conduct. Moreover, venture capital investors have highly diversified experiences relevant to my inquiries. Many venture capital investors have experienced good and bad returns on literally dozens of investments and have considerable insight into what makes those investments good and bad, and many of these investors have had previous careers as entrepreneurs themselves, giving them a more complete perspective.

This Part proceeds in three steps. First, I discuss the goals of the venture capital investors, which relate to the likelihood that a portfolio firm can differentiate itself from its competitors in a reliable way. Second, I analyze the ability of patents to provide that reliable differentiation. Although the discussion evinces pessimism about patents as a mechanism for appropriating the value of innovations in the software industry, it also emphasizes a shift in the efficacy of patents as firms pass through the early stages of growth. To that end, I discuss a number of reasons why patents are not likely to allow such appropriation in the earliest stages of a software firm, before it has revenues or begins shipping a product. Then, I show how once the firm moves beyond infancy-to a stage with revenues or a productpatents can have a variety of beneficial effects. That section provides a framework for relating the evidence drawn from my interviews to the existing analytical literature, teasing out of that pattern a set of direct and indirect positive effects that patents can have in various circumstances.

A. Venture Capitalists and Sustainable Differentiation

The development of software is expensive and time consuming. Thus, it is not common for a successful product to be developed by an individual developer working in his spare time. Rather, most commercial software products are the result of years of effort. That effort, in turn, inevitably requires the expenditure of considerable monetary resources. Of course, young firms can-and normally are expected to-go a considerable way toward developing their concept without using the funds of third parties. At some point, however, the individuals within young firms will exhaust their own resources and the readily available resources of friends and family members.68

In most cases, the firm then will turn to institutional investment. One of the most prominent and common69 sources of that investment is a venture capitalist.70 Venture capital firms are intermediaries that raise funds from institutional investors (corporate pension plans and the like) and invest those funds in startup companies in technology areas.71 There is a vast literature on the structure of the venture capital industry, including detailed studies of many aspects of the contract structures that firms use in dealing with their investors and with the portfolio companies in which the firms invest.72 For the purposes of this project, however, the structure of the venture capitalist is relatively unimportant. What is important is to understand as precisely as possible what characteristics of a portfolio that lead venture capitalists to invest. Although there is little quantitative empirical work on that question,73 the most obvious role that IP protection might play in that process is that the monopoly that IP protection grants on the exploitation of a covered technology could cause investment to flow into the firm that has created the technology. The monopoly supports such a flow of investment-at least in theory-by creating market power that allows the firm to earn supranormal profits by exploiting the technology in question.

The first point to understand about startup companies is that the uniqueness of a firm's product is not likely to be one of the primary issues a potential venture capital investor will analyze in deciding whether to invest in the firm.74 The investor is likely to examine a large number of plans and invest in only a small number of them-perhaps 6 out of every 1,000.75 In deciding whether to invest, the investor is likely to start by focusing on issues that validate the firm's competency to execute its concept successfully.76 For example, investors will be interested in such things as experience in the relevant market and the skills of the management team.77 One remarked: "Every company of mine that has failed has been [due to] mismanagement of executives, not technical failure."78 Similarly, even before investors consider whether a firm can protect a market leader position, they will want to know whether the product is one that customers need so desperately that the firm could earn significant revenues from sales of the product.79

Still, for firms that have a credible product idea and the expertise to implement it, venture capitalists plainly accept the idea that their goal is to identify firms that will have sufficient market power to earn extraordinary profits.80 IP protection is important only indirectly, as a tool that might provide that market power. The key is "sustainable differentiation"81: something special about the particular firm that will enable it to do something that its competitors will not be able to do for the immediate future. The interviews reflect more picturesque terminology-referring to "secret sauce"82 or "magic dust."83 But it is clear that the key to a desirable investment opportunity is in the expectation of market power, and all other attributes of the company are indirect predictors of that ultimate goal.84

For example, investors commonly referred to lead time or first-mover advantages.85 The premise is that a portfolio company that truly is the first to provide a sophisticated and functional response to an important problem can expect to earn a supranormal return for years to come. Interestingly enough, that expectation rested on the perception that a firm can maintain a lead on its rivals as long as it keeps improving its technology as quickly as its competitors.86 I rarely if ever heard investors (as opposed to developers) voice an expectation that portfolio firms obtain and keep a strong market position through "lock-in" or "bandwagon" effects.87

That is not to say that IP protection is unimportant. It is clear, however, that different investors have different views about it.88 Some feel that intellectual property always is important and claim that they never invest without strong patentable technology.89 Even those investors, however, go on to say that they are not as interested in the IP protection as they are in technology that is sufficiently cutting edge to warrant protection.90 Others, however, particularly those that emphasize early-stage companies, say IP protection is unimportant for software investments.91 Still others take a smiddle position, holding that IP protection matters some, but not all, of the time.92 Most of those who addressed the subject recognized differing perspectives on the point and argued that those with the other perspectives are misguided.93 The most likely explanation is that investors are simply implementing different investment models based on their particular expertise.

B. Patents and Sustainable Differentiation

If understanding what venture capitalists want answers the first question, the second question is whether they believe that patents can provide it. This subpart looks quite closely at the various stages of relatively small private firms and suggests that investors and developers discern a balance of interests that shifts as firms grow from the earliest stage, where patents are not often helpful, through intermediate stages to the terminal stage of venture-backed firms (just before an acquisition or IPO), where patents are almost universally viewed as useful.

1. The Basic Problem: Patents and Approprlability.-A basic problem for software firms at all stages is the sense that even with a patent it often is difficult for a firm to "appropriate" the value of its invention.94 Specifically, my interview subjects agreed that competitors usually could, without infringing a patent, implement most of the aspects of a patented software product.95 One reason for that problem is the multifarious nature of software innovation, which permits many solutions to any particular problem. Another contributing factor is the poor match between patents and products in the industry: it is difficult to patent an entire product in the software industry because any particular product is likely to include dozens if not hundreds of separate technological ideas. Thus, it may take a number of novel ideas-and patents-to build a defensible barrier96 around a product.97 Another problem is that technology tends to develop so rapidly that by the time a patent is issued-and the formal right to exclusivity commences-the technology may be obsolete.98 Litigation at that point will involve efforts by the patent owner to challenge technology of a subsequent generation to which application of the patent may be less clear.99 Yet another problem emphasized in my interviews is the problem of detectability-the difficulty of being sure that a competing product infringes a patent.100

The problem of appropriating the value of software through patents is not universal. As I discuss below, some patentees manage to obtain patents of sufficient breadth to cover all possible solutions to an important problem. More broadly, some knowledgeable observers attribute the difficulty of appropriation not to the nature of software technology, but to the infancy of the industry. Because the industry is developing so rapidly, some argue, the nature of technology and even of technological developments is so poorly understood that firms do not recognize the value they could appropriate from patents if they pursued them in an informed way.101 From this perspective, the relatively limited appropriability provided by software patents should be compared not to the relatively high appropriability of hardware patents,102 but to the even more limited appropriability that software patents provided a decade ago. Those observers expect that in a matter of decades software technology will be as effectively subject to patent protection as the related hardware technology is at this time.

For now, despite those qualifications, the relevant point is that for most firms, most of the time, there is little prospect that the patents they obtain will provide market power that they can use to exclude competitors. That point is underscored by the relative infrequency with which venture-backed software startups obtain patents. For comparative purposes, consider that only about twenty percent of venture-backed software companies have a patent within five years of their first financing (with each of those firms holding, on average, about two patents), while more than half of biotech startups have patents by that time (with each of those firms averaging about seven patents).103 Thus, even the tenuous exclusivity such patents could provide is usually not a factor for early-stage software startups because so few of them in fact have patents.

The point is further underscored by the statistical relation between the success of venture-backed software startups and their patenting practices. The general theme of the preceding pages is that there are many factors that play into the ability of a startup firm to obtain funding and success and that intellectual property has a low place on the list of factors, if it appears on that list at all. The data I have collected with Tom Sager supports that viewindicating that patenting practices have at best a minuscule ability to predict the success of a venture-backed software startup.104

2. Patents and Prerevenue Startups.-In addition to the problem that patents often are not an optimal mechanism for appropriating the value of software innovation, a number of considerations make it particularly difficult for early-stage companies to employ patents effectively. The key points here are the limited efficacy of litigation for those firms, the constraints on resources that make it infeasible to focus on patenting, and the limited value to prerevenue firms of excluding competitors.

a. The Perils of Small-Firm Litigation.-On the first point, even if an early-stage company had a patent, it is unlikely that it would have resources available to enforce the patent through litigation against a competitor.105 That is particularly true when the competitor is a large firm. One problem is the disparity in litigation resources. One investor emphasized the concern that a large defendant would "rain lawyers on your head and tie you up in court for the next ten years."106 A somewhat different concern is the likelihood that the large firm might have a patent that the small firm infringes. If so, the lawsuit might simply alert the large firm to the presence of the small firm.107

A related concern is that firm culture is degraded when a firm must rely on licensing revenues instead of developing its own product.108 Interestingly enough, that sentiment was expressed even at firms that rely heavily on licensing revenues. Those firms encouraged efforts to maintain a product-centered culture that emphasizes production of the firm's own products.109 One executive explained that a cultural risk arises when a company pursues patent litigation and licensing, explaining that such a strategy unfavorably affects the company's needs. "You don't need sales people; you need attorneys. You don't need solutions architects; you need accountants. So you wind up losing the very people who are, who were, and who continue to be constructive . . . and innovative and help you build things and would give us a continuing competitive advantage."110

Indeed, the one interviewee whose firm had a major licensing program related that the entire program was entrusted to third-party professionals so that it would not interfere with the focus of the onsite software engineers.111

b. Diversion of Focus.-Similarly, many investors and developers emphasized that attention to patents can be damaging to a startup because it has the potential to divert limited time and resources from what is likely to be a highly time-pressured effort to develop a product and convince customers and investors of its worth before the firm runs out of capital resources.112 One investor explained: "[We] typically find that the companies that focus on just patents don't have the right view of what is important, and they really are therefore not successful in business. And they're usually not around to prosecute their patents."113 Developers understand the point well. As one said: "Every dollar we spend on [patenting] is a dollar we can't spend on a software engineer."114 Another, with a patent-leaning background from his days at IBM, commented: "Patentability is something we will pursue, but let's get the product out first.115

Thus, a young company is presented with a challenging task. If the nature of the firm's innovation is such that IP is ever likely to be important, it must spend sufficient resources on the protection and development of intellectual property from the earliest days of the company-as an investment in the possibility that the firm might grow to the point at which IP is useful.116 The firm that fails to protect its IP at the earliest stage is like a desperate ship at sea that empties its drinking water in the hope of evading a faster pursuer: it might survive for the time being, but it may have sown the seeds of its inevitable failure if it survives to a later stage.117 On the other hand, it must not spend so much that the company fails before it is able to recoup its investment.

Firms have developed a number of strategies for dealing with that problem. Some involve using half measures to protect the IP, such as filing provisional applications118 or omitting standard practices related to documentation of the work of engineers.119 Those practices do not directly abandon the IP, but they may make it more costly and difficult to protect it in the future. The bottom line is that even for companies that have begun to earn substantial revenues it often does not seem appropriate to devote the resources necessary to ensure that all of the firm's innovations are patented.120 Others-it must be said that executives with prior experience at large IP-sensitive firms like IBM121 or Bell Labs populate this category-seem to relish the discipline of making sure that the IP is pinned down no matter how difficult it may seem to find the time and resources to do so.122 The difficulty of this strategic choice, coupled with the difficulty in accurately predicting the future prospects for their products and their IP, is a problem about which startup software executives worry constantly.123

Investors, of course, are aware of this problem. Their approach typically does not extend to forcing (or even urging) their portfolio companies to seek patent protection.124 However, they do go to considerable lengths to evaluate the IP that potential portfolio companies have. In a typical process, the venture capitalist (VC) knows most of the reputable patent attorneys in the local community. If one of those attorneys has filed a patent for a potential portfolio company, the VC discusses the patent with that lawyer. If an attorney with whom the VC did not have a preexisting relationship filed the patent application, the VC has the patent studied by an attorney in whom he has confidence. In the context of the interviews, it was clear that the intent of this examination is not purely technical-whether this is a patent likely to be granted-but also a broader exercise to understand what type of market power the patent might (or might not) provide.125 Indeed, the current practice-oriented literature characterizes examination of market power as central to careful due diligence practices.126 The practice at first seems to be in tension with the thesis of this section-that patents have little value for the earliest-stage startups. In fact, however, it leads into the point of the next section: the firms that survive their earliest days may reap substantial value from patents.

c. The Limited Benefits of Exclusion.-One final element of the patenting calculus for small firms may seem obvious, but is so important to the overall framework as to warrant brief notice. Because those firms do not yet have a product, they have no opportunity for revenues. Thus, the benefits they reap from excluding competitors are minimal at best. Only if they survive to a later stage-in which they can hope to profit from their own exploitation of that product-will they be able to reap any substantial value from the exclusion of competitors.

3. The Increasing Value of Patents for Later-Stage Startups.-When firms mature to the point of having revenues, the systematic difficulties that plague the efforts of prerevenue startups to obtain and exploit patents dissipate. That does not mean, however, that patents suddenly become a philosopher's stone that will turn their creative endeavors into IPOs. Rather, the underlying problem of appropriability continues to plague most efforts to use patents directly to exclude competitors. Still, my interviews suggest a series of benefits that patents might provide for later-stage software startups. This section describes those benefits by reference to the bodies of existing literature that have offered them as theoretical possibilities.

a. Direct Effects: Protecting a Space for Innovation.-The most important point concerns the direct ability of a software patent to carve out for the firm a space in which it can innovate without competition. I explain above that my interview subjects often complain about how difficult it is to use patents to exclude competitors. Although there is some truth in this complaint, it is an overgeneralization, at least once a firm reaches the stage at which it has designed a product that it can market to customers.

First, it is clear that some firms in the industry obtain a substantial amount of revenues by licensing the use of their patents to competitors that need to use the patented technology in their own products. Indeed, even in my limited sample, three small Austin companies-Applied Science Fiction,127 Bluecurrent,128 and Forgent129-have obtained substantial revenues from patent licenses. I do not believe that industry wide statistics quantify the size of that market, but it plainly is substantial. Those transactions-and others like them-demonstrate that some software patents are sufficiently robust to allow their holders to appropriate substantial value from the underlying inventions. Licensing transactions are noteworthy given the difficulties small firms face in enforcing patents against large firms. As discussed above,130 the small firm with a revenue producing product must be quite confident in the value of its technology before it wisely can cross swords with a company like IBM.

More generally, it seems clear that the received wisdom that patents are not useful to appropriate software-related inventions is overstated. Two separate points are important. The first is the distinction between the relative rarity of observed offensive patent use-for out-licensing or litigation-and the use of patents to exclude competitors. The relative rarity of offensive use of patents does not prove that the patents are insufficiently robust to exclude competitors. As discussed in the previous section, there are many reasons why a firm might want to wait until late in its development to advertise the nature of its technology and its proprietary claims to that technology.

A firm can refrain from offensive use of its patents and still derive important value from the patents as an exclusion device. Contrary to the perception that patents tilt the playing field in favor of large incumbent firms to the disadvantage of small firms,131 patents in this context afford a unique opportunity to the small startup.132 The patent system grants the small firm an automatic stay of competitive activity that remains in force long enough for the firm to attempt to develop its technology. For large firms, the marginal increase in appropriability that comes from patents may have little benefit: IBM could compete quite successfully against smaller firms even if it did not have patents protecting its product from copycat competitors.133 For the smaller firm, however, the ability of the implicit threat of patent litigation to prevent incumbents like IBM and Microsoft from taking its technology can be the difference between life and death. As one executive put it: "What's protected me from other people ripping [off our product] has been the specter of patent infringement."134

It is instructive to think of the offensive uses of software patents reported in the press. Among the most famous incidents are the successful attempts by small firms-Stac in the mid-1990s135 and Eolas in 2003(136)-to force alterations in Microsoft products that arguably infringed patents held by those firms, and the similar attempt by InterTrust to assert rights to digital rights management technology that was important to several Microsoft products.137 The profits from suing other small firms seem to be so much smaller that it is easy to see why that kind of "horizontal" litigation is apparently less common. I note that the story depicted in the press is consistent with empirical work suggesting that patents held by small firms are more likely to be litigated than patents held by large firms.138 This also finds strong support in Bronwyn Hall's recent work suggesting that patent rights in complex product industries are more valuable for younger firms than they are for older incumbent firms.139

Second, the ability of a patent to appropriate the value of an innovation is often said to vary along several dimensions. One of the most common is the nature of the particular innovation. Thus, it is often thought futile to rely on a patent if the innovation lies in a method of writing software code.140 At the other end of the continuum, patents that protect an ultimate functionality that the software provides or an algorithm necessary to provide that functionality are more likely to be important in excluding potential competitors.141 Interestingly, that distinction seems to undermine the conventional wisdom dividing patents along another dimension, in which "process" patents tend to be less valuable than "product" patents.142 In the software industry, a patent on the product tends to have relatively little value because of the ease of designing a distinct product. A patent on the process that the product implements is much more likely to be valuable, if only because it often is possible for the claimed process to be defined broadly enough to include all practicable methods of competition.143

A sectoral variation in patenting appears to be related to those dimensions of analysis. As the empirical data presented in a related paper I authored with Tom Sager demonstrates, there is a strong variation in the rate of patenting by venture-backed firms in different sectors of the software industry.144 For example, the average number of patents in the dataset was about 0.6 patents per firm.145 Several sectors, however, had markedly higher rates, including graphics and digital imaging, expert systems and natural language, multimedia, and security.146 At the same time, some relatively important sectors had unusually low rates of patenting, including email and internet software, applications software, and financial software.147 That variation is interesting because it can be discerned in a quantitative way even though patents are thought to be less valuable for software than they are for hardware, and even though patents are much less common in the software industry than in some other industries.

This is not a topic on which my interviews produced constructive results. I focused several of my interviews on firms known to me to have commercially valuable patents. Notwithstanding the discussion and empirical evidence above, executives of those firms that expressed a view as to why their patents were valuable generally believed it had nothing to do with the nature of the innovation. Rather, they attributed it to the firm's ability to obtain a patent that staked out a sufficiently large field to cover all plausible variations on the relevant technology.148 Still, it seems likely that variation is related at least in part to the nature of innovation in different sectors, with higher rates of patenting associated with types of innovation more susceptible to appropriation by patent. However strongly my interview subjects rejected such a distinction, the data powerfully suggest that further inquiry is warranted. For now, my intuition is that the patterns of patenting rates most likely relate to the distinction between product firms (for which patents would be more useful) and service firms (for which patents would be less useful).149

Taken together, those two points portray a world in which small firms struggle to innovate, facing the pervasive concern that a competitor might appropriate any useful invention at any time. Given the difficulties those firms face in sustaining differentiation, the possibility that patents can provide shelter for some firms is important. The extent of shelter may be difficult to predict because it depends primarily on the breadth of market protection a patent has by the time it is issued.150 Furthermore, the frequency of shelter is open to doubt; it plainly was not relevant to most of the firms that I interviewed. The interviews that I discuss above, however, do suggest that shelter is real in the place where its effect would be most important-in the minds of firms doing the innovation. It certainly would be valuable to know more about the frequency with which small firms obtain patents of sufficient strength to use in this way. However, even without quantitative information, it is difficult to believe that this is not a major part of what makes patents and their breadth an item of interest to investors.

b. Indirect Effects.-The most intriguing information from my interviews relates to patent benefits that do not involve direct exploitation of the patents. Those benefits fall into two classes: facilitation of a litigation-free zone through a pattern of cross-licensing; and a set of what I call "information" effects-beneficial effects that patents have on information related to the firm's technology.

(1) Facilitating a Licensing Equilibrium.-As suggested above, many in the industry completely deny any substantial use of patents to exclude competitors. Rather, the most prominent explanation for patents in the interviews was that patents are useful as "barter" in cross-licensing agreements that the firm enters if it reaches a sufficiently mature stage to be a significant player in the industry.151 Those with whom I discussed the subject articulated a common vision of the way that firms gain access to protected technology in the industry: they obtain a cross-license from existing industry members that have patents related to their technology. To the extent that a growing firm has patents on its own technology, it can reduce the cost of licensing technology from existing market players by providing that technology as part of a cross-license agreement.152 The likelihood that a firm will be asked at some point to enter such an agreement seems quite high. Interview subjects acknowledged, for example, that their products might infringe a patent in IBM's large portfolio of software patents.153 Yet, a patent to offset IBM's potential claim is of little value until IBM demands royalties, and IBM usually does not ask for royalties until a firm is earning sufficient revenues to justify the inquiry.154

As a matter of policy, it is difficult to know how to evaluate that arrangement. One perspective is that it reflects a classic instance in which sophisticated parties with repeat dealings can reach a state of equilibrium, operating in the shadow of the law with relatively little active conflict.155 The premise is that firms with large portfolios will refrain from "mutually assured destruction" litigation-a situation produced by two large-portfolio firms in conflict. The destructive capacity of those portfolios is enhanced by the nature of technology in the industry, which could involve dozens of patentable innovations in any single product that a large firm might bring to market. From that perspective, it is natural to believe that litigation occurs most often when equilibrium fails, when someone other than an active developer holds the patents,156 or when the developer fails and loses all incentive to cooperate.

More broadly, however, it is difficult to see how equilibrium can be regarded as a positive benefit attributable to patents. The only benefit that cross-licensing agreements provide is freedom from patent litigation. The cross-license agreements in question provide only freedom of action; they do not involve the disclosure of technology or transfer of any knowledge beyond material on the face of existing patents.157

I can postulate ways in which cross-licensing might provide net benefits. For example, patents might provide an effective way to evaluate the value of each firm's technology and thereby determine the amount and direction of payment that is appropriate for each cross-license. This might be particularly important in an industry, like the software industry, in which there are a large number of players with widely varying patent portfolios. By contrast, in an industry with a small number of relatively equal participants, a straight patent pool (without pair-by-pair determinations of value) would make more sense.

Those types of arguments, however, strike me as ad hoc justifications for a practice that at best simply reduces the transaction costs firms face in gaining access to patent-protected technology. We know little about the terms of license agreements in the industry, or even the frequency and extent to which license agreements involve the payment of license fees. We do know, however, that it cannot be costless to acquire the patents that firms use to enhance their licensing position. If those costs are incurred solely to minimize the costs of patents that other firms have, and if patents provide no other benefit, then it seems plain that patents are not providing any net benefit to the industry. Thus, however pervasively they were discussed in my interviews, and however important they are to understanding actual patenting practices, I do not give great weight to the benefits of cross-licensing as a policy justification for patents in the industry.

(2) Information Effects.-The last set of effects relates to information generated through a firm's participation in the patent system. Moving along the course of a firm's development, these effects fall into three classes: (A) the ability of patents to facilitate the firm's efforts to codify tacit knowledge, (B) the firm's subsequent ability to signal the discipline and technical expertise that allowed it to codify that knowledge, and (C) the use of the patent as a signal of the underlying technology.158

Facilitating the Codification of Tacit Knowledge. One of the most intriguing benefits of patents relates to Ashish Arora's recent writings on innovation. He argues that a key problem in transferring knowledge between firms is the ability to convert tacit knowledge-which is difficult to verify or transfer-into codified knowledge, which can be readily evaluated and transferred.159 Because a patent by definition-at least if it satisfies the statutory criteria-includes the knowledge necessary to enable a person having ordinary skill in the relevant art to replicate the invention, the existence of a patent is strong evidence both that there is substantial knowledge of some kind and that the knowledge is not so bound up with the abilities of particular individuals as to be immovable.

To the extent that a patent facilitates that process, it can provide real benefits to the firm: codification of knowledge enhances its transferability and thus its value.160 This analysis has found its way recently into the law review literature in Paul Heald's work on transaction costs and patents.161 More generally, the idea resonates with the notion that productive assets can have no value until they have been brought into a documentary system in which they readily and reliably can be transferred from person to person.162 Thus, although my interview subjects do not discuss "codification" of knowledge, they do emphasize the importance that patents play in the acquisition of a startup firm.163 It seems plain to me that there is more work to be done in understanding how patents facilitate the transfer of knowledge-and whether alternate systems could serve the same purpose without the costs of exclusivity. I discuss those questions briefly below,164 but it is clear that more research is necessary to formulate any definitive views on the topic.

Signaling Discipline and Expertise. Once knowledge has been codified in a patent, the existence of the patent itself can send a signal regarding the skills necessary to obtain it, primarily engineering discipline and market understanding. The premise is that firms that obtain patents tend to be more careful in their engineering work and have a better understanding of what is special about their products than competitors that do not have patents. One serial startup developer explained:

[I]n my experience, all a software patent buys you is the fact that you are disciplined in your engineering approach and that it is reflected in your ability to execute technically. Not that it is a means of protection for the investors to believe that you're gonna be the only person that's gonna be able to solve this particular problem.165

Those who articulate this line of reasoning generally view the signal as a true one-a plausible indicator of valuable information about the firm that otherwise would be difficult to discern.166 Notice, of course, that this use of patents says nothing about the uniqueness of the technology or the firm's ability to exclude competitors. Rather, it reflects something positive about the ability of the management team to focus and execute. That does not mean, however, that the signal is not taken seriously. As discussed above, many investors think that inadequate market analysis and execution are among the most common reasons startup companies fail.167

Signaling Technology. When a firm reaches the stage at which it considers acquisition or a public offering, its patents may send a more direct signal regarding the underlying technology. As the discussion below explains in more detail, larger firms are likely to value patents for reasons quite different from those that motivate small firms; larger firms value patents because they facilitate freedom of action by helping the company avoid claims of infringement.168 Thus, investors consider the existence of a patent to play a key role in influencing the "build-or-buy" decision of a larger company. The hope is that the potential cost of patent infringement will make it cheaper for the larger company to purchase the portfolio company rather than build the technology in-house.169 Even in these situations, however, it is clear that the sophisticated acquirer focuses on the business model the company has adopted and whether that model makes sense apart from the IP that might protect the underlying technology.170 For example, consider the following comments about Google (a firm that, incidentally, has two patents):

Do you think the big asset for Google is patents? No, it's a business model that's working and making money. Do you think patents are something they're not being stupid about[?] They're filing patents and being careful on the off fear that Microsoft might sue them. Do you think they've built into their S-1 or their business plan that they plan on using their patent portfolio directly? No way.171

That same explanation can be seen in a much more negative way if it is thought that patents will not ultimately bring value to the balance sheet of an acquiring firm. This view of the role of patents in acquisition characterizes them as valuable for "marketing,"172 convincing investors in public markets that the company's technology is valuable.173 The idea is that sophisticated investors at the early stage can evaluate the "true" value of the technology based on a careful analysis of such factors as the company's product, the customers' needs for that product, and the personnel the company employs to execute its business plan. Thus, the patent has only secondary significance to those investors. Customers or later-stage investors, by contrast, are said to be less willing to undertake such careful evaluations.174 Thus, the argument goes, they tend to rely (less thoughtfully) on the mere existence of patents in the company's portfolio.175 That argument is made particularly with respect to protecting the downside in the event of a company's failure.176 Interestingly, developers often present a similar argument about venture capitalists, arguing that companies obtain patents that have no real value to them, in part because they will look good to venture capitalists.177 The truth of that view in any particular context is of course difficult to assess.

4. Patents and Large Firms.-Although the bulk of my interviewing base is small venture-backed startups, the interviews and publicly available information do provide enough information to give a good idea of the role that patents play in large firms. Because much of the information is plain from the discussion above, I discuss that topic briefly solely to complete the picture.

First, as discussed above, large firms gain relatively little through litigation or the exclusion of competitors from patented spaces because large firms often can compete successfully with small firms without excluding the competitor. The saying that "nobody ever got fired for buying IBM [or Microsoft]"178 is not baseless, and in a contest between IBM and a small startup, both with equivalent products, IBM (or Microsoft) often will prevail. In contests among large firms, litigation is rare because of cross-licensing of portfolios.179

Second, patents provide considerable benefits to large firms by enabling them to participate in cross-licensing agreements that give them the freedom of action to design and deploy products as they wish, without regard to the IP portfolios of competitors. It may be, as I argue above, that large firms that use their portfolios solely for that purpose would be better off without the costs of developing and maintaining those portfolios, but in the existing milieu, each firm has a strong incentive to collect patents for that purpose.180

Third, many-though certainly not all-large firms obtain substantial revenues from directly exploiting their patent portfolios. IBM, for example, earns literally billions of dollars each year exploiting its patent portfolio; a significant share of the revenue comes from its software patents.181 Thus, although different firms have different strategies, the potential for large firms to earn substantial revenues from direct exploitation of patents does exist. Information about the amount of those revenues would be valuable in assessing the net effect of patents, but that data is not readily available.

C. Summary

Much remains unclear about the ability of patents to induce commercialization in the software industry. For example, although the ability of small firms to use patents to protect themselves is important, it is difficult to tell from the available data how widespread that benefit is. If it is widespread, then it may contribute to the fragmented and highly competitive structure of most sectors of the software industry by providing startups a sufficient time to commercialize their products without competition.182 Similarly, it is difficult to disentangle the local effects that motivate firms to obtain patents-as cross-licensing collateral, for example-from the direct effects and the information effects that might provide a justification for the system as a whole. Further, any understanding of those effects must account for the differentiation of their weight as firms progress through the development cycle. Finally, even if those effects elicit funds for the firms that have patents,183 we cannot be sure that they increase total investment. It is also possible that they simply alter the direction of investment towards patent-protected investments without altering the total amount of investment.184 Another possibility is that larger firms against whom small firms' patents are enforced would have invented the same products almost as quickly. If so, then small firms' patents may be imposing costs on the industry that exceed the value of the innovation they stimulate.

To clarify the overall import of the discussion, Table One summarizes these effects. As that table shows, my research indicates a basic tradeoff between several effects that are not readily quantifiable. The major burden I discern is the acquisition and use of patents for cross-licensing purposes, which seems to be a deadweight loss for the industry. On the other side are three benefits. The first two largely accrue to later-stage startups: the benefits patents provide in sheltering those firms and the information benefits patents deliver to later-stage startups. Both of those are difficult to quantify: the first because it depends on an understanding of whether and to what extent the startup firms invent technology sooner than large firms would have invented it without a patent system, and the second because it is almost inherently subjective. The third benefit is the potential licensing revenues that accrue most commonly to IBM and other large firms. That might not be difficult to quantify but, as discussed above, it is not something about which quantitative data is easy to locate.

IV. Potential Costs of Patents in the Software Industry

A. Patent Thickets

The literature criticizing software patents for the most part focuses on a single potential problem: the costs that patents impose when they exclude third parties from development. Most famously, Larry Lessig argues that the proliferation of software patents has created an "anticommons"-a term Lessig draws from Michael Heller's work with Becky Eisenberg185-or a patent "thicket" (Carl Shapiro's term186). Lessig supports this claim largely by reference to the work of Jim Bessen (by himself and with other co-authors).187 Specifically, the concern is that there are so many overlapping patents in the industry that potential innovators cannot readily obtain the approvals necessary to conduct their research.188 The thesis gains some credibility from the nature of software innovation because, as explained above, a software product might involve dozens of innovations and several firms might hold patents on one or more aspects of a firm's technology.

I address first the claim that proliferation of patents has stifled R&D spending. I then address the anecdotal claim that a thicket of patent claims deters small firms from pursuing promising innovations.

1. R&D Spending.-The two papers advancing the claim that patents have stifled R&D spending are Bessen's unpublished papers with Maskin, Sequential Innovation, Patents, and Imitation,189 and with Hunt, An Empirical Look at Software Patents.190 Bessen and Maskin compare a dataset of software patents-defined by patent classes-held by large publicly traded firms to R&D expenditures of those firms.191 They find a correlation between increases in software patenting and declining R&D expenditures.192 The evidence, however, terminates in 1995-a very early stage of the software industry-and includes only leading software patentees, rather than leading firms in the industry.193 Thus, their dataset includes companies like Ford, General Electric, and Japanese firms like Mitsubishi and Matsushita, but excludes companies like Microsoft and Oracle, which were not large software patentees at that time.194 Furthermore, given the limited importance of software development to the firms they examine, it seems important to determine the amount of R&D spending those firms allocate to software development; Bessen and Maskin are unable to do so with the Compustat data on which they rely. More fundamentally, the basic object of the Bessen and Maskin paper is not to analyze the software industry specifically, but to analyze firms that obtain software patents. As they explain in the paper, the software industry is not the principal industry that obtains those patents-so their paper does not specifically address the relation between patents and R&D in that industry.

Bessen's more recent paper with Robert Hunt collects a dataset of patents based on key words that are reasonably likely to reflect software innovation.195 The paper plainly reflects the most serious effort to date to collect a dataset of software patents; the authors have collected all patents including the relevant key words from 1976 to 1999-about 131,000 patents.196 Most of the patents in their dataset are assigned to large manufacturers.197 Software publishers own only five percent of the patents in their dataset. Their dataset also excludes most private firms.198 Thus, they do not analyze the firms that generate about one-third of the successful patent applications they collected from the PTO database.199 The primary finding of Bessen and Hunt of relevance here is that the propensity to patent has increased in the software industry since the late 1980s.200 For the reasons discussed in Part II, that finding is not surprising.201

Bessen and Hunt also analyze the relation between R&D spending and patenting, concluding that the data is consistent with the view that patents are a substitute for R&D spending. The implication is that R&D spending would be higher if changes in the law had not made software patents more attractive.202 They conclude that the results are more consistent with strategic patenting than with incentive effects of patents.203 Although the data is intriguing and the analysis considerably more robust than that in the Bessen and Maskin paper, the paper has the same basic focus on software patents rather than the software industry. As Bessen and Hunt emphasize, "software patenting [as they define it] by and large has little to do with the pre-packaged software industry."204 Rather, their emphasis is on the other industries in which many software patents are held.205 Thus, because such a large portion of the dataset lies outside any plausible boundary for the software industry, it is entirely possible that their findings reflect issues of firm governance and industrial structure that have little to do with the software industry.206 Also, as in the earlier paper, the analysis relies exclusively on data about relatively large firms and thus has almost no overlap with the topic of relevance here: whether patents held by large firms stifle investment in software startups. Thus, at most the paper establishes doubts about the role patents play in large public firms. The preceding sections of this Article, of course, articulate a similar degree of skepticism about those benefits.207

Against these studies, which suggest a puzzling relation between software patents and overall R&D outside the software industry, it is easy to examine data that directly describes the current state of R&D in the software industry. That data suggests a different picture, one in which software R&D is impressively robust. For example, Technology Review's, Corporate R&D Scorecards report the annual research and development spending of the world's top 150 technology companies.208 Each company is assigned to one of 12 sectors based on its primary business.209 The scorecard figures are derived from annual reports and U.S. Securities and Exchange Commission filings.210 Data from the Scorecards indicate that R&D spending for the software industry is higher than in similar high-tech industries. For example, R&D spending as a percentage of revenues in the software industry for 2002 was 14.5%.211 By way of comparison, R&D spending in the same period was 6.7% for computer hardware, 7.4% for electrical/electronics, and 8.1% for telecommunications.212 Thus, for the top technology companies, the R&D intensity ratios are high in the software industry in comparison to other industries.213 National Science Foundation data regarding industrial R&D intensity provides a similar picture. That data shows that R&D intensity for firms in the software industry (NAICS code 5112) was 19.3%, 20.0%, 16.8%, and 20.5%, for the years 1997-2000, respectively.214 This is far above the average for all industrial R&D firms of about 3.6%.215 Indeed, according to the National Science Foundation, the software industry for the past four years has had an R&D intensity substantially higher than any industry other than Scientific R&D Services (NAICS code 5417).216 Because software development does not depend heavily on the existence of manufacturing facilities and other fixed assets, those high figures should not be surprising. However, it is hard to credit the argument that R&D spending in the industry is systemically depressed.

As the data shows, software R&D spending tends to be relatively stable over time as a percentage of sales. Indeed, company size seems to be more important in explaining variations in R&D spending within the industry. For example, data from Compustat indicates that median R&D spending for large public companies (over $100 million in sales) in SIC 7372 (prepackaged software) is only 15.9% of net sales, while mid-sized firms (between $30 million and $100 million) spent 22.6%, and small firms (under $30 million) spent 32.8%.217 Those figures have not changed substantially over the last three years.218 The Software 500 provides similar statistics for the 500 largest firms in the industry, including both public and private firms. For 2001, firms with more than $100 million in sales had an average R&D intensity ratio of 12.83%,219 firms with sales between $30 and $100 million had an R&D intensity of 20.49%, and firms with less than $30 million in sales had an R&D intensity of 23.89%. If patents facilitate a fragmented industry structure by sheltering small firms, they may help to support the high level of R&D spending characteristic of those firms.220

The questions that Bessen, Maskin, and Hunt raise are answered best by looking at patenting practices and R&D spending in the software industry. That work, however, is beyond the scope of this Article. For now, perhaps the most that can be said with clarity is the basic point with which I began this section: the patent system is not systematically preventing the initiation of product development. Beyond that, it is plain that the system is functional. In the world that we have-which includes patents-there are literally hundreds of small firms using institutional financing to develop new technologies. The smaller firms are spending relatively more on R&D than the bigger firms. It is as difficult to be sure that all of those firms would exist if there were no patent protection as it is to be sure that there would not be even more firms if there were no patent protection.

2. Stifling Small Firms.-Turning from the empirical evidence about R&D spending to the more fundamental question of industry practice, my interviews and the publicly available information I have located about the industry make it difficult to credit the idea of a "thicket" or "anticommons" in the software industry.221 When raised in my interviews, that thesis universally was rejected.222 The premise of the model is that assets will go unused because of the costs of obtaining the permissions necessary to use them.223 There is of course nothing theoretically impossible about that outcome. The important question, however, is whether this is in fact what has happened in the software industry.

A few objective indicators suggest that the patent thicket thesis has little to contribute to an understanding of the software industry. For one thing, none of the startup firms to which I spoke suggested a practice of doing prior art searches before beginning development of their products.224 As far as I can tell, the only occasion in the software industry in which a startup is likely to experience such costs is when the startup is founded on a specific piece of existing technology spun off from an existing company or from a university laboratory.

For another thing, no investor suggested any concern about the possibility that their portfolio firms might be infringing the IP of others in the industry. That is not because they were sure that the startups were not infringing; it was because they thought it was unlikely to pose a significant difficulty if they were. As discussed above, industry executives do accept one premise of the patent thicket thesis: that software patents are multiplying so rapidly that it is likely that many products that startups develop ultimately will infringe patents held by large existing companies.225 The textbook example is IBM, which apparently holds far more software patents than any other company in the industry. Indeed, as I explained above, several of my interview subjects joked that they thought it likely-without any investigation or particular knowledge-that their product did infringe something in IBM's portfolio.

Yet that poses no significant concern for those firms. It is perhaps an artifact of the particular history of the industry, but IBM has firmly set a course of relatively lenient enforcement of its IP rights.226 The lenience of its practices is attributed to an attitude developed during its long subjection to government antitrust scrutiny, an attitude of wishing to refrain from conduct that would interest federal antitrust regulators in its practices.227 It is now a circumstance long forgotten by many (as IBM is now regarded most prominently as a dominant hardware manufacturer), but there was a time when IBM's dominance in the software market was as complete as any dominance it has had in the hardware market.228 Indeed, the most authoritative history marks the beginning of the commercial software industry as the date when IBM began to sell its proprietary software unbundled from its hardware products.229 If the antitrust litigation tempered IBM's willingness to press its advantages to their fullest, it has limited the rise of a patent thicket in the industry.

IBM's relative lenience also is attributed to the asymmetric risks IBM faces in patent litigation. A finding that IBM's widely distributed products infringe a valid patent is likely to cost IBM much more than a finding of infringement by a small party with a limited customer base involved in litigation with IBM.230

That is not to say that IBM allows people to use its IP freely.231 It is to say, however, that licenses to use its IP are freely available to all legitimate users.232 Indeed, it appears that the principal, if not the only, reason that IBM would be unwilling to grant a license to use its patents would be if the party requesting the license refused to grant IBM parallel access to the party's own IP.233 Thus, IBM has followed a consistent two-pronged strategy: gaining as much access as it can to all IP in the industry (giving IBM the freedom to market and sell as freely as possible),234 and at the same time generating a steady stream of revenues from its now massive software patent portfolio.235 As Rosemarie Ziedonis has shown, this strategy of heavy patenting is common for large firms like IBM in an environment characterized by fractionation of technology.236

The focus on freedom of action is a rational strategy for a large firm like IBM. Considering its advantages in prestige, resources, marketing, and other forms of infrastructure, it is reasonable for IBM to conclude that it can succeed in the marketplace without using the relatively ineffective tool of IP to appropriate the value of its inventions. Thus, the principal relevance of IP to IBM is to ensure that it is able to commercialize whatever products it desires. If the patent portfolio that it uses to ensure that freedom also happens to generate substantial revenues, that is a useful thing, but not nearly so central to the firm's core strategy.

Nor is this strategy unique to IBM. Microsoft, for example, has an impressively large portfolio, but does not appear to enforce it aggressively.237 Its recent adoption of an open licensing policy that resembles IBM's policy238 suggests at least an implicit acknowledgment that IBM has discerned the correct strategy. There is the possibility, of course, that Microsoft's current strategy is motivated as much by its experience with antitrust litigation as IBM's is. Finally, other large firms that I interviewed in related industries suggested that their IP strategies were similar.239

The noted paper by James Bessen and Eric Maskin240 articulates a contrary view, reasoning that sequential innovation in an industry with complementarity of inventions is likely to lead to an anticommons.241 The paper is flawed by its central logical step: it reasons from the wide dispersion of IP rights to the conclusion that IP is not generally available to firms in the industry. The paper does not account for the literature indicating that the effectiveness of licensing depends to a great degree on context.242 Rather, the paper rests on the explicit assumption that firms are not able to reach value-increasing licensing agreements that make technology widely available in the industry.243

To the extent that the literature supports any theoretical conclusions that are independent of historical context, Merges and Nelson argue that a positive outcome is particularly likely in cases-as in the software industry-in which there is not a single pioneering patent or group of patents that gives one firm control.244 The numerous sectors into which the software industry is divided have made it difficult for any single patent or group of patents to control a major part of the whole industry.245

James Bessen's recent paper articulates a more complex model of the same problem. He recognizes the possibility that optimal incentives for research and development can occur when firms develop a culture of "mutual non-aggression."246 He argues, however, that "aggressive" cross-licensing is a distinct pattern that is likely to lead to sub-optimal incentives for innovation in industries in which patent standards are too low, particularly in cases in which mature incumbents populate the industry.247 Whatever the merits of that analysis, there is little reason to think that it is applicable to the software industry. As discussed above, it is clear that the licensing culture in the industry depends to a considerable degree on the practices of the industry's leader. In addition, it is clear that IBM has determined for reasons of its own-influenced to be sure by federal antitrust enforcement-that it should refrain from pressing its patent portfolio aggressively.248 Thus, although it is always possible to imagine that aggressive practices could lead to a suboptimal pace of innovation,249 the historical events that have made IBM and now Microsoft as cautious as they are make it difficult to argue that those patterns have emerged,250 whatever might happen in the future.251

What this means for purposes of this discussion is that innovators know that IBM and Microsoft hold a large number of patents and that they are likely to use those patents to seek some share of revenues from any major new product. There is, of course, nothing wrong with that. IBM does spend billions of dollars each year on research and development related to the software technology on which it receives patents. It is not alone in that practice.252 The fact that it can earn royalties from those patents through freely licensing them to all comers does not suggest a patent thicket. On the contrary, a patent thicket would exist only if industry licensing practices were such that firms in the industry commonly were unable to agree on terms for licenses and thus retreated from the field of innovation. That is not a realistic portrait of the commercial software industry as it now exists.

Another more plausible possibility is that the "tax" created by crosslicensing fees is detrimental to innovation in the industry. From this perspective, it is not that existing patents are so widely distributed as to make it impossible for firms to obtain access to the technology; it is just that the cost of paying for access to that technology lowers the return on investment in the industry to the point that investments in innovation are less than they would be without patents. Part of the difficulty in assessing that possibility is the intractability of determining whether a typical five percent license fee is a sufficient drag on a small firm's profitability to amount to a substantial burden. The problem would be more severe if firms often had to pay multiple licensing fees, but that seems uncommon based on the interviews I conducted.253 In the end, my instinct is that it is not a substantial burden. I am driven particularly by the point that licensing fees normally are paid only on revenues-not on the use of the patented technology-and thus impose no costs on firms that are in a prerevenue development stage.254 I am also driven by the reliance of many of those firms on venture capital investments. Given the large returns on investments that are necessary for venture capitalists to obtain an adequate return on their risky portfolios, it seems unlikely that individual investments are often forgone because of the likelihood that a firm will be sufficiently successful to earn revenues against which IBM will likely seek royalties.

B. Patents and Open Source Development

In my mind, the biggest question about the effectiveness of software patents involves the interrelation between commercial software development-the topic of this Article-and open source development. This Article explores almost exclusively the commercial software industry, where software is developed and commercialized in an institutional way. My evidence suggests that within that framework patents may be useful, and that if they are it will be largely because they offer more benefits than costs to small firms. However, coexisting with the commercial software industry is an open source community attempting to foster the development of software largely without commercial investment or affirmative IP protections.255 Those who work in that community may have little need for patents. The cooperative nature of development in that environment obviates any need for actual or implicit cross-licensing that disseminates access to technology throughout the commercial software sector. Similarly, because open source developers at least theoretically do not depend on outside equity investment to any significant degree, the limited ability to appropriate a software invention might pose little harm to them.

One problem, however, is that the open source community does not exist in a vacuum. It exists in a world in which the commercial software industry is building up large portfolios of protected IP, portfolios that pose a serious threat to open source development.256 To put the matter in a current context, suppose for a moment that the Linux operating system in fact infringes in a substantial way patents held by some major proprietary software firm. That could result in liability for all of the many firms using the Linux operating system. The problem is that the open source community has set itself outside the cooperative IP framework of the mainstream software industry. Thus, its members often have no patents of their own with which they might protect themselves in litigation. At the same time, this community has developed its software with the same cavalier attitude regarding the possibility of patent infringement as commercial software firms exemplify. Those two habits cannot coexist in the end.

That raises the question, in turn, whether the potential for high quality software development through the open source model justifies eradication of software patents for the commercial software industry. To put it another way, one potential cost of permitting ready enforcement of software patents is the disabling of the open source model.257 It is difficult to answer that question definitively without evidence that would allow a comparative weighing of the benefits of open source development against the benefits that the commercial software industry derives from IP. It does seem relevant, however, that the reluctance of the open source community to obtain patents is largely a political statement, not something necessary to the development of the improvements in functionality that the movement promises.258 This is proven most clearly by the recent applications for patents filed by noted Linux distributor Red Hat.259 To the extent the open source community is put at risk solely because of moral distaste for patents, the claim that the proprietary community should not be able to use patents to advantage its own operations is weakened.

In any event, a thorough analysis of that question is beyond the scope of this Article. Among other things, such an analysis has to account for the rapid convergence of commercial and open source licensing models: even proprietary licenses now commonly allow access to source code and purportedly open source licenses regularly permit commercial development of proprietary works derived from the covered products.260 It also must account for the effects of those licenses on the character of commercial investment in open source software. For example, it is plausible that restrictions in common open source licenses might tend to tilt the scales of proprietary investments in favor of service firms rather than products firms. If, as seems likely, it is more difficult for startups entering the industry to compete in service sectors than in product sectors, this suggests in turn that the spread of open source software in fact could promote concentration in the software industry. I note these issues here only to define the bounds of my analysis. The primary goal of this Article is to consider the role of IP in the commercial software industry. I leave for another day the relation between open source development and that industry.

V. The Role of Other Existing Systems

If the ultimate question is whether patents facilitate the ability of software firms to appropriate gains from innovation, the picture must include some understanding of alternate methods that firms might use to appropriate the value of inventions. These alternate methods are particularly important given the clear evidence that in many circumstances patents will do little to allow a firm to exclude competitors from a firm's innovation. This Part discusses the two main existing legal systems that complement the protections afforded by patents: copyright and trade secrets.

A. Copyright

Like patent protection, the role of copyright protection changes markedly as a firm develops. My interviews suggest that copyright is of relatively little value in protecting the startup from competitors. It does have value, however, in two particular circumstances: preventing piracy at a company's later stage when it has developed a product and preventing "theft" of materials by outgoing employees.

1. The Role of Copyright in Startups.-For purposes of this Article, the key question regarding copyright is the extent to which copyright protection can provide the kind of sustainable differentiation that is important to investors. The preceding Parts reveal considerable ambiguity on the ability of patents to provide that differentiation. With respect to copyright, however, the question is much less ambiguous. Rather, it seems quite clear that copyright is not suited to providing that protection. The problem with copyright protection for software is that the legal system for copyright is not designed to protect functionality. Because functionality in most cases is what makes software products attractive to customers,261 the differentiation that is important to investors is differentiation in functionality. Thus, if the legal system works as designed, copyright is useless at this point.

My interviews strongly supported that perspective. For example, a typical startup executive explained that copyright protection "is not useful to us [because of its inability to protect functionality]. The other person could do just the same thing in a different manner and get around it very easily."262 Another argued: "I've been in the software business for 20 years. Copyrights are worthless. They are totally worthless."263 One thoughtful executive opined:

Copyright solves one problem, which is the whole or partial copying of an expressive application. The whole or partial copying of an application by a pirate you can get. But it doesn't really protect us from sharing our technical information broadly and a company then understanding how our products work. Patents are inter-industry mechanisms for creating value. Copyright is creating protection between the industry and the channel or end customers.264

The most obvious problem with copyright protection for software relates to reverse engineering.265 Generally, the expression protected in computer software by copyright is the lines of code of which the program consists.266 Thus, although copyright does not prohibit a competitor from writing a completely new program that includes the functionality of the existing program, it does bar a competitor from taking the existing code to produce that program.267 Thus, the effect of that protection turns on an empirical question: how effective as an exclusionary device is it to require a competitor to rewrite a competing program instead of reusing the code?

Surprisingly, my interviews indicate quite strongly that it is not effective. The perception is that in most instances a software engineer who observes a program in operation can readily understand the functionality the software provides and with that understanding easily can write code that provides the same functionality: "[S]oftware in general is very malleable and is easily reverse engineered."268 As one venture investor explained: "The difficulty normally is managing the people, not solving the problem. The code won't look the same, but the functionality will."269 One developer explained that the difficulty in coming up with a successful enterprise software product is not writing the code, but understanding the problem that needs to be solved.270 The only IP protection that makes it difficult to duplicate a program's functionality is patent protection, which bars a competitor from writing code that includes any patented aspects of a software's functionality. If copyright systematically stops short of providing that protection, it cannot provide the vehicle for appropriating value that persuades potential investors.

2. The Role of Copyright in Later-Stage Firms.-Yet, it is plain that copyright plays a crucial role in the industry's ability to appropriate returns from the innovation it produces. That role has several aspects, but two are sufficiently pervasive to be characterized as structural: the prevention of piracy and the control of "theft" of code by departing employees and the like.271

a. Piracy.-On the first point, the discussion above explains that copyright protection is unimportant for a startup firm because literal copying of code is not that important to competitors. There is one group, however, that is quite interested in a free right to copy the startup's product: its customers. Thus, although patent and trade secret protection are more important in limiting the ability of competitors-horizontal firms-to take technology from the innovator, copyright is more important in limiting the ability of customers to obtain the product without paying the product's owner.272

That problem-piracy, to use the industry's preferred term-affects different types of software differently. For example, it is less important in the enterprise software market in this country. To be sure, there are reliable methods of limiting piracy in that market. In some cases, firms operate as application service providers, so that the code for their program resides entirely on their own server, which can be protected more readily than their customers' servers.273 Other companies, particularly in the enterprise software sector that is the source of most innovation in the industry, emphasize the practical value of dealing only with large and fiscally responsible "Fortune 500" customers.274 A typical example: "[W]e're selling to an enterprise customer. We're not on a store shelf. So I'm not at risk of somebody copying the disk and just cloning what I do."275 Key to the effectiveness of these arrangements is the likelihood that the customers are large and creditworthy firms. Those types of firms are unlikely to participate in illicit distribution of software code, if only because of the likely financial exposure they would incur if their participation were discovered.

In some markets, however, such protections are not useful. Most obviously, they are not valuable in markets-such as consumer markets-in which software code (in any form) is freely distributed.276 In this country, consumers commonly violate the terms of license agreements, copying and transferring software in ways that would require payment from the new user.277

Executives recognize that in other countries the problem is a serious one even in the enterprise software context.278 For reasons that range from an intentional governmental design to foster piracy to mere lackadaisical toleration of piracy, the extent of piracy in many foreign countries is shocking: industry statistics indicate that in many countries less than twenty percent of the software actually distributed is acquired through lawful channels.279 To be sure, piracy in those countries is not a total loss to the software developer because it helps to develop a network of users that make the product more attractive to others. Still, it does reflect a loss of revenue that the software developer could obtain if its IP rights were enforced effectively.

For several reasons, copyright is the only effective IP protection against piracy.280 For example, even if pirated software is protected in part by a patent, a suit against the pirate challenging patent infringement necessarily is more difficult because of the need for the software owner to establish the patent's validity. Because of the relatively high standard of patentability-compared to the copyright standard of originality-it always will be difficult for the patent owner to get over the threshold of patentability.281 Because of the low threshold of copyrightability, it never will be difficult for the owner of copyrighted software to establish that the software includes copyrightable innovation.282 In addition, the limitations on copyright protection discussed above, which make copyright useless for the startup trying to protect the functionality of its software, are irrelevant in the case of piracy: the pirate by definition has copied all or substantially all of the product. These problems associated with patent protection in this context are particularly salient in the enforcement of criminal sanctions for piracy. Statistics from the Department of Justice suggest that the federal government often sues pirates for criminal copyright infringement; there is not a statute for criminal patent infringement.283

b. Premarket Protection.-Copyright also plays a role before a product goes to market by helping a firm prevent technology from leaking out through the actions of employees and business partners. The obvious problem is policing the activity of departing employees. It is common in all startup sectors-including the software sector-for employees to rapidly move from firm to firm. Indeed, Ron Gilson argues with considerable force that California rules limiting the ability of firms to prevent those moves are crucial to the success of the venture capital industry in Silicon Valley.284 Yet, it is one thing for employers to allow cross-pollination of employees and their human capital and experiences. It is another for their employees to take substantial pieces of "product" out the door with them and sell that product from their new company.285 As Rob Merges explains, it is optimal for firms to have some control over that activity, and it is not clear that parties can protect themselves adequately through contracts alone.286 Although at least theoretically this could work to the benefit of early-stage startups, it is much more likely to be a benefit to larger firms (with more code worth plundering) and a detriment to smaller firms (trying to begin operations with employees recently departed from larger firms).

Thus, although the discussion in the next section evinces considerable skepticism about the social value of robust trade secret protections, my intuition is that copyright protection serves an important function. In this context, patent protection is relatively ineffective because of the litigation costs and uncertainty of such litigation. But in cases in which the employees attempt to reuse a substantial amount of code from their previous firm,287 copyright law provides a simple and effective remedy against the new firm.288 The importance of that constraint is evident from the discussions of corporate counsel about their diligence with respect to new employees289 and from venture capitalists about their investigation of potential investments. The only instance in which I heard venture capitalists express concern about preexisting IP of other firms constraining the ability of their potential portfolio firms is the case in which a startup has engineers with previous experience designing a similar product, which raises the risk of code contamination.290 Although state law causes of action based on misappropriation of a trade secret, unfair competition, or breach of employment agreements might remedy some of those problems, the clarity and simplicity of the copyright action, the ready availability of federal jurisdiction, and the statutory remedies combine to make it a significant tool in policing such conduct.291

A similar problem occurs for large companies engaged in collaborative development projects. In that context, the copyright protection that attaches during the development process is an important part of preserving exclusive rights to the code as it passes among the participants in the process.292 Although the participants in that process are free to use contracts to define the rights each of them has in the various portions of the project, the lawyers that participate in the process argue that injunctive remedies and statutory damages available under the Copyright Act play an important role in establishing a robust enforcement backdrop for those arrangements.293 As one executive explained:

The protection of [a major new product], as it is being designed, and built, and tested, and being distributed to third parties, is critical. And anything that diminishes the protection of that pre-release code will impede our ability and willingness to get outsiders to look at the code and test that.294

B. Trade Secret

Trade secret protection plainly plays an important role in the software industry if only because many companies have no formal IP protection for their products295 and take significant steps to keep the details of their technology secret. There are limits, however, to the extent that trade secret protection can provide a robust appropriability mechanism in the software industry.

First, as noted above, my interviews strongly suggest that it is easy for competitors who observe a new product to design and deploy products that include the functionality of that new product. Such conduct does not violate trade secret law.296 More broadly, trade secrecy does nothing to provide the "foothold" protection that is useful to smaller firms trying to fend off large firm efforts to market competing products. In an industry in which innovation involves many firms trying to do similar things at the same time, it is likely that a large firm like IBM or Microsoft might make the same advance that a small firm has made, even if that large firm has no access to the small firm's technology and thus no responsibility under trade secret law. Although patents arguably give small firms some shelter in those contests, trade secret law does not offer the same protection.

Finally, as suggested above, there is some reason to think that vigorous enforcement of trade secret protections in some contexts-against departing employees, for example-is itself costly to the industry. Saxenian and Gilson present a rich descriptive account arguing that the development of a rich innovative culture in Silicon Valley depends in part on the free transmission of noncodified information297 by employees moving from firm to firm.298 Given the picture of cross-pollinating innovation I describe in this Article, that effect should be particularly valuable in the software industry.

Copyrights and trade secrets, then, play an important role in protecting investments in software. But it is a role weighted in the opposite direction of the role that patents play: copyrights and trade secrets, to the extent they are useful, tend to support the efforts of large incumbent firms and to hinder the efforts of smaller entering firms seeking a foothold for competition.

VI. Responding to the Problems

I do not intend to propose a new system for IP protection in the software industry.299 Indeed, our international treaty obligations make it difficult for us to substitute any system for the systems now in place. I do think, however, that it is useful to explore in a summary way the possibility that some reform short of a full-scale abolition of patents might solve the problems that patents cause while leaving in place the benefits they provide. I start with a brief discussion of a more radical response-a registration regime that would separate disclosure and filing from exclusivity. I then discuss some more incremental possibilities. My discussion is frankly skeptical in tone, reflecting my view that it is difficult to be sure that radical interventions would improve the existing system, but I remain hopeful that minor revisions could limit problems that threaten to destabilize the existing equilibrium.

A. Registration

The direct benefits of patents necessarily depend on the right of the patentee to exclude competitors. That is not so clear, however, with respect to the indirect benefits discussed above.300 In particular, the indirect benefits that seem to produce social value depend on a range of effects that do not require that codified information be deployed in order to exclude competitors from deploying any particular product. This suggests that those benefits could be obtained through a simpler registration system-one in which the tec