Since late 2023, the CMA has opened five merger investigations in relation to AI partnerships (four of which came after the CMA’s update report into AI foundation models) – the cases will give important insights into the CMA’s merger policy, both in terms of the limits (or indeed, breadth) of the CMA’s jurisdiction as well as its approach to substantive assessment in (the most) nascent but fast-moving tech markets. In this post, we summarise the state of play on the five investigations, reflect on why they matter beyond AI, and highlight what we’ll be looking out for in the CMA’s Phase 1 decisions when they come.
What's AI the fuss about?
As part of its strategic review of AI foundation models (FMs), the CMA issued a report in April 2024, in which it outlined growing concerns regarding competition in the sector. The report highlighted the existence of an interconnected web of partnerships and strategic investments, including so-called ‘acqui-hires’ (which consist of acquiring a company for its staff expertise, or simply hiring the staff), between the large tech incumbents and FM developers, and queried whether these could exacerbate existing positions of market power through the value chain. These concerns were also articulated in the recent joint statement issued by the CMA, the European Commission and the US’s Federal Trade Commission and Department of Justice (as outlined in our recent blog post on the joint statement).
The AI case docket
The CMA has opened five investigations, three of which are ongoing. Each is interesting for its own reasons – we summarise their status below.
Cleared on 17 May 2024 on the basis that the CMA did not have jurisdiction (a ‘found not to qualify’ decision): The CMA found that Microsoft would have a shareholding in Mistral of less than 1% which, along with other factors, indicated that it would not have the ability to influence Mistral’s policy. In addition, the CMA found that Microsoft’s other agreements with Mistral were not sufficiently significant as to create a dependency on Microsoft, and thus an ability for Microsoft to influence Mistral’s policy.
The CMA found that it had jurisdiction over the merger, but that Inflection’s presence in the UK was limited and thus the merger did not give rise to competition concerns.
In determining that Microsoft had acquired an enterprise, the CMA placed considerable weight on the fact that the core Inflection team had transferred to Microsoft (bringing knowhow and expertise with it), combined with access to Inflection’s IP.
In pre-notification: The CMA issued an invitation to comment on 30 July which closed on 13 August 2024.
Beyond AI: why these cases matter
The fact that the CMA has chosen to open these investigations is revealing, and not just in the context of its interest in digital markets and AI models. In particular:
Many had speculated that the CMA might become less bold following the Microsoft/Activision saga. These cases make clear that the CMA’s desire to position itself at the forefront of merger control policy remains unwavering.
None of these partnerships look like a typical merger or acquisition – put at their highest, they could be said to be akin to a minority (from an EU perspective non-controlling) stake. The CMA’s jurisdiction is not bound by such technical rules, but instead by whether the transaction involves 'enterprises' which 'cease to be distinct'. The cases will collectively give significant insight into where the CMA views the bounds of its jurisdiction (albeit unless any decisions are challenged through the courts, this can only be said to reflect the CMA’s view of its own powers rather than an established legal position).
By the time we have the CMA decisions in all five cases, its approach to assessing dynamic competition will be clearly signalled. The CMA is being called upon to assess current and future competition in nascent markets. It is hard to imagine more fast-moving markets than AI, given developments over the past 12-18 months: the CMA’s AI update report found that 120 new foundation models were launched between September 2023 and April 2024. Against this backdrop, the CMA’s conclusions (and indeed the framework for analysis) will be hugely important.
What Platypus will be asking our AI assistant to watch out for
The CMA’s decisions in these cases promise some great fireside reading as the nights begin to draw in. Platypus will be looking particularly closely at how the CMA approaches:
Jurisdiction: while each partnership turns on its facts, the decisions will shed light on where the CMA draws the boundaries of its jurisdiction (both in terms of the concept of an enterprise and as to the low bar for material influence). Key issues will include what combination of staff and assets will be sufficient to fall within the merger rules. More generally, given the low (if any) shareholdings across all these partnerships, the cases will illustrate how to assess the impact of financing arrangements and other commercial arrangements (i.e. agreeing to use the acquirer’s cloud services) on the relationship between ‘acquirer’ and ‘target’, and how to take account of the incumbent acquirer’s experience/scale/significance in determining its influence over the target (building on existing precedent such as BskyB/ITV).
How the CMA moderates its competitive assessment to reflect the level of influence: assessing the impact of a merger which involves a partial transfer of a business (e.g. acqui-hires) or non-exclusive licensing arrangements is complex. The CMA may adopt a cautious approach in its interpretation of the impact of these arrangements in practice. In previous cases involving minority shareholdings, the extent of the 'material influence' has been relevant (as this provides an insight into how any competitive harm could plausibly be effected) – it will be very insightful to see how, if at all, this issue is dealt with here.
Dealing with uncertainty: the CMA’s merger assessment guidelines confirm that the existence of uncertainty regarding changes in future competitive conditions does not preclude or reduce the likelihood of a merger giving rise to competition concerns. Given the high bar for challenging the CMA’s decisions, the CMA has significant discretion in weighing evidence and assessing risk. With an almost free hand (at least in its Phase 1 decision), Platypus will be keeping a close eye on how interventionist the CMA is willing to be in an industry as nascent and fast moving as AI (and whether/how the incumbent’s identity is taken into account in this assessment).
The tipping point: unlike in the EU, the ‘first past the gate’ principle does not technically apply in the UK as a formal procedural safeguard (as discussed in our recent blog post on merger control risk in parallel transactions). However, one can imagine the CMA considering whether arrangements could become incrementally more concerning as their number increases – the CMA may need to grapple with the complex history of investments in these AI companies, and questions will be asked as to the cumulative effect of these partnerships.
The CMA’s remedial powers: finally and critically, in the event that the CMA ultimately finds both jurisdiction and competition concerns in any of these cases, it will need to consider what remedies may be appropriate to address its concerns. In relation to acqui-hires in particular, such remedies would likely be particularly complex, owing to the limit on the CMA’s ability to require individuals to change employers (although it has previously imposed work-arounds to incentivise individuals to do so, e.g. in Facebook/Giphy).
We have some initial insights from the two cases concluded so far (although for Microsoft/Inflection, only a summary of the CMA’s decision is currently available), noting that less than half the picture has been painted. First, while the CMA recognises that there are limits on its jurisdiction, especially in cases involving small investments, it will interpret jurisdiction purposefully where it considers that a ‘business activity’ has been transferred, regardless of the form of the arrangement (as we have seen in Microsoft/Inflection). Second, thus far the CMA has not found any competition problems, and the 'clean bill of health' seems anchored in the evidence gathered as part of its investigation (including in particular third party views; as highlighted by the CMA in a recent post on LinkedIn). Given the scale of opportunities offered by partnerships with big tech, we look forward to seeing how the pro-competitive elements of these partnerships are assessed by the CMA in its final decisions.
It remains to be seen whether future partnerships in the AI arena evolve ‘intelligently’ to an ongoing cat-and-mouse game with the CMA. Outside of local markets cases, it’s highly unusual to have this number of cases considering the same fast-evolving market in parallel. By the time all five decisions have been delivered, we hope and expect to have a rich understanding of how the CMA is considering these issues.
