The long-awaited Google Android decision is out (see the press release here or the full text here). The General Court has dismissed Google’s action almost in its entirety, slightly reducing the record fine from 4.343 billion euros to (still a record) 4.125 billion euros. This post gives a very first overview of the judgment and highlights some of the most relevant aspects.


The facts

The conduct in question is more complex than in the Google Shopping decision as it involves multiple layers and a larger number of, albeit interconnected, markets. Google Shopping was about a classical leveraging from one market (the one for general search, in which Google was clearly dominant) to an adjacent market (the one for specialized search, in which Google was initially quite unsuccessful). The abuse consisted essentially of Google giving preferential treatment to Google Shopping in its general search results (see for a brief reminder of Google Shopping here).

In Google Android, things are far more complicated:

First, the case involves four interconnected product markets; (i) the market for the licensing of smart mobile device operating systems (AndroidOS); (ii) the market for Android app stores (Play Store); (iii) the market for the provision of general search services (Google, Yelp…), (iv) the market for non- OS-specific mobile web browsers (Chrome, Opera, Firefox…).

Second, like the markets, Google’s conduct was multi-layered and interconnected. The Commission had identified three types of restrictions: (i) through distribution agreements, Google required manufacturers of mobile devices (e.g., Samsung) to pre-install the Google Search and the Chrome app to get a licence to use the Play Store on their devices; (ii) through ‘anti-fragmentation agreements (AFAs)’, Google granted the operating licence for the pre-installation of Google Search and Play Store apps only to manufacturers that did not sell devices running on Android versions not approved by Google (so-called non-compatible forks); (iii) lastly, Google had entered into ‘revenue sharing agreements’ (‘RSAs’) with mobile device manufacturers and mobile network operators under which Google shared some of its advertising revenue on the condition that no competing general search services were pre-installed on a predefined portfolio of devices.

Third, the conduct ultimately aimed at making consumers use Google Search – which anyway most consumers did and do-. It, therefore, does not perfectly fit into the leveraging category, but the conduct is also about building a protective moat around the already existing dominance in the market for general search.


Dominance in markets or ecosystems?

Google had highlighted that its various products (Google Search, Android, Play Store, Chrome…) should not be looked at individually but rather as part of a broader Google ecosystem, which is subject to vigorous competition from the Apple ecosystem.

The concept of ecosystems has received quite some attention recently (see for example the contributions to an OECD expert hearing on the matter). The Court, following the Commission, stuck to a rather traditional approach and defined individual product markets (paras 102 ff.). It reiterated that “the definition of the relevant market is […], as a general rule, a prerequisite of any assessment of whether the undertaking concerned holds a dominant position.” (para 105) This will remind some readers of the US Supreme Court’s statement in the Amex case (footnote 7) and will disappoint those who favour direct measurement of market power (e.g., through mark-ups). However, the Court (and previously the Commission) recognized that in interconnected markets, competitive constraints can be exerted by companies even if they are not necessarily active in the same product markets (para 122). Thus, irrespective of the market definition, it was important to assess whether Apple and the iOS ecosystem sufficiently constrained Google and the Android ecosystem. The Commission had done so with the conclusion that this was not the case. This was confirmed by the Court (para 143 ff.). The Commission was hence correct to conclude that Google was dominant except in the market for non-OS-specific mobile web browsers.

Concerning Google’s dominance in the market for licensable OSs, the Commission conducted a SSNDQ (‘small but significant and non-transitory degradation of quality’)-test. The Court accepted how the Commission carried out this test, recognizing that “defining a precise quantitative standard of degradation of quality of the target product cannot be a prerequisite for the application of the SSNDQ test” (para 180).

This is an important insight for antitrust in zero-price (and potentially in all) markets: while economic quantification, where possible, is useful, the limits of such quantifications do not at the same time demarcate the limits of antitrust law. Other arguments the Commission relied on were switching costs and user loyalty to operating systems. The Court accepted the Commission’s assessment and rejected Google’s argument that the open-source nature of Android would assert competitive pressure on the company. Despite Android being open source, any company wishing to develop an Android fork as a reaction to a decrease in the quality of Android would have to incur significant costs (para 227). Moreover, Google prevented the emergence of Android alternatives through the AFAs (para 228).

The Court also accepted the Commission’s assessment that Google was dominant in the market for Android app stores (paras 235 ff.). While the market definition seems to already anticipate this result (who can name another Android app store than the Play Store?), the Commission nevertheless had considered the competitive constraints that could result from ‘app stores for non-licensable mobile OSs’ (i.e., Apple’s App Store). It linked this argument to the previous one related to the competition between operating systems. Apple’s App Store and Google’s Play Store competed only indirectly through the operating systems to which the respective app stores are linked. The Court followed this line of argumentation (paras 248 ff.).

Google’s dominance in the market for general search will hardly surprise the reader. Google did not try to challenge dominance as such but argued that the Commission had contradicted itself by defining the market vis-à-vis end users but assessing conduct that was not directed at end users but at original equipment manufacturers (‘OEM’S). The Court rejected Google’s claim, holding that “Google’s dominant position on the national markets for general search services was both the starting point and the objective of the practices […], which were in reality designed, according to the Commission, to preserve and to increase Google’s power on the national markets for general search service and to prevent the emergence of any competitor on that market”. (para 262) In a pure leveraging case, such as Google Shopping, the argument might have worked. In a ‘protective moat’ case, it could not.


The first abuse: pre-installation conditions (tying)

While Android is free and open source, other Google apps aren’t. This includes in particular the Play Store. To obtain a licence to use the Play Store on their devices, OEMs had to enter into Mobile Application Distribution Agreements (‘MADAs’) with Google, which required them to pre-install Google Search (general search app) and the Chrome browser. This abuse could be characterized as leveraging: Google used its dominance in the market for Android app stores to gain market shares in the markets for browsers (Chrome) and general search (Google general search and Google Search app).

This abuse is framed in terms of a classical tying case, relying on the Microsoft precedent (paras  275 ff.). The argumentation is therefore rather formalistic. It gets interesting only when it comes to the restriction of competition (paras 304 ff.). The Commission’s main argument was that the pre-installation gives Google Search a competitive advantage that cannot be offset by any competitor.

Here, one needs to keep in mind the interconnection between the different affected markets and the different conducts. Google Chrome uses Google Search as the default and does not allow other default settings (para 493). Because of revenue sharing agreements, other browsers (that could be and sometimes were pre-installed) typically also had Google Search as the default search engine (para 448 ff.). Certainly, the pre-installation of Google Search and Google Chrome (with Google Search as a default) leads to a competitive advantage over search engines or browsers, be it due to a ‘status quo bias’ or a ‘default setting’ (para 331).

When it comes to the question of whether this advantage was based on competition on the merits or not, the judgment in my opinion remains a little bit vague. Basically, the Court found that Google competitors were not able to offset the competitive advantage conferred by the MADA pre-installation conditions. They would not have been able to compete effectively with Google either by entering into pre-installation agreements themselves or by trying to reach users through different channels (e.g., through the Play Store where competing general search apps can be downloaded).

Google’s argument that “a competitive disadvantage does not amount to anticompetitive foreclosure” (para 543) could have been taken more seriously. However, the formalism here is linked to the categorization of the conduct as tying: the criteria to find tying abusive do not entail an explicit reference to competition on the merits. The link here is created by the criterion that “the undertaking concerned does not give consumers a choice to obtain the tying product without the tied product” (para 284).

In essence, this means that in tying cases, the dominant undertaking bears the burden of proof that the success of the tied product (in the Google case, for example, the Google Search app) is due to it being superior to other products. Here, the Court seems to apply a very strict standard, stating that “even assuming that Google Search and Chrome are superior in terms of quality to the services offered by rivals, that would not be decisive since it is not claimed that the various services offered by the rivals are not technically capable of meeting consumer needs.” (para 577) This appears a rather harsh statement and it is not entirely clear if this is really the standard of proof that the Court is applying. Eventually, it also considers the Commission’s assessment that the “quality assessment of the various competing services remains similar” (para 582).

The Court also rejected Google’s arguments as to an objective justification for its conduct (paras 599 ff.). In particular, it rejected the argument that the practice enabled “Google to monetise its investment in Android and in its non-revenue generating apps” (para 604). If and how a dominant undertaking makes money, is not the Court’s (or Commission’s) problem. Google is not obliged to offer the Play Store (or any other product) for free (para 617).


The second abuse? Revenue sharing agreements under the condition of pre-installation

Google shared some of its advertising revenue (from search queries via Google Search, Chrome, and the URL bar of other web browsers if Google was set as a default) with OEMs or MNOs, if they exclusively pre-installed Google Search on all devices within a predefined and agreed portfolio. Were an OEM or MNO to pre-install another search app on any device within that portfolio, it would have to forego the revenue in respect of the entire portfolio. According to the Commission, this restricted competition because it eliminated any incentive to pre-install competing general search services. Google’s competitors could not have offered a similar revenue for being pre-installed on all devices within the portfolio as much as Google did. Here, the concern is that Google used its dominance in general search (where it gained the revenue that was to be shared) to build a protective moat and hinder competitors from entering the market.

This is one of the most interesting parts of the judgment as it becomes quite technical. The Commission, in order to show the practice’s exclusionary effect, conducted an as-efficient-competitor (‘AEC’) test in form of a quantitative analysis. The Court found this analysis to be erroneous, largely for procedural reasons and annulled the Commission’s decision insofar as it considered portfolio-based revenue sharing agreements as a separate abuse (para 802).

Repeating the findings of the Intel judgment, the Court considered that where the dominant undertaking claims “that its conduct was not capable of restricting competition and, in particular, of producing the alleged exclusionary effects, it is then for the Commission, in order to establish the culpability of that undertaking, to analyse the various circumstances that demonstrate the restriction of competition resulting from the contested practice” (para 639). Thus, the Court also requires the Commission to assess whether the practice excludes competitors that are at least as efficient as the dominant undertaking (para 640). To do so, “a test known as the ‘as efficient competitor test’ (‘the AEC test’) can be useful” (para 641). This statement is quite remarkable: the Court avoids to state that such a test is always necessary (see also para 643)! However, where the Commission applies the AEC test, it must do so “rigorously” (para 644).

According to the Court, this did not happen:

First, the Court did not agree with the Commission that the agreement covered a ‘significant’ part of the markets for general search services. The coverage of the conduct did not exceed 10-20 %. A higher coverage would have been found, had the Commission also included the exclusivity payments between Google and Apple (Google pays Apple 8 to 12 billion dollars annually to be set as the default search engine for Safari, Siri and Spotlight, see the US DoJ’s lawsuit against Google, para 118) – but the Commission had chosen not to include the agreement with Apple in its decision (paras 691 f.).

Second, the purpose of the AEC test, as conducted by the Commission, was to find out whether an as efficient competitor to Google could have matched the payments under the RSAs (equal to the loss for OEMs and MNOs if they installed a competing general search app alongside Google Search). The Court recalls that an “as efficient competitor is a competitor which […] at the very least […] has the same ability to generate revenue and is faced with the same costs as those of the undertaking in a dominant position” (para 739). Those costs, according to the Court, refer to incremental costs rather than operational costs. The Commission, however, had relied mainly on the latter and had not considered economic data from Google itself (paras 738 ff). In this regard, it is also important to note that even though Google did not on its own initiative submit such data, the Commission could and should have requested these data (paras 747 f.). Ultimately, therefore there were “doubts as to the correctness and validity of the AEC test carried out by the Commission” (para 752).


The third abuse: anti-fragmentation agreement (bundling)

Google made licences for the Play Store and Google Search conditional on the acceptance of an anti-fragmentation agreement. Google’s AFAs require minimum quality standards, such as that the device enables the installation of apps, basic security features, a complete set of Android Application Programming Interfaces (‘APIs’) and correctly reporting of screen size to apps (para 806). Android forks (i.e., OSs developed from the Android source code) have to pass a series of compatibility tests to demonstrate that they comply with these requirements. The most famous example of a non-compatible Android fork is Amazon’s FireOS on which the Play Store and other Google apps are not pre-installed. The Commission had not challenged these compliance obligations as such but only one specific feature: OEMs that wanted to get licences for Play Store and Google Search for some of their devices, had to comply with the AFAs for all of their other devices as well. An OEM would therefore not have been able to offer one device with Play Store and Google Search pre-installed on a compatible fork and another one without those apps on a non-compatible fork (para 810). The concern here: leveraging of dominance in the market for Android app stores in the market for general search. While the Commission succeeded with its arguments here, this appears to nevertheless be one of the most disputed aspects of the decision. There were several interveners, including Gigaset, HMD and Opera, arguing that the practice was actually pro-competitive (para 824). They highlighted that Google’s proprietary APIs on Android forks was necessary to make apps developed for Android function properly. App developers would therefore benefit from Google’s AFAs which secured Android to be a reliable platform for all users.

The Commission’s analysis with regard to the third abuse, as with the tying (first abuse) was rather formalistic. It focused on the exclusionary effect in the market for Google Search. This is interesting insofar as there is only an indirect relationship here.

First and foremost, the conduct affected the market for licensable operating systems – it restricts market access for non-compatible Android forks as competitors of the Android operating system. It is thus little surprise that Google emphasized that the AFAs did not prohibit installing competing search services. The Court, however, accepted the Commission’s assessment. The link between competition in the market for operating systems and competition in the market for general search is the following: non-compatible Android forks could not have certain Google products, e.g., Play Store and Google Search, pre-installed. Therefore, were they successful, they could have emerged as an important distribution channel for competing general search services. One could imagine, for example, a privacy-friendly phone running on a non-compatible Android fork that has DuckDuckGo rather than Google Apps pre-installed.

This argument appears somewhat circular: because Google did not allow Google Search on non-compatible Android forks, those could have become a threat to Google in the market for general search, so Google must not have impeded the success of Android forks. What about the counterfactual in which Google did allow the installation of Google Search and on non-compatible Android forks? Would this have also been anti-competitive? Then more forks might have been developed but without making market entry easier for other search engines. In this scenario, there would have been a trade-off between competition in the market for Android (forks) OSs and competition in the market for general search services. It appears a bit hasty to simply conclude that more competition in the first market necessarily leads to more competition in the second market.

Google also put forward a number of arguments for an objective justification of its conduct. In particular, it claimed that the AFAs were necessary to ensure the “integrity and quality of the Android platform” (para 868), in particular as regards the open-source nature of the platform (para 869) as well as Google’s reputation. This argument does indeed appear to have some merits. Previous open-source platforms like Unix, Symbian and Linux Mobile suffered from incompatibilities. Nevertheless, Google cannot explain why the AFAs are linked to the installation of the Play Store and Google Search rather than a condition for all uses of Android (para 872). The Court then also rejects Google’s arguments. First, in light of Google’s dominance and success with the Android operating system, it does not accept that incompatible Android forks could seriously threaten the Android ecosystem (para 880). Second, regarding Google’s reputation, the Court finds that Google could have put measures in place to avoid confusion between “the commercial origin of devices running on Android-compatible forks” and those running on incompatible forks, such as through trademarks (para 883).

Thus, the standard applied here was relatively low: no AEC test needed to be conducted and the formalistic approach to tying and bundling cases established in Microsoft was confirmed.


The rights of the defence

Like already in the Qualcomm judgment, the Court expressed some concern as to how the Commission takes and shares notes of meetings with third parties. Whenever an interview (even if it is conducted by the Commissioner for Competition herself or a member of her cabinet!) takes place for conducting information, the Commission must record such an interview in a way that enables it “to provide an indication of the content of the discussions which took place during the interview, in particular the nature of the information provided during the interview on the subjects addressed (para 912). The Court would find it “useful and appropriate for the record of each interview […] to be made or approved at the time when that interview is held or shortly afterwards” (para 933). This is once more a reminder of the Commission’s procedural duties. Ultimately, the failure to fulfil those duties did not impact the decision because Google did not establish that absent the procedural errors it would have been better able to ensure its defence (para 939).

Google also contended that it was refused its right to a hearing regarding the application of the AEC test. The Commission had submitted to Google a letter of facts which, according to the Court, substantially amended its initial approach eight months after the statement of objections (paras 981 ff.). The letter (and some subsequent letters) included significant adjustments to the analysis of the AEC test. It did not grant an oral hearing on the observations submitted in response to those two letters and thereby, “circumvented Google’s right to be able to develop its arguments on those observations orally, and infringed Google’s rights of defence” (para 997). This was enough to partly annul the decision. By coincidence, this only affected the part of the decision concerned with the revenue sharing agreements – which anyway was annulled on substantive grounds as well.


The fine

Although the Court annulled the Commission’s decision, insofar as it considered the abusiveness of the revenue sharing agreements, it decided to relieve Google only from a very small fraction of the fine. The Court reminds the reader that it has unlimited jurisdiction to rule over the fine and that “it is not necessary to draw automatic inferences from [the] partial annulment” (para 1031). In other words: the Court could have also upheld the entire fine, despite the partial annulment of the Commission’s decision. It allowed that the revenue sharing agreements, despite not constituting a distinct abuse, were considered “as elements of the factual context or the purpose of assessing the exclusionary effects of the first and second aspects of the single and continuous infringement” (para 1018). One has thus still to look at the bigger picture of the interconnected conducts and markets. This is what the Court did, emphasizing that Google pursued a “carrot-and-stick” strategy (the ‘smoking gun’ used in an internal Google presentation) that included a range of practices (para 1045).

The Court followed the Commission’s methodology to first identify a basic amount based on the value of sales in relation to the infringement that can then be further adjusted (para 1054). Google challenged here that the Commission considered the value of sales in 2017, which was the last full year of the infringement and not the average per year sales over that period. The Court deemed the last year the appropriate benchmark because in this year “Google was able to reap all the economic benefits of the practices it had implemented since 2011” (para 1059). Certainly, a compelling argument: Google can’t claim that another year should be the benchmark for the fine if the higher revenues obtained on this year resulted from the abusive practices. The Court also followed the Commission’s argument that the relevant value of sales should include revenues generated by the Google general search website rather than only via pre-installed Google apps (paras 1061 ff). Because dominance in general search is the ultimate starting and end point of the whole case, this appears to be reasonable. The Court only disagreed with the Commission regarding a fixed gravity coefficient of 11 % because it regarded the gravity of the infringement as varying over time (paras 1081 ff.). Ultimately, the Court set the fine at 4 125 000 000 EUR. No reduction was granted for the “novelty” of the conduct (paras 1103 ff.). The Court considers this fine sufficiently deterrent even for an undertaking the size of Google (paras 1109 f.).


What to make out of the judgment?

As the reader might have already noticed, I find the Google Android judgment somewhat less convincing than Google Shopping. I believe, however, that the case has nevertheless a lot of merits. The key to understanding the judgment is not to look at the practices in isolation but to consider them as part of a bigger strategy with combined elements of leveraging market power and elements of building a competitive moat. One can see throughout the judgment that the Court has a vision of how a world with a less dominant Google could look like: more privacy, adaptation to particular linguistic features and more focus on value-added content (paras 65, 578, 1028). Open-source products, such as Android, on the other hand, seem to be less of a concern to the Court. Of course, just because something is open-source (or for free for that matter) does not mean that it cannot be anti-competitive or used for anti-competitive purposes.

If the anti-competitive nature of the practices can only be explained as part of a bigger strategy, however, it appears questionable that different standards are applied. On the one hand, there is the rather formalistic application of the Microsoft case law when it comes to tying and bundling and then the strict application of the AEC test where the Court gets really into the technical details of how this test can be applied. This is not entirely consistent. Whether an as efficient competitor could have offset the advantage imposed on it by the revenue sharing agreements tells us very little about the anti-competitiveness of the combined conduct. Even if the AEC could offset this particular disadvantage, would it also have been able to offset the additional disadvantage by the other practices?

In this regard, it should be recalled that the Court still did not find that the Commission was required to conduct an AEC test. So far, the Commission has not fared very well when it conducted such an AEC test (see Intel). One cannot avoid asking the question of why the Commission still does it. Perhaps it fears that the Court of Justice applies a stricter standard in this regard? The General Court seems to recognize that economic quantifications can be useful tools but at the same time also accepts their limits. Just because we cannot properly quantify something does not mean that it cannot be anti-competitive. From a legal standpoint, this should be a no-brainer. It is nevertheless important to keep it in mind.


To make sure you do not miss out on regular updates from the Kluwer Competition Law Blog, please subscribe here.

Kluwer Competition Law

The 2022 Future Ready Lawyer survey showed that 79% of lawyers are coping with increased volume & complexity of information. Kluwer Competition Law enables you to make more informed decisions, more quickly from every preferred location. Are you, as a competition lawyer, ready for the future?

Learn how Kluwer Competition Law can support you.

Kluwer Competition Law
This page as PDF

Leave a Reply

Your email address will not be published. Required fields are marked *