Recent European Commission merger control practice suggests a renewed willingness to scrutinise conglomerate mergers. This is in stark contrast to the Commission’s past practice and to the US approach. In this article, we detail exactly how conglomerate mergers generate customer benefits, but also—under certain circumstances—how they may end up harming competition and consumers. We apply theory to practice by discussing the recent NVIDIA/Mellanox merger, and explore which novel theories the Commission may pursue in the future.
The European Commission’s recent merger control practice suggests a renewed willingness of the Directorate-General for Competition (‘DG Competition’) to scrutinise conglomerate mergers, and to intervene on the grounds of conglomerate concerns.
In particular, DG Competition imposed behavioural commitments to address conglomerate concerns raised in relation to seven merger cases between 2016 and 2021. The Commission approved Densply/Sirona, Worldline/Equens/Paysquare, Microsoft/LinkedIn and Broadcom/Brocade in Phase 1, subject to behavioural commitments. In Qualcomm/NXP, Telia/Bonnier and Google/Fitbit, the Commission confirmed its conglomerate concerns following lengthy Phase 2 investigations, and the parties only secured clearance after addressing these concerns with behavioural commitments.
The Commission also looked closely at potential conglomerate effects in relation to many other transactions before eventually dropping these concerns, sometimes only after a Phase 2 review (Bayer/Monsanto and Essilor/Luxottica).
This practice is in stark contrast to the Commission’s approach in 2005–15, after the European courts overruled the Commission’s prohibition Decisions in GE/Honeywell and Tetra Laval/Sidel on the grounds of conglomerate effects.. It also contrasts with the less interventionist approach of US authorities towards conglomerate mergers.
So is the Commission right to look closely into conglomerate mergers and to intervene where it deems necessary? Or is the Commission overshooting the mark?
The current economic thinking is that, as is the case for vertical mergers, conglomerate mergers tend to bring significant benefits to customers. However, and in contrast to some previous views on conglomerate mergers, they can—under certain circumstances—also harm competition and consumers.
In this article, we first explain what conglomerate mergers are and how they compare with vertical mergers. We then explain why conglomerate mergers are often thought to bring significant benefits to customers, but also how, and under which circumstances, conglomerate mergers may harm competition and consumers. Finally, we apply theory to practice by discussing the recent NVIDIA/Mellanox merger review by the Commission and, using this case, reflecting on potential novel theories of harm concerning conglomerate mergers.
What is a conglomerate merger?
The European non-horizontal merger guidelines (‘NHMG’) define conglomerate mergers as ‘mergers between firms that are in a relationship which is neither horizontal (as competitors in the same relevant market) nor vertical (as suppliers or customers)’.1
The guidelines explain that in practice, the focus of the Commission is on conglomerate mergers between firms that are active in closely related markets, which the Commission defines as either of the following.
- Markets for complementary products—i.e. products that are consumed together and whose demands increase jointly. More specifically, product A (e.g. skis) and product B (e.g. ski boots) are complements if an increase in the price of A (skis) leads to a reduction not only in the demand for product A (skis) but also in the demand for product B (ski boots). In other words, the cross-price elasticity of complement products is negative—whereas it is positive for substitute products.
- Products that belong to a range of products that are generally purchased by the same set of customers for the same end use, but which are not strictly speaking complementary in the economic sense. These products can be independent, but also weak substitutes—i.e. substitution is not strong enough to include them in the same relevant market (for instance, glasses and contact lenses).
Conglomerate mergers may benefit customers in various ways
Conglomerate mergers may generate various types of efficiencies that benefit customers.
First, conglomerate mergers generally allow savings in transaction costs for customers, enabling them to purchase several products from a single supplier even absent any bundling by the merged entity—so-called ‘one-stop-shopping’.
Second, when conglomerate mergers involve complementary products, they may lead to lower prices. This is due to the so-called ‘Cournot effect’, which is the equivalent of the elimination of double marginalisation in vertical mergers.2
When two companies offering complementary products set their prices independently, they fail to take into account that by reducing the price of their own product, demand for the other product also increases. If the two companies merge, this effect is internalised, and the merged entity may have an incentive to decrease the price of each product and/or a bundled offer containing both products.
Internalisation of externalities
Third, the logic of internalisation of externalities in the Cournot effect can be extended to other aspects of the supply of complementary products.
For instance, a merged entity selling complementary products may have an incentive to increase its investments in R&D in view of increasing product quality: a higher-quality product will also lead to more demand for the complement product.
Conglomerate mergers can also harm competition and consumers under specific circumstances
While conglomerate mergers can bring significant benefits to customers, they can also—under specific circumstances—harm competition and consumers. The main concern is that a conglomerate merger may enable the merged entity to engage in exclusionary practices. However, according to the Chicago School, conglomerate mergers (as well as vertical mergers) are very unlikely to lead to anticompetitive foreclosure—as discussed below.
The single monopoly profit theory and subsequent thinking
Chicago economists formalised their reasoning in the so-called ‘Single Monopoly Profit’ (SMP) theory.3 Intuitively, a monopolist in market A (skis) would have no incentive to leverage its market power from market A into a perfectly competitive market B (ski boots). This is because the monopolist already extracts the full monopoly rents to be made on the bundle by charging a monopoly price on product A. According to the SMP theory, this monopolist would not be able to extract more profit by gaining market power in market B: the most profitable price for the bundle is already the sum of the monopoly price for product A and the perfectly competitive price for product B. Increasing its sales of product B would not bring any additional profits at current prices, and raising the price of B would be counter-productive, as this would lead to a price for the bundle that is higher than the monopoly price, and would therefore decrease the monopolist’s overall profit.4
As a consequence, the Chicago School concluded that conglomerate mergers were driven by the willingness of firms to exploit the efficiencies that we presented above. However, the Chicago School models underpinning the SMP rely on strong assumptions. For instance, this theory no longer holds when the tied good market (market B) is not perfectly competitive before the merger.
In the 1990s, economists started studying the conditions under which conglomerate mergers may lead to anticompetitive foreclosure by means of bundling or tying. The Commission’s Non-Horizontal Merger Guidelines (NHMG) are largely inspired by the results of this post-Chicago School economics literature.
The focus of the NHMG is on the risk of foreclosure.5 The concern is that by combining the supply of two products in related markets under the same roof, the merged entity may have the ability and incentive to leverage a strong position in one market to the other market by means of bundling or tying practices. These two types of practices are described in the box below. Such leveraging may then impact the ability and incentive of rivals to compete effectively, eventually allowing the merged entity to increase prices (and/or reduce innovation or quality).
What are bundling and tying practices?
Bundling and tying are two practices where a firm combines and jointly supplies products in related markets.
Pure bundling is a practice whereby two products are only sold together, in fixed proportions. With mixed bundling, the two products are still available separately, but it is cheaper to buy the bundle than the standalone products separately. An example of mixed bundling is the ‘triple-play’ package offered by telecoms companies, combining fixed telephony, broadband internet and TV.
‘Tying’ refers to a situation where customers that purchase one good (the tying good) from a producer are also required to purchase another good from the same producer (the tied good). It may be an outright obligation or an obligation focused only on customers who actually need both products. In the latter case, a customer would still be allowed to buy the tying good alone, but if they also need the other good, they would be required to buy it from the same producer.
Tying is contractual when the buyer is committed by contract to buy only the tied product (and not the alternatives offered by competitors). Tying can also be technical, where the product is designed such that it works only with the tied product and not with alternatives offered by competitors.
Source: Official Journal of the European Union (2008), ‘Guidelines on the assessment of non-horizontal mergers under the Council Regulation on the control of concentrations between undertakings’, footnote 1, para. 96; Niels, G., Jenkins, H. and Kavanagh, J., (2016), Economics for Competition Lawyers, second ed., Oxford University Press, para. 2.141; Official Journal of the European Union (2008), op. cit., footnote 1, para. 97.
The post-Chicago School economics literature has identified certain circumstances under which conglomerate mergers may harm competition.6
Ability and effect…
In brief, a conglomerate merger may lead to the anticompetitive foreclosure of rivals if the following three conditions hold. These are partly reflected in the ‘ability’ and ‘effect’ sections of the NHMG’s ability-incentive-effect framework of analysis.
- First, the merged entity must have significant market power in at least one of the closely related markets. Without significant market power in one of the two markets, the merged entity would not be in a position to leverage its position in one market to affect the outcome of the second market. According to the NHMG, ‘one of the merging parties’ products [must be] viewed by many customers as particularly important and there [should be] few relevant alternatives for that product’.7
- Second, the merging parties must have a considerable overlap in their respective customer bases. If a large portion of customers in market B does not buy product A, then the merged entity will not be able to leverage its strong position in product A to significantly foreclose rivals in market B. These rivals would at least still be able to compete for customers in market B who do not need the merged entity’s product A. In the NHMG’s words, the merged entity may only have the ability to foreclose if ‘there is a large common pool of customers’.8
- Third, market conditions must be such that reduced sales prospects for rivals would limit rivals’ ability and/or incentive to compete. As recognised by the NHMG, this could be the case because of the presence of significant economies of scale or network effects.9
Moreover, in the case of pure bundling or tying, the merged entity may not have the ability to foreclose its rivals if it does not have the ability to commit to a lasting bundling or tying practice. If it cannot commit to not selling the products on a standalone basis, the merged entity—faced with a customer determined to buy product B from a rival (and therefore willing to forego product A from the merged entity)—would have an incentive to sell its high-margin product A on a standalone basis and allow the mix-and-match option. This may be anticipated by those customers preferring the mix-and-match option, such that the merged entity would be unable to leverage its position in product A to influence the choice of the supplier for product B in its favour.
The NHMG acknowledges this commitment problem and explains that one way to make the bundling/tying strategy a lasting one would be ‘through technical tying or bundling which is costly to reverse’.10 Technical tying and degradation of interoperability was indeed at the core of the conglomerate concerns that the Commission used to justify its interventions in Microsoft/LinkedIn, Broadcom/Brocade, Qualcomm/NXP, and Google/Fitbit.
…but no incentive?
The above conditions and considerations relate to the ability of the merged entity to foreclose. However, it is not sufficient for the merged entity to have the ability to foreclose its rivals. Indeed, a firm may be trivially able to foreclose its rivals through bundling or tying, but it may not have the incentive to do so.
The importance of analysing the incentive to leverage market power and foreclose competitors is illustrated by the SMP theory introduced above. Importantly, even where market conditions are such that the SMP theory would not hold—which is most often the case—this does not mean that the merged entity would necessarily have an incentive to engage in bundling or tying.
More specifically, bundling and tying practices involve a trade-off, as they lead to lost sales on the leveraging product side (some consumers of the separate products will not buy the bundle) but also gained sales (and possibly increased margins) on the leveraged product side. The merged entity would only have an incentive to engage in bundling or tying if the gains on the leveraged product side outweigh the loss of profits on the leveraging product side.
Case study: NVIDIA/Mellanox (2019)
The Commission’s clearance Decision in NVIDIA/Mellanox is a good example of how the Commission applies its framework of analysis in practice. Even more interesting is the door left open by the Commission to intervene where the merged entity would not have the ability to significantly foreclose rivals—i.e. where a tying practice would not reduce rivals’ sales prospects sufficiently to hinder their ability or incentive to compete.
More specifically, the Commission assessed whether, in the absence of a realistic prospect of hampering rivals’ abilities and incentives to compete, the merged entity could still harm a subset of customers.
The NVIDIA/Mellanox merger involved two producers of complementary products:
- NVIDIA, the leading supplier of graphics processing units (GPUs);
- Mellanox, a leading supplier of network interconnect solutions for data centres.
GPUs and network interconnect solutions are complementary components purchased by data centre customers such as HPE, Dell, AWS, and others to build data centres. NVIDIA was found to have a market share of more than 90% on a relevant market for discrete data centre GPUs, where it was facing emerging competition from Intel and AMD.11 On the other hand, Mellanox was found to have a market share between 60% and 70% in two relevant markets: a high-performance fabric market and a market for Ethernet network interface cards (NICs) of at least 25Gb/s.12
The Commission focused its assessment primarily on the risk that the merged entity would have the ability and incentive to leverage Mellanox’s potentially strong market position in the high-performance fabric market (with its InfiniBand fabric), or in the market for Ethernet NICs of at least 25Gb/s, into the discrete data centre GPU market. The Commission explored whether this would have a significant detrimental effect on competition in the discrete data centre GPU market, and therefore harm customers.13
Ability: beyond market shares
On the ability assessment, two main lessons can be drawn from the Commission’s Decision.
First, high market shares (e.g. 60–70%)14 are not sufficient for the Commission to conclude that the merged entity would have a sufficient degree of market power to leverage its position in one market to influence customers’ choice in the related market. The Commission also needs to show that customers do not have access to sufficient credible alternatives.
The Commission only found that the merged entity would have sufficient market power in relation to the high-performance fabric market. This is because on the basis of the parties’ internal documents and responses to its market investigation, the Commission confirmed that Mellanox’s InfiniBand fabric had a considerable edge over its competitors in terms of performance—i.e. competing products were not considered to be credible alternatives to which Mellanox’s customers could switch. This contrasts with the Commission’s finding with respect to Ethernet NICs of at least 25Gb/s.
Second, the Commission’s ability test is not limited to the ability to leverage (i.e. to sell more products in one market by bundling with a must-have product in another), but should also assess the merged entity’s ability to foreclose rivals—i.e. whether a bundling or tying strategy would reduce rivals’ sales prospects to such an extent that their ability and/or incentive to compete would be reduced. This is one part of the Commission’s assessment where economic analysis played a crucial role in this case. Two elements fed into this analysis:
- the extent of the overlap between NVIDIA’s and Mellanox’s customer bases;
- the importance of economies of scale in the supply of discrete data centre GPUs by NVIDIA.
On the first point, the Commission concluded that the vast majority (somewhere between 70% and 100%) of NVIDIA’s discrete data centre GPUs sales are used to equip data centres that do not use Mellanox’s InfiniBand fabric. This meant that if the merged entity imposed the purchase of NVIDIA GPUs to all customers of its InfiniBand fabric needing GPUs, rivals would (in the worst case) still be able to compete to supply somewhere between 70% and 100% of GPU customers—i.e. those who want a GPU but do not want a Mellanox InfiniBand fabric.
On the second point, the question was the following. Considering the significant economies of scale (because of, for instance, high fixed R&D costs), the Commission had to determine whether having access to ‘only’ 70–100% of the market for GPUs would have been sufficient for AMD and Intel to operate and compete effectively with NVIDIA, or whether this would have implied operating below their minimum viable scale.
The Commission concluded that even if they were foreclosed from the segment of the GPU market linked to Mellanox’s InfiniBand fabric, AMD and Intel would still have sufficient sales prospects to reach their minimum viable scale. This was particularly true given that AMD’s own documents predicted a total addressable market by 2021 that would be several times higher than in 2018.15
Overall, the Commission therefore concluded that the merged entity would not have the ability ‘to hinder Intel’s effective long-term entry and AMD’s expansion into the discrete datacentre GPU market’.16
Beyond anticompetitive foreclosure: the potential for an exploitation theory of harm?
The Commission could have left it there, in line with the NHMG. However, it also assessed whether, in the absence of a realistic prospect of hampering rivals’ abilities and incentives to compete, the merged entity could still harm the subset of customers who buy GPUs in order to equip a data centre for which they also need Mellanox’s InfiniBand fabric.17 The potential concern for this subset of customers was that they would be ‘forced’ by the merged entity to use NVIDIA GPUs (reduction of choice)and that NVIDIA would then be able to raise its price.
The Commission suggested that the merged entity may not have an incentive to ‘force’ customers to use NVIDIA GPUs by way of tying or bundling practices because of the SMP theory described above.18 However, as the discrete data centre GPU market was not perfectly competitive before the merger, the SMP predictions are not guaranteed. Therefore, the Commission also carried out a proper incentive assessment.
Incentive: beyond the static trade-off between lost and gained sales
The NVIDIA/Mellanox Decision shows that the incentive assessment cannot be limited to a static assessment of the trade-off between lost sales for the merged entity on the leveraging product side and gained sales (and possibly increased margins) on the leveraged product side. Such an assessment should also consider counter-strategies and possible retaliation by competitors and customers. In NVIDIA/Mellanox in particular, the Commission took into account the possible retaliation by Intel and AMD.
From a purely static perspective, as the margins on the GPU side were several times higher than the margins on the InfiniBand fabric side, the Commission considered it likely that in case of tying, the gains on the GPU side would have outweighed the loss of profits on the InfiniBand fabric side.19
However, this was without considering any reaction from NVIDIA’s main competitors—i.e. Intel and AMD. Importantly, Intel and AMD were the two leading suppliers of central processing units (CPUs), a third type of component present in every single data centre.
In essence, the Commission considered that Intel and AMD would react to the merged entity’s tying strategy by degrading the interoperability of their respective CPUs with NVIDIA’s GPUs and Mellanox network interconnect. According to the Commission, the retaliation threat was so strong that the merged entity would not have taken the risk to upset Intel and AMD. This is because CPUs are at the heart of every server.
The Commission found that other components, including GPUs and NICs, need to interoperate seamlessly with the CPUs in order to be valuable for data centre customers—and this interoperability was controlled by the CPU suppliers. The Commission considered that such threat of retaliation would have eliminated all incentives to engage in tying practices, given that ‘the vast majority [of] NVDIA’s GPUs and Mellanox’s network interconnect products sales depend on being interoperable with Intel’s and AMD’s CPUs’.20
Novel theories of harm in conglomerate mergers
In line with its NHMG, the Commission’s focus so far has been on exclusionary theories of harm, whereby customers may be harmed as a result of competitors being foreclosed. However, in NVIDIA/Mellanox, the Commission suggested that it may also consider intervening in the absence of a realistic prospect of hampering rivals’ abilities and incentives to compete, if the merged entity would be in a position to exploit a subset of customers.
The Commission eventually dropped its concerns in that case because the merged entity would not have had an incentive to engage in bundling or tying practices, in fear of retaliation by rivals. But the Commission’s Decision does leave the door open to theories of harm that go beyond those covered by the NHMG—and particularly exploitative theories of harm.
The economics literature has already identified various possible exploitative theories of harm, whereby customers may be harmed without foreclosure. This includes the recent portfolio effect theory of harm developed by Chen and Rey (2021).21 According to these authors, consumption synergies as a result of the ‘one-stop-shopping’ efficiency presented above may increase product differentiation between the merged entity and single-product rivals. Under specific circumstances, this may soften competition in a way that decreases consumer surplus.
The economics literature has also highlighted a theory of harm by which a conglomerate merger could lead to increased bargaining power as a result of the strengthening of a product portfolio.22 Interestingly, the Commission imposed remedies in Qualcomm/NXP specifically to address a concern by which the merged entity, by combining complementary IP portfolios, would have improved its bargaining position and therefore its ability to raise royalties.23
It remains to be seen whether the Commission will further explore alternative theories of harm, or instead stick to its guidelines, focusing on cases where rivals are likely to be foreclosed.
1 Official Journal of the European Union (2008), ‘Guidelines on the assessment of non-horizontal mergers under the Council Regulation on the control of concentrations between undertakings’, para. 5.
2 Ibid., footnote 1, para. 117.
3 See Bork, R.H. (1978), The Antitrust Paradox, Free Press, New York, pp. 378–9.
4 The monopoly price is profit-maximising for a monopolist. This means that any further increase would trigger a loss of sales that would outweigh the increased margin on the retained sales.
5 To be complete, the NHMG also discuss the risk of coordinated effects; but in practice, the Commission has never raised serious doubts on this basis.
6 For a summary of the literature on the topic, see Church, J. (2004), ‘The impact of vertical and conglomerate mergers on competition’, prepared for the European Commission, section 4.5.
7 Official Journal of the European Union (2008), op. cit., footnote 1, para. 99.
8 Ibid., footnote 1, para. 100.
9 Ibid., footnote 1, para. 101.
10 Ibid., footnote 1, para. 102.
11 AMD already had a 5–10% market share, while Intel still had to enter the market, with entry expected in 2021. Its entry and future expansion was found to be credible, as it won a tender to supply discrete data centre GPUs for one of the fastest data centre in the world.
12 GPUs are specialised semiconductor devices that are optimised for processing graphic images. In data centres, GPUs are used to accelerate the data centre workload computing/processing. They are necessarily used in complement to central processing units (CPUs), which are always present in data centre servers. A discrete GPU is a GPU that is separate from the CPU. In contrast, integrated GPUs share system memory with the processor. Network interconnect solutions facilitate efficient data transmission between servers, storage systems and communications infrastructure equipment within data centres. They are composed of NIC, switches, routers, cables and related software. ‘Ethernet NICs’ refers to NICs that use the Ethernet communication protocol. High-performance fabrics are integrated systems of custom hardware including NICs, switches, and cabling, which have particularly low latency and high bandwidth. They are designed to run on custom protocols (such as InfiniBand) and orchestrated by custom software.
13 The reverse theory of harm whereby the merged entity would leverage NVIDIA’s strong market position in the discrete data centre GPU market into any possible network interconnect markets did not get much traction, as the Commission concluded that despite NVIDIA’s market share being above 90% in discrete data centre GPUs, AMD’s and Intel’s GPUs offered suitable alternatives to which NVIDIA’s customers could turn.
14 And even more than 90%, as the Commission concluded that NVIDIA, despite its more than 90% market share in the market for discrete data centre GPUs, ‘would not have sufficient market power to leverage its position in the market for discrete data centre GPUs into the markets for network interconnects’. See Commission Decision of 19 December 2019 in Case M.9424 – NVIDIA/Mellanox, para. 264.
15 See Commission Decision of 19 December 2019 in Case M.9424 – NVIDIA/Mellanox, para. 218.
16 Ibid., para. 219.
17 Ibid., footnote 232.
19 Ibid., para. 226.
20 See Commission Decision of 19 December 2019 in Case M.9424 – NVIDIA/Mellanox, para. 242.
21 Chen, Z. and Rey, P. (2021), ‘A Theory of Conglomerate Mergers’, mimeo.
22 OECD (2020), ‘Roundtable on Conglomerate Effects of Mergers – Background Note’, 24 May; Spulber, D.F. (2017), ‘Complementary monopolies and bargaining’, The Journal of Law and Economics, 60:1, pp. 29–74.
23 See Commission Decision of 18 January 2018 in Case M.8306 – Qualcomm/NXP Semiconductors, section 7.5.