European Commission publishes Competition Policy Brief on generative AI and virtual worlds

On 19 September 2024, the European Commission published a Policy Brief on competition in generative AI and virtual worlds.  The Policy Brief explores emerging market trends and perceived risks to competition in these sectors.  It also examines the tools in the Commission’s armoury to tackle potential concerns.  Its publication follows the Commission’s related call for contributions in January 2024, as well as workshops and interviews held with stakeholders over the summer. 

The Policy Brief represents the views of the staff of the Commission, rather than the official position of the Commission itself.  However, it provides valuable insights into the Commission’s current thinking on generative AI and virtual worlds, and re-affirms the Commission’s commitment to keeping these sectors competitive and contestable.  

Key findings on emerging market dynamics 

The Policy Brief identifies several recent market trends characterising AI-related markets. These include, in particular:

• A tendency towards firm interconnectedness, resulting from increased vertical integration, an increasing number of partnerships, investments and strategic agreements between firms operating at different levels, as well as strategic hires of key talent.  The Policy Brief acknowledges the benefits of these types of partnerships, highlighting how they allow smaller players to enter and grow within the market by providing them with access to key input resources and distribution channels.  However, the authors also question whether some of these partnerships may lead to the concentration of key resources in the hands of a few major players – noting that some of these partnerships do not always fall within merger control rules. 
• A tendency towards more efficient, smaller models, particularly those that can operate efficiently on small mobile devices and with limited internet connectivity.  This trend challenges the notion that only AI developers with significant scale can be competitive.

The Policy Brief emphasises the importance of data, AI accelerator chips, computing infrastructure, cloud capacity and highly specialised technical expertise as key components for the successful development and deployment of generative AI systems.  According to the authors, each of these key inputs could pose significant barriers to entry or expansion for market players.  Other notable barriers include the high marginal costs related to the operation of and pre-training of AI foundation models, network effects and data feedback loops.  These findings are in line with those of the UK Competition and Markets Authority’s recent review of AI foundation models. 

‘Theories of harm’ and tools for competition enforcement

To assess risks to competition in AI markets, the Policy Brief distinguishes between upstream and downstream markets.  Upstream markets may involve the purchase of necessary inputs by AI developers, such as cloud capacity from service providers, with potential segmentation and product differentiation (for example, generalist cloud compute versus specialised cloud services).  Downstream markets may involve the sale or supply of generative AI services, with further segmentation and differentiation between consumer-facing and business models, cloud-based and on-device models.

The authors of the Policy Brief note that revenue streams are not necessarily a proxy for market power or a useful metric in these markets, given the nascent state of the industry where AI foundation models are not yet fully monetised.  Proposed alternative metrics to estimate market power include activity volume, cost and power of processing capacity used for inferencing, and the number of purchases or downloads of the model.  These alternative methods echo the Commission’s recent revisions to its guidance on market definition, which was updated earlier this year to better reflect the increased digitalisation of the economy (see our previous blog).

The Policy Brief lays out five key ‘theories of harm’, that is, reasons why certain market features or behaviour by market participants may lead to consumer harm and breach competition law: 

• incumbent, large digital players granting exclusive access to key components or preventing competitors from accessing them;
• large players using their market power to limit choice or distort competition in downstream markets when distributing and commercialising AI applications;
• agreements between competitors that reduce competitive constraints or enable the exchange of competitively sensitive information;
• vertically integrated players adopting pricing policies aimed at margin squeezing other competitors; and
• ‘killer’ or ‘reverse killer’ acquisitions of nascent competitors. 

The Policy Brief suggests that markets and competitive harm should be assessed holistically to capture ecosystem dynamics and market realities in the digital age. 

To address these risks, the Policy Brief proposes a combination of robust antitrust, merger control and Digital Markets Act (DMA) enforcement: 

• As regards alleged ‘killer’ acquisitions, to the extent that such transactions do not meet the turnover thresholds for review under the EU Merger Regulation, the Commission will “work together with Member States and the parties to assess whether they will be reviewed under national merger control regimes or referred to the European Commission in line with the legal requirements for such referrals”, as recently clarified by the EU Court of Justice (see our detailed analysis in our Competition newsletter).  Recent reforms allowing national authorities to call-in below threshold deals in strategic sectors, and therefore refer deals for review to the Commission, could create significant uncertainty for merging parties active in the AI space.
• The DMA will be relevant for generative AI services in two ways.  First, a generative AI player may offer a core platform service and meet the DMA’s ‘gatekeeper’ requirements.  Second, generative AI-powered functionalities may be integrated or embedded in existing designated core platform services and fall under the DMA obligations. 

The Policy Brief identifies similar competition concerns in relation to ‘virtual worlds’, and outlines barriers to entry and theories of harm resembling those identified in respect of generative AI.  An additional concern raised is the risk of ‘kill zones’, where the presence of dominant ecosystems could discourage other companies from investing or competing in this space.

Comment

The Commission is expected to continue actively monitoring these industries, keeping AI technology and its emerging use cases as a top policy priority.  With the recent nomination of Teresa Ribera for the post of EU Competition Commissioner, it remains to be seen how the Commission’s new leadership will tackle these challenges (see our recent Competition newsletter). 

The direction of EU competition policy is expected to be highly influenced by the recent report by Mario Draghi on the future of European competitiveness, in which AI features prominently.  One of Draghi’s most striking EU merger control proposals, which is expected to be particularly relevant in the AI space, is the introduction of an ‘innovation defence’, allowing transactions to be cleared on the basis that they would lead to increased innovation.  The report points out that, in certain sectors, pooling resources is essential to cover high fixed costs and achieve the necessary scale to compete globally.  In this context, innovation could be viewed as an efficiency that outweighs potential negative impacts on competition.  It will be interesting to see whether the new Commission takes this proposal on board, and if so, how this will affect M&A deals in the sector.