Should we break-up Big Tech?

In recent years, digital technologies have profoundly changed many aspects of our daily lives, from e-commerce to internet search, travel, communication or entertainment consumption. While for the most part these changes have benefited consumers, certain voices have started to speak up against the power and influence of the Big Tech companies – Google, Amazon, Facebook, Apple in particular, accusing them of stifling innovation, dealing unfairly with their suppliers, and violating our privacy among others. Elizabeth Warren, one of the most prominent candidates to the Democratic investiture in the U.S., recently called for a much tougher policy approach towards Big Tech, proposing in particular to dismantle some of these companies, a call that has received a certain echo in the press and among politicians.

To understand whether we should break-up – some of – the big tech companies, it is important to understand why they have become so big, whether such a situation is actually harmful to consumers, and whether a break-up is an appropriate solution.


Many digital markets are characterised by the existence of economies of scale and of network effects (see Shapiro, Carl and Varian, 1998). The former corresponds to the idea that the average cost goes down with the number of units sold, which is typical of information goods: their production entails a large fixed cost, but they can be reproduced for a small marginal cost. For instance, once a search engine algorithm has been developed – at a considerable cost, answering an individual query is virtually costless.

Network effects are the demand-side equivalent of economies of scale: a product is more valuable the more users it has. If a social network like Facebook is a natural example of direct network effects, other platforms may exhibit indirect network effects: Android users exert a positive externality on each other, not because communication is easier between Android devices, but rather because more Android users attract more application developers to the platform (see Caillaud, Bernard and Jullien, 2003).

The use of data by technology companies is a particularly important source of returns to scale and network effects: as firms get more data, they can offer better products or services, or produce them more cheaply. Big Data also allow firms to realise economies of scope, that is to enter new markets thanks to the insights generated on their primary market – having access to your email data allows to offer a better calendar app.

By giving an advantage to larger firms, economies of scale and network effects can result in market tipping, that is in one firm becoming dominant as a natural result of the competitive process. The perspective of monopoly is worrying, but two forces push in the opposite direction. First, while possible, tipping is not guaranteed even in the presence of network effects. When these effects are intermediate, they can even intensify competition, as the fight for additional users becomes more intense. Second, even when they lead to monopoly, network effects and economies of scale can induce firms to compete harder to be the early leader: competition for the market, rather than in the market.

Breaking-up a monopolist in such a market, by creating several smaller networks, could result in increased competition. For instance, competing social networks could be induced to offer better privacy protection in order to attract more consumers. But breaking-up a network results in the fragmentation of the market, with some groups of consumers being unable to interact with others. This could make consumers switch network in order to enjoy more interactions, and eventually lead back to market tipping, thereby undoing the break-up.

The big technology firms have not passively enjoyed the rents of their position of natural monopolists, but have instead used a variety of strategies to protect or extend it, some of which have been deemed anticompetitive. Google, for instance, has been fined three times by the European Commission. One set of practices consisted of imposing restrictive clauses – exclusivity, tying –  to its trading partners, thereby preventing its rivals from competing on the merits. For instance, a rival search engine would have had to develop its own application store – or to pay a lot of money – in order to convince a device manufacturer to choose it over Google – and its very popular app store Google Play (see De Cornière and Taylor, 2018).

Another practice consisted in systematically favoring Google Shopping at the expense of other comparison shopping services on Google’s search engine. This issue of “own-content bias” has taken a new dimension with the emergence of internet gatekeepers such as Google or Amazon, the latter having also been accused – but not yet fined – of favoring its own brands on its platform. Own-content bias may also take other forms, such as when Spotify is required to pay Apple a fee when consumers subscribe through iOS, which puts it at a disadvantage compared to Apple Music. Platforms leveraging their dominant position on complementary markets is a key motivation for the proponents of breaking-up these firms.


Despite these legitimate concerns over exclusionary practices by multiproduct incumbents, it is not clear that a break-up – say, separating the search and the shopping activities of Google – would be desirable. First, in the presence of complementary products, common ownership enables firms to better coordinate their production decisions and achieve superior outcomes, which is the reason why competition authorities view vertical mergers more favorably than horizontal ones. Second, being able to use the data acquired on their dominant market on another market gives these firms further incentives to improve their core product. Forcing, say, Amazon to divest its personal assistant business would probably marginally weaken its incentives to offer cheap products on its platform. Third, a break-up in itself would not be sufficient to ensure neutrality of the platform, since they could use other contracts with some of the participants ensuring preferential treatment in exchange for a commission, a common practice in many industries (see De Cornière and Taylor, forthcoming).

A more sensible course of action consists in monitoring more closely the behavior of dominant platforms, and to intervene more quickly. At the moment antitrust actions take too much time to be carried out, and by the time they are the markets have changed, usually to the detriment of smaller rivals. Several recent reports make related arguments,  advocating a more responsive competition policy or the creation of a sectoral regulator (see the UK report “Unlocking digital competition: report from the digital competition expert panel”, or Cremer, Montjoye and Schweitzer, 2019).

Tech giants have also been accused of using acquisitions to cement their market power, buying out the start-ups that could potentially represent a threat to their dominant position. The typical illustration of this phenomenon is Facebook, with its acquisitions of Instagram and WhatsApp – and failed bid for SnapChat.  Google and Amazon have also been very active acquiring start-ups: over the past ten years, these three firms have bought around 300 companies, often relatively young. Most of these acquisitions have not been reviewed by competition authorities because they do not meet the various turnover thresholds.

One concern is that some of these acquisitions are “killer acquisitions,” i.e. made only for the purpose of shutting down potential competition, a phenomenon recently studied in the pharmaceutical sector (see Cunningham et. al 2018). Things look different in the tech sector, as many of the targets offer products that are complementary to the incumbents, and the perspective of being bought out by a big firm is a strong incentive to innovate. At the same time, economies of scope might turn a firm that offers a complementary product today into a rival tomorrow, but it is hard to predict when this is the case.

In markets such as these, with young firms and rapidly evolving technologies, competition authorities are bound to make errors, either of type I – blocking a pro-competitive one – or type II – approving an anticompetitive merger. The current situation is very asymmetric, as none of the reviewed acquisitions by the Big Tech firms have been blocked. This is certainly suboptimal, especially given that the cost of a type II error, namely elimination of competition, is probably much larger than that of a type I error. While recognising that predicting the effects of a merger is especially difficult in innovative markets, moving the needle towards a stricter approach to mergers in the digital sector seems warranted.

As I tried to show in this brief essay, ensuring effective competition in the technological markets will require a more elaborate answer than a break-up, the efficacy of which is highly doubtful. Several approaches have been proposed, and the debate is still raging. These are exciting times to be an industrial economist!

By Alexandre de Corniere



Caillaud, Bernard, and Bruno Jullien. “Chicken & egg: Competition among intermediation service providers.” RAND Journal of Economics (2003): 309-328.

Crémer, Jacques, Yves-Alexandre de Montjoye and Heike Schweitzer, “Digital policy for the digital era”, 2019

Cunningham, Colleen, Florian Ederer, and Song Ma. “Killer acquisitions.” Working Paper (2018).

De Cornière, Alexandre and Greg Taylor. “Upstream Bundling and Leverage of Market Power”, CEPR working Paper, 2018

De Cornière, Alexandre and Greg Taylor. “A Model of Biased Intermediation”, Rand Journal of Economics, forthcoming

Shapiro, Carl, and Hal R. Varian. Information rules: a strategic guide to the network economy. Harvard Business Press, 1998.

UK report, “Unlocking digital competition: report from the digital competition expert panel”, 2019.

AI: How are companies preparing?

The article “Why Are We Paying Different Prices? Hint: Artificial Intelligence is Learning What You’re Willing to Pay” published on LinkedIn by Ted Gaubert, discusses how the progress of Artificial Intelligence (AI) pricing engines are allowing them to influence the functioning of markets. AI is defined as the ability of machines to mimic human behaviour so that they can solve problems. Because of our vast cognitive skills, there are many problems where human beings have traditionally performed relatively well. Developments in AI allowed it to be able to mimic such behaviour, and firms are testing AI pricing mechanisms’ readiness for implementation and trying to understand what kind of role it may have in modern markets.

There is no simple answer to this question. However, as the author argues, one thing that is clear is that AI’s solution to the revenue maximisation problem of the firm is better than auctions. It has become a very attractive way to pricing sales items for companies in the e-commerce sector. However, this new technique has become a big concern for regulatory authorities. In fact, auctions that have long been viewed as optimal mechanisms to maximise revenue are currently failing at getting the attention of most experienced e-commerce firms. So, does it mean that it is time to let AI pricing mechanisms take over, for instance, for the items sold in eBay?

Consider a hypothetical first-price auction on eBay of Jean Tirole’s book The Theory of Industrial Organisation. All online participants, including you, submit their bids and the individual with the highest bid gets the book. I am sure you know the result! You are a TSE student with a high valuation for it – we all are – so I suppose you won. However, I can certainly tell you that you did not pay what you are really willing to pay. Why? Because you are a rational individual who realises that bidding a little more than your rival, instead of bidding the value of your actual willingness to pay, will get you the book and you will go back home with some money left in your pockets.

Now, imagine a similar scenario. This time, instead of you personally going online, you bought an option on the platform beforehand, which allows eBay to participate on your behalf in this auction. Suppose eBay will use AI for this purpose. Well, if you are an active customer, the platform has data on your preferences and your willingness to pay for certain types of goods. The AI will start solving the revenue maximisation problem of the retailer, coming up, most probably, with the same result as previously (i.e. you as a winner). These are intelligent algorithms that know that your valuation is highest among the bidders and will therefore sell the book to you. But, guess what? You will not go back home with some money in your pockets! Why? Because the algorithm finds your true valuation of the book, but it does not bid tactically on your behalf. You had to pay a premium to eBay for not having to allocate time to participate in it.

In the end, the AI pricing mechanism does its job for both the seller and the buyer. It reveals your true willingness to pay for the book, creating revenue for the seller. Additionally, it helps you to avoid having to dedicate your time to participate in the purchase. This is the first reason why AI could currently be more attractive than auctions for experienced e-commerce companies such as eBay or Amazon.

Moreover, if one considers the case of airline operators, one can easily reach the same conclusion. As Gaubert (2017) writes, AI is able to “learn about local events that are happening in real time on a global basis far more economically that what could ever be achieved by a host of humans”. It comes up with an optimal solution for the revenue maximisation problem of the firm. In such a case, it may consist of setting high prices for flights going to a city where a big festival is taking place. Therefore, uninformed competitors serve individuals with low valuations, while the informed competitor serves the bucket of individuals with high valuations or the residual demand.

Let’s go back to our hypothetical auction for Jean Tirole’s book, and suppose that the book has a special value for bidders because it is the only book that has been personally signed by him. I guess you will win again, it is the signed copy after all! But, were you tempted to bid more than what you are really willing to pay? Let me answer this question for you, YES! It is the only one on earth that has been signed by him, which means that if you get it, you will be able to sell it for thousands of dollars in the future at the pawn shop. So, you started bidding and you could not stop, right? Congratulations, you are another example of what is called ‘escalating commitments’: you committed to get it and therefore you could not stop bidding.

Nevertheless, if you had allowed the platform to participate for you in this auction, eBay would have avoided this issue. How? Its AI pricing mechanism is good at estimating your willingness to pay because it constantly collects data on your behaviour. AI is so efficient that it could have made a good prediction of your behaviour when participating in an auction. Thus, when solving the revenue maximisation problem of the firm, AI would have stopped bidding on your behalf once it had realised that you would start behaving irrationally. Once again, why? Because it is at that point that revenue is maximised without giving you the disutility you would have received from your escalating commitments.

AI did an excellent job once more: it solved the problem under different conditions, which is the second reason to believe that it is very attractive for the e-commerce industry. Several other reasons can support this fact. For instance, naive consumers – unable to bargain – would also have chosen AI to act on their behalf, because even though they are naive they know that buying the book at a fixed price comes with a premium for the convenience of an immediate purchase, a higher price. This means that consumers also find it attractive.

Ultimately, the examples above and the ones provided by the author support the idea that an AI pricing machine is a very attractive mechanism for the e-commerce sector. It reveals consumers’ willingness to pay accurately, avoids escalating commitments, constantly collects asymmetric information (that is not available to competitors), attracts consumers, and maximises revenue. The concern it is imposing on regulatory authorities, if anything, is just a display of its supremacy and its potential in the near future.

by Maria Alejandra Arenas