The House Antitrust Subcommittee Report makes clear the weakness not just of competition in the tech sector, but also of competition policy toward the tech sector.  We argue that this is the result of the failure to employ one important tool of antitrust, namely, structural separation.  Economics and experience teach that breaking up such firms is almost surely necessary in order to make the companies focus on serving customers better rather than on handicapping their rivals.  We review the evidence, discuss the alternatives, and provide guidance for how structural separation can be employed in bringing competition to the tech sector.

By John Kwoka & Tommaso Valletti1



The House Antitrust Subcommittee has concluded its milestone study of Amazon, Apple, Facebook, and Google with a lengthy report entitled “Investigation of Competition in Digital Markets.” Despite that title, the report is in fact not so much about the weakness of competition in digital markets as it is a report on the weakness of competition policy toward digital markets. After all, the bulk of the report documents the unimpeded rise to dominance of these giants over online search, ecommerce, social media, and advertising, and how this dominance has “diminished consumer choice, eroded innovation and entrepreneurship in the U.S. economy, weakened the vibrancy of the free and diverse press, and undermined Americans’ privacy.” Since these objectives and values are very much the mission of antitrust and other public policies, there is no escaping the fact that it is the failure of those policies and institutions that has resulted in these harms.

To be sure, there have been other contributing factors, from novel technology to agency resource constraints, but evidence of the responsibility of policy is in plain sight. In the face of a tsunami of acquisitions by the tech companies — some 800 over the past 20 years — the antitrust agencies have been spectators. The Federal Trade Commission and the Antitrust Division of the Justice Department have conducted formal investigations in only a handful of cases, and have prohibited exactly none, zero, out of these hundreds. There have been endless reports of anticompetitive conduct by these companies, including self-preferencing and other forms of bias, tying and predatory conduct, misuse of competitors’ and consumers’ data, and the like. In the few cases where the agencies did not dismiss these concerns out of deference to apparent “efficiencies,” they have ended their inquiries with remedies that have proved largely ineffective.

In short, the tech companies could hardly have hoped for a more tolerant policy toward their rise to dominance. In response, the House report now offers a total of 13 recommendations across three broad areas: Restoring Competition in the Digital Economy, Strengthening the Antitrust Laws, and Strengthening Antitrust Enforcement. All of these have merit. Many are familiar from experience in other industries. A number seem likely to attract support in congress and the agencies.

But it is our view that among all of the report’s policy recommendations, one is of over-riding importance. It is of over-riding importance because it is the single most potent policy initiative; because without it, other recommended policies are not likely to suffice; and indeed, because this one policy can lessen the need for other policies.

This singularly important policy is structural separation, that is, breaking up the firms. As we explain below, structural separation is likely the only way to truly re-orient these companies’ incentives toward customers, rivals, and competition, to limit their ability to engage in anticompetitive actions, and to permit the antitrust agencies to step back from on-going intense regulation of these companies. Structural separation would drastically diminish if not eliminate self-preferencing, tying, and other such practices by these companies, relieving the agencies of the burden of identifying, challenging, proving, and seeking to remedy each such instance of anticompetitive behavior. Structural separation would embed these companies firmly in markets where they face rivals that could challenge their positions and where they would have to focus on serving customers better rather than handicapping their rivals.

It is therefore to the credit of the House report that it lists as its first recommendation “Structural separations and prohibitions of certain dominant platforms from operating in adjacent lines of business.” But in the past and predictably in the case of the House report, that recommendation is greeted with the dismissive statement that it is operationally impossible to undo big firms, or more costly than any possible benefits, or inferior to other policies. While evidence for these objections is notably scarce, this criticism has served to sideline that strategy whenever it is advanced as a possible solution.

We disagree with these criticisms.2 The reality is that antitrust and regulatory policy has not infrequently broken up firms in a wide variety of markets. Companies themselves routinely divest nearly as many businesses as they acquire each year. Digital companies in particular engage in an endless pattern of acquisition, assessment, and divesting of businesses. Claims that structural separation is impossible do not survive even the most cursory examination of the evidence. Moreover, when breakups and divestitures do happen, they have been accomplished without a record of disastrous consequences for either the core company or the divested operations. Indeed, the record suggests that breakups of most such companies have been operationally successful and competitively beneficial. And finally, when structural separation has been rejected in favor of other policies, that record is replete with failures of the latter. Indeed, it is these very failures of weak policy toward the tech companies that have brought us to this point.



A few examples illustrate these various points. With respect to the feasibility of breakups, consider the cases of American Tobacco and Standard Oil, both restructured by judicial action in 1911. Standard was the easier case since the mandated restructuring — into a total of 34 separate companies — was along geographic lines. American Tobacco was a more fully integrated national company but it was nonetheless successfully turned into three distinct companies and in a period no greater than eight months. In both cases true competition was slow to emerge but the break ups themselves were accomplished without the disruption or disaster that was predicted.

More recently, the AT&T divestiture was asserted by the company and many outsiders to be literally impossible, or certain to cause the collapse of the industry if accomplished. Neither of those predictions was remotely accurate. The vertical and horizontal breakup of this enormous company into eight parts was achieved in the two-year time frame imposed by the court, with widely acknowledged benefits to consumers and businesses alike. While some have criticized the extended role of the court, it is useful to note that was in no small part because of the obstructive actions of the post-divestiture companies, an experience worth studying to ensure that future breakups of firms proceed more smoothly.

In the more recent Microsoft case, the trial court ordered the company to be broken up in ways that reflected the natural divisions within the company. One division would consist of its core operating system, that is, Windows, while the second would encompass its numerous derivative applications, headed by Office. This structural remedy was rejected on judicial appeal, resulting in protracted hand-to-hand combat between the company and the court in an effort to gain compliance with a complex rules-based decree. As we discuss below, this effort to find an alternative to structural separation has its own lesson, namely, the near-impossibility of using a conduct-oriented order to prevent anticompetitive actions by such a company.

It is also worth bearing in mind that regulators in many countries have restructured numerous large companies in several industries in order to enhance competition. British Rail and British Telecom (“BT”) in the UK; telecoms throughout Europe as well as AT&T in the U.S.; and scores of large electric power companies in the U.S. and throughout the world have all been broken up. For reasons worth studying, some breakups have been more successful than others in fostering competition, but operationally there are numerous cases in which two or more companies have been successfully carved out of a single dominant firm as a result of policy action.

Interestingly, breakups are often not regulators’ first choice. They typically have come to the decision to require divestiture only after lengthy efforts to impose operating rules and constraints on these companies have failed. This was the case with the AT&T and other telecom breakups, where divestiture was pursued when the regulator could not or would not act, as well as with the countless divestitures of electric power companies, where mandatory access and tariff policies were tried and ultimately failed.

And finally, we note that breakups initiated by firms themselves, including in the tech sector, are quite common. While the motivation for these is companies’ own interests rather than regulatory or antitrust imperatives, our point is that experiences such as eBay/PayPal, Pfizer, GE, HP, and a great many others establish the feasibility of successful corporate separation.



The above experiences underscore the frequency and success of structural separation. Also relevant are the failures of the alternative rules-and-remedies approach toward dominant firms. Rules and remedies have a poor record since they are fundamentally an effort to make the company act against its own interest in maximizing profit by instructing it to avoid certain actions that would raise its rivals’ costs, deny rivals competitive opportunities, or otherwise unfairly disadvantage them.

But no written order or instruction alters the firm’s incentive to engage in these actions, and so the firm will predictably make every effort to avoid or evade the constraint it faces. Where that order affects an operation or transaction of greater value to the company, it is likely to make a correspondingly greater effort to do so. In addition, the company is well positioned to do so since it has much better information than the regulator about its products, divisions, technology, transactions, customers, and so forth. As a result, it has considerable advantages over the regulator in interpreting and complying with the order in ways that minimize or avoid its impact.

Examples of these difficulties abound. Evasion by interpretation is illustrated in a merger context by the U.S. Justice Department’s conduct remedy accompanying its approval of the merger of Ticketmaster and Live Nation in 2010. The order sought to preserve rivals’ ability to compete against the merged company by specifying that it must not “condition or threaten to condition the provision of Live Nation Entertainment Events” on whether a venue owner had contracted for ticketing services with a servicer other than Ticketmaster. The company engaged in precisely this conduct, arguing that the language of the order prohibited it only from denying all Live Nation events to a venue owner, not just one or several events. Only after a decade of anticompetitive practices was the order finally changed to prevent this.3

Evasion by stalling is illustrated in the tech area by one provision in the 2001 Microsoft decree. That provision required the company to license to third parties its communications protocols for connecting servers to desktop computers, and to provide documentation within three years. The company repeatedly claimed it was unable to compile the necessary documentation, either because of its inherent complexity, or because of the large number of other requirements in the order, or because it could not “find[…] and hire[…] competent employees with the necessary experience in and training for these highly specialized tasks.” Without a good basis for challenging each excuse, the court repeatedly extended the deadline. The Justice Department eventually declared Microsoft’s work “substantially complete” (even though hundreds of unresolved issues remained) six years later and nine years after the initial order.4

The EU has been more active with respect to the tech companies but its remedy approach has similarly proven inadequate. In its proceedings against Google Shopping, it acknowledged there might be more than one way to resolve the competitive concern and so, instead of imposing a specific fix, it ordered Google to propose its own method for “treating competing comparison shopping services no less favorably than its own comparison shopping service.” Not surprisingly, Google responded with proposals that, many said, worked to its further advantage. One such proposal, for example, would create an auction in which various comparison shopping services (including Google’s own) would bid for placement. Rivals complained that having to pay Google to cure its anticompetitive actions was not much of a remedy. Moreover, since Google’s payments for its own placement would simply go into a different company account, this was no real disincentive to its continued dominance.

In other instances, the tech companies have sought to evade the imposition of a remedy by promising to adhere to specific standards that would seem to cure the problem. In a number of such cases competition authorities have subsequently discovered either flagrant violations or clever wording to justify violations. In its effort to acquire WhatsApp, for example Facebook promised the EU that it would be impossible to create automated matching between the two companies’ user accounts. Evidence emerged that Facebook knew full well that it was possible and two years later began that very process, with a miniscule penalty. Google is now promising the EU that as a condition of its acquisition of Fitbit it will “not use individual/personal Fitbit data for advertising.” But this promise could easily be evaded by using a more limited data set to gain insights and then train its algorithm accordingly

In the U.S., Amazon addressed concerns that it was using data on its marketplace sellers to launch its own branded competing products by assuring Congress that it did not use “individual seller data directly to compete” with such businesses. But the company soon had to acknowledge that is uses “aggregated data” from independent sellers in exactly this way, leaving unanswered the question of the meaning of “aggregated.”

These examples illustrate key distinctions between a rules-and-remedies approach and structural separation. Most fundamentally, structural separation creates incentives for each entity to maximize its own separate profit, much like in a competitive market, rather than to evade constraints on the ability of the single large firm to maximize profit. Separate and independent companies have sharp and visible boundaries, which firms would be reluctant to cross (as with collusive practices). By contrast, rules and remedies are characterized by blurred lines, compromised incentives, and unobservable actions. As some of our examples have shown, rules and remedies often depend crucially on the specific language of an order, inviting the firm’s own interpretation and efforts to explain away potentially problematic conduct. In addition, many troublesome actions are deeply buried in a company and difficult for an adversely affected company to identify and nearly impossible for a competition agency to observe. Algorithms that steer business, misuse of data, and other practices cited in the above cases illustrate these problems.

For all these reasons, it is difficult for a competition agency to monitor and enforce rules and remedies. Competition agencies are not designed to be regulatory institutions with constant oversight. The tech companies have features that make this reactive approach especially unlikely to work. Their production processes — digital technologies — are opaque and therefore not readily observed by an outsider. Their services are malleable — changeable at their discretion — permitting endless ways to circumvent an order or rule. The technology is constantly changing in ways that are difficult to predict but can render existing, static constraints irrelevant. And these firm’s incentives to evade or avoid are enormous, measured by the profitability of the affected parts of their businesses.



This discussion makes clear that a rules-and-remedies approach toward tech company competition problems is unlikely to be satisfactory, and very possibly altogether ineffective.5 This is important to bear in mind since the major objections to breaking up tech firms are that it may be costly or not work well. But the correct basis for judging the breakup approach is not whether it is costless or perfect, but simply whether it is superior to actual alternative approaches. In fact, we have run the experiment on the latter, and we know how it comes out. Tolerance of large firms while attempting to control their specific anticompetitive behavior has not worked. While it is appealing to policymakers since it holds out the promise of “having one’s cake and eating it, too,” time after time this approach has failed, often predictably and miserably. It is for these reasons that we argue that any competition policy toward the tech sector must include the possibility of breakups — and will not be effective unless it does.   

To be clear, we do not advocate breaking up the core platforms of the tech companies, but as we have discussed elsewhere, these companies in fact have identifiable fault lines along which structural separation appears feasible.6 These fault lines are in two dimensions. The first is based on whether they have resulted from acquisitions or by internal development, the second by whether they constitute plausible substitutes or complements to the core platform (or possibly neither). We would argue that generally speaking business operations that have been acquired are likely to reveal clearer fault lines than those developed internally. We would further argue that businesses that are plausible substitutes for core operations constitute more important candidates for divestiture on competitive grounds. This typology would suggest as candidates for separation, for example, Facebook’s Instagram and WhatsApp operations, both because they were the result of acquisition and also because they represent plausible alternatives to Facebook’s core social media platform.

Elsewhere we have categorized many more parts of Amazon, Apple, Facebook, Google, and Microsoft into these categories.7 Here we simply emphasize that a close examination of the companies can reveal some quite visible fault lines along which structural separation is entirely plausible, and that such an exercise should be an important tool of competition policy toward these companies. Indeed, nothing short of that will succeed in the objectives of the House committee report, namely, reducing the market power of these companies by limiting their ability to thwart and distort competition.

1 Respectively, Northeastern University and Imperial College London.

2  Our full case is made in “Scrambled Eggs and Paralyzed Policy:  Breaking Up Consummated Mergers and Dominant Firms, J. Kwoka & T. Valletti, forthcoming, Industrial and Corporate Change, 2021.  Available at

3 J. Kwoka, “Conduct remedies with 2020 hindsight: Have we learned anything in the past decade?” CPI Chronicle, April 2020.

4  A. Gavil & H. First, The Microsoft Antitrust Cases.  MIT Press, 2014.

5  Perhaps worse yet, rules and remedies can give the appearance of agency action, even when they have little prospect of success.

6  Kwoka & Valletti, op. cit.

7  Ibid.