Antitrust

On 15 April 2025, the UK Competition and Markets Authority (“CMA”) published its long-awaited decision on the Microsoft / OpenAI partnership, marking the end of a cycle of five merger investigations launched into partnerships between AI developers and big tech companies since December 2023. 

The Microsoft / OpenAI case potentially marks a turning point in the CMA’s application of its merger control powers, not just to AI partnerships but more broadly.  Having subjected the parties to a 15-month investigation only to determine it had no jurisdiction, the CMA is now facing legislative curbs on its “unusually broad” powers to review mergers, alongside a broader push to create greater business confidence in the UK merger regime.  

In this post we examine what the CMA’s review of AI partnerships to date, and the wider UK merger reform programme, may mean for UK antitrust scrutiny of the AI sector in future.   

How did we get here?

In May 2023 the CMA launched an initial review of AI models, discussing the evolution of the market, opportunities, risks and guiding competition principles.  It has since issued two update reports, including setting out its proposed AI principles to guide competitive markets in the space and outlined the perceived risks that the CMA sees to effective competition.

In April 2024, the CMA CEO Sarah Cardell gave a speech to the ABA Antitrust Law Spring Meeting where she warned that the CMA would step up its use of merger control to assess whether, and in what circumstances, AI partnerships fall within the merger rules and whether they raise competition concerns.  She was upfront that the CMA would use its merger control powers to build that understanding, flagging concerns that incumbent technology firms may have the ability and incentive to shape AI markets for their own interest for example by seeking to use partnerships and investments to quash competitive threats and leveraging already strong positions in key inputs such as compute, data and talent.  

Having opened its Microsoft / OpenAI investigation in December 2023, following the dismissal and reinstatement of OpenAI’s CEO, Sam Altman, the CMA then proceeded to open reviews into a further four AI partnership cases, namely Microsoft / Mistral AI, Microsoft / Inflection AI, Amazon / Anthropic and Alphabet / Anthropic. 

Jurisdictional clarification or the last staging post of an unpredictable era of merger control?

The CMA only has jurisdiction to investigate a “relevant merger situation”, which requires an actual or anticipated change of control of an “enterprise”.  Focussing on such a nascent sector has provided the CMA with opportunities to explore whether it can further extend its jurisdiction.  As the AI partnerships that the CMA has investigated are not typical mergers or acquisitions, the CMA’s discussion of its own jurisdictional limits provides a useful insight into how the CMA views its current powers.   

As we discuss below, however, the lessons drawn from these cases may be short lived, with the CMA’s jurisdictional powers set to be reviewed in the coming months. 

Material influence in the context of exclusive supply arrangements

The UK merger control regime recognises three levels of control, the lowest level that can give rise to a relevant merger situation being “material influence”. This threshold generally is designed to catch transactions where there is an ability to influence the commercial policy of an enterprise.  The CMA has been testing the limits of material influence over several years in the context of minority shareholdings, but the investigations into AI partnerships presented an opportunity to consider the role of contractual arrangements in providing material influence.

In Microsoft / Mistral AI, the CMA accepted that Microsoft does not have material influence over Mistral.  The decision nonetheless sets out the CMA’s framework for assessing this issue in the context of AI partnerships.  Of particular interest for future cases – including in the context of supply and distribution arrangements beyond technology – is the CMA’s analysis that an exclusive supply agreement that “locks in” an AI model developer, or includes other terms compromising the target’s “commercial freedom” such as how IP is commercialized, could be sufficient for a finding of material influence.  Similarly, a distribution agreement could result in an acquisition of material influence where that distribution agreement creates a dependency that enables the distributor to materially influence the commercial policy of the AI model developer. 

This approach to determining material influence was reiterated in both Amazon / Anthropic and Alphabet / Anthropic.  In the latter, the CMA followed the same framework as in Mistral and concentrated in particular on the non-exclusive nature of the compute supply agreements and distribution agreements in finding that Google does not have material influence over the commercial policy of Anthropic.

Acqui-hires as an “enterprise”

In order for a relevant merger situation to arise, two or more enterprises must cease to be distinct.  In Microsoft / Inflection AI, the CMA considered the definition of an “enterprise” in a situation where the co-founders of Inflection moved to Microsoft and were followed by the majority of its employees.  Whilst the CMA’s jurisdictional guidance states that a “transfer of … employees alone may be sufficient to constitute an enterprise”, ‘acqui-hires’ of this nature have not been considered by CMA in practice previously.

The CMA ultimately concluded that the transfer of employees, know-how and a non-exclusive IP license enabled Microsoft to substantively acquire Inflection’s pre-acquisition model and chatbot development capabilities. As a result of this, the CMA considered that there was “economic continuity” between at least part of the prior activities of Inflection and those of Microsoft following the hire of the employees.

This will not be the case for every employee transfer, and bare assets alone remain insufficient to constitute an enterprise, but it represents a significant expansion of the CMA’s application of its jurisdictional powers.  To date, the discussion around reform to the CMA’s jurisdictional thresholds has focused on the material influence and share of supply tests.  It therefore appears likely that the CMA will continue to view acqui-hires as potentially falling within its purview.

Partnerships and the 4-month jurisdictional clock – can the CMA turn back time?

The CMA has jurisdiction to review both anticipated and completed deals, whether or not proactively notified by the parties.  Significantly, the CMA retains power to refer a merger for an in-depth investigation until the later of (a) four months after completion of the merger or (b) four months from the date on which “material facts” about the merger entered the public domain.  Based on CMA guidance, “material facts” was understood to mean information about the parties and whether the transaction remains anticipated or has completed.

However, in 2024 the CMA investigated Google’s partnership with Anthropic, comprising a cloud partnership and investments in 2023, which were widely reported at the time.  The CMA’s initial enquiry letter was not sent until June 2024, well after the 4-month time period had elapsed.  In claiming that the 4-month time period had not expired, the CMA relied not on the cloud agreement or the investment agreements, but a side letter first executed in October 2022.  The CMA argued that as the most recent version of the side letter had been signed in August 2024, but not made public, this restarted the 4-month clock.   The CMA became aware of the side letter in the context of the Amazon / Anthropic merger investigation and relied on this as notice of “material facts”.  This is despite the fact, based on the publicly available decision, that the contents of the side letter appeared not to be material to the CMA’s decision.

This has potentially significant implications for partnerships and joint ventures more widely — even if basic facts regarding the relationship have been made public, details of a side letter or similar agreement will be sufficient to constitute “material facts” and (re)commence the 4-month clock for the CMA to commence an investigation. 

Substantive assessment recognises rapid pace of development in AI sector

The CMA’s focus on the AI sector to-date has likely been driven by fears of repeating the past, with antitrust enforcers accused of under-enforcing antitrust rules in some ‘Web 2.0’ markets.  However, as it becomes clearer that early fears that FM technology might cause markets to “tip” are starkly at odds with commercial reality, the CMA’s AI investigations highlight the challenges of identifying concrete competition concerns in rapidly evolving markets.  This is particularly the case where, as the CMA has rightly acknowledged, partnerships “can bring significant benefits to the parties involved and lead to increased innovation and efficiencies”.

The CMA found that it did not have jurisdiction in four out of the five cases it has investigated.  Microsoft / Inflection AI is, therefore, the only case in which the CMA has needed to reach conclusions on the competitive analysis.  The CMA focused on potential horizontal unilateral effects arising from the loss of competition in two markets: the development and supply of consumer chatbots and FMs.  Ultimately, the CMA found that Inflection was not a sufficiently strong competitor in either market to raise concerns.  Nevertheless, broader insights regarding the CMA’s approach can be gleaned from the decision:

• Given the rapidly changing nature of the market for FMs, and the significant degree of differentiation in the FM offerings of different developers, the CMA recognised that static shares of supply are of limited utility in assessing competitive dynamics.
• Instead, the CMA placed significant weight on third party views of Inflection’s strengths, as well as an assessment of Inflection’s innovation capabilities.  The CMA’s finding that “Inflection did not have any strong innovation capabilities compared to its competitors” was likely central to its clearance decision. 
• The presence of a “wide range of innovative competitors” was another key factor enabling the CMA to distinguish AI-related markets from the more cautious approach taken to dynamic competition concerns by the CMA, for example, in Meta / Giphy.  

The CMA’s Phase 1 clearance provides a welcome recognition of the highly dynamic nature of competition in the FM space.  Indeed, even in the months since the CMA’s decision a surge of alternative AI models (including China’s DeepSeek) have entered the market, demonstrating that competition in the FM space is highly unlikely to be “winner takes all”, with a multiplicity of models likely to persist.  

The CMA outlined the potential theories of harm it investigated in Microsoft / OpenAI, but without any further substantive analysis (given its lack of jurisdiction).  In particular, as well as considering horizontal overlaps in the development of FMs and FM-based services (including chatbots), the CMA was concerned that Microsoft might be able to restrict rivals’ access to OpenAI’s FMs in downstream markets and/or that the partnership could impact competition in the emerging market for the supply of accelerated compute, given OpenAI’s potential to act as an important customer in this market. 

These areas of focus reflect the CMA’s concerns outlined in its April 2024 Update Paper.  Given the sustained high levels of competition and innovation observed in the development of FMs, it is hard to see how these concerns could be substantiated. The CMA’s decision (and those in Amazon / Anthropic and Google / Anthropic) will be seen by some as a missed opportunity for the CMA to provide greater clarity on how it will assess when partnerships which promote competition in FMs might give rise to concerns in other markets.   

Reform on the horizon: a more restrained approach to reviewing AI partnerships in future?

As discussed in our previous post, in February 2025 the CMA set out proposals for major reforms to its approach to reviewing mergers.  This includes proposals to clarify and delineate how the CMA asserts jurisdiction over mergers, to “strengthen business certainty about which deals might attract the CMA’s attention”.  AI partnerships will be an area where the application of the CMA’s ‘4Ps’ approach will be particularly welcome.  In four of the five AI partnership cases to date, the businesses involved were subject to investigations ranging from two to fifteen months before the CMA determined it had no jurisdiction to review. 

On 17 March 2025, the UK government announced that it will consult on legislative reform to improve the pace, predictability and proportionality of the UK’s regime, including proposals to address “uncertainty with the existing share of supply and material influence tests”.  This reflects the CMA’s own signals that it believes legislative change may be required to enable it to truly deliver on its ‘4Ps’ approach. 

With the CMA also reviewing how it assesses global deals – and subject to a strategic steer to support economic growth and enhancing the attractiveness of the UK as a destination for international investment – it is questionable whether the CMA will continue to use its merger control powers to review AI partnerships in the same way in future, or a more targeted approach will be taken going forward.  Indeed, it is hard to think of another sector where an uncertain application of competition rules and unduly interventionist approach could be more at odds with the CMA’s stated ambitions to support productivity, innovation and growth across the UK economy.