Open and shut? Lessons from the CMA's Microsoft/OpenAI merger inquiry

03 June 2025

Following an extensive 15-month investigation, the CMA concluded that it did not have jurisdiction to review Microsoft’s partnership with OpenAI. We consider below the implications of this decision for future cases concerning changes in the nature of control over time, and what this might tell us about the CMA’s approach to similar partnerships in the AI space going forward.  

Introduction

OpenAI develops and supports several leading AI foundation models (FMs1), as well as FM-based services–notably the “chatbot” ChatGPT. OpenAI was established as a for-profit company in 2019 by OpenAI, Inc. (OpenAI Nonprofit), a non-profit AI research and development organisation that continues to control the day-to-day management of OpenAI. 

On 8 December 2023, the CMA opened an investigation into OpenAI’s relationship with Microsoft (at the time its largest investor), in response to the dismissal and subsequent reappointment of Sam Altman as CEO of OpenAI in November 2023, and suggestions that Microsoft may have had an important role in his reappointment. The decision to close the investigation, owing to a lack of jurisdiction, was announced over a year later on 5 March 2025. 

Quite why an investigation focusing primarily on jurisdictional issues took 15 months to complete2 was not entirely clear from the 5 March announcement. However, the full found-not-to-qualify decision, published on 15 April 2025, reveals that the CMA undertook a detailed investigation that involved the consideration of a wide range of evidence, including large volumes of the parties’ internal documents. Additionally, material aspects of the partnership changed during the course of (and indeed, potentially even in response to) the CMA’s investigation, adding to its complexity.

The investigation took place against a backdrop of the CMA reviewing a number of other AI partnerships (for more on this, see our article from October 2024) whilst also carrying out a broader study on AI FMs (as it had been requested to do following a 2023 UK government White Paper), working closely with other UK regulators such as the Information Commissioner’s Office.

The (not so bright) line between material influence and de facto control 

In order for a “relevant merger situation to have been created” under the Enterprise Act 2002 (EA02), control (meaning “material influence”, de facto control or de jure (legal) control) over a business (or part thereof) must be acquired by one or more parties that did not previously have that same level of control. This includes situations in which there has been a change in the nature of control, for example where an existing shareholder/investor moves from material influence to de facto control or from de facto control to de jure control.

The ability to exercise material influence is the lowest level of control that may give rise to a relevant merger situation. As detailed below, the CMA concluded that Microsoft had material influence over OpenAI since the partnership was launched in 20193. The jurisdictional question for the CMA was therefore whether developments in the parties’ relationship meant that Microsoft had subsequently acquired de facto control over OpenAI.

The CMA’s guidance provides that de facto control consists of the ability unilaterally to determine a company’s policy. It also notes, however, that there is no ‘bright line’ between factors which may give rise to material influence, and those giving rise to de facto control. For example, an investor’s industry expertise, which can be a factor in establishing material influence, might also be a relevant factor in determining whether de facto control has been acquired (on the basis that such expertise may lead to the investor’s advice being followed to a greater extent that its shareholding or investment would seem to warrant).

What were the key factors the CMA looked at when establishing whether de facto control had been acquired?

1. Investment and corporate governance

The CMA observed that Microsoft (at the time of the decision) was OpenAI’s largest investor and could exercise a high degree of influence over OpenAI, over and above its relatively few formal governance rights. In particular, although Microsoft’s veto rights were limited to “typical financial investor protections”, the parties’ internal documents showed Microsoft engaging regularly with Open AI’s senior management and using its position as the largest investor to exert considerable influence over OpenAI’s commercial policy. However, the CMA noted that, while Microsoft acquired the right to appoint an observer to the OpenAI Nonprofit board in December 2023, it relinquished this right in July 2024. The CMA also concluded that Microsoft’s role in the reappointment of Sam Altman as CEO in November 2023 was one of an “influential actor, rather than driver of these events”. Further, more recent events (including the appointment of additional independent board members, and additional third-party investment into OpenAI) indicated that “Microsoft’s influence over OpenAI has limits, despite its position as its largest investor”.

2. Compute supply

Computing capacity is an essential input for the development and distribution of FMs. The CMA concluded that the evidence pointed to Microsoft being able to use the supply of compute to exert a high degree of influence over OpenAI. Microsoft had been the exclusive provider of the majority of OpenAI’s compute requirements for a number of years and the CMA observed that the parties’ internal documents, among other evidence, indicated that the compute arrangements provided Microsoft “with a means to exert significant pressure on OpenAI, which may impact OpenAI’s incentives to act independently.” However, in January 2025, it was announced that these exclusivity arrangements had been renegotiated, enabling OpenAI to obtain compute from third parties in certain circumstances, with Microsoft retaining a right of first refusal for new compute capacity and exclusivity in relation to OpenAI API4 workloads (i.e. the compute powering Open AI’s interactions with external applications). Despite certain limitations on OpenAI’s ability to source compute from other parties, the CMA found that the amendments did reduce OpenAI’s overall reliance on Microsoft, “which may reduce Microsoft’s ability to use compute supply to exert pressure on OpenAI”. 

3. IP and commercialisation rights

The CMA considered it relevant that Microsoft had an exclusive licence to OpenAI’s IP (with some carve-outs, and with OpenAI also retaining rights to commercialise its IP). However, the CMA also found that OpenAI had taken steps to pursue commercial opportunities that might not necessarily align with Microsoft’s interests, including entering into negotiations with Apple regarding the integration of certain ChatGPT features into Apple devices, and the rolling out of ChatGPT-related features to both Microsoft’s OneDrive product and Google’s competing Google Drive product. 

 

A high degree of material influence, but not de facto control

The CMA’s overall conclusion was that, while the above evidence showed that Microsoft was able to exert a “high degree of material influence” over OpenAI’s policy (and sought to use that influence), there were also examples of OpenAI acting independently and in a way that was not aligned with Microsoft’s interests. Taking the evidence in the round, the CMA did not believe that Microsoft was able to determine OpenAI’s commercial policy. 

This suggests that, where parties can point to tangible examples of the ‘target company’ making decisions or pursing opportunities which may not necessarily be in line with the relevant investor’s interests, this will be an important part of the evidence base demonstrating that a position of de facto control has not been reached.

Factors related to investment and corporate governance are likely to be relevant in every case involving an assessment of potential de facto control. In IP-heavy sectors, control over the exploitation of key IP rights is also likely to be a determining factor. Additionally, while ‘compute supply’ is clearly more specific to AI and FM markets, one can envisage other scenarios in which a relevant factor in establishing de facto control will be whether the relevant investor is the exclusive (or near-exclusive) supplier of a key input, which may be a means through which to exert pressure on the “target company”.

The CMA’s substantive concerns

The decision suggests that the CMA had initially been concerned that an increase in Microsoft’s control over OpenAI could allow it to restrict rivals’ access to Open AI’s leading FMs in markets where they are likely to be important and where Microsoft already holds a strong market position (for example, the supply of cloud compute services and productivity software). In addition to this potential input foreclosure theory of harm, the CMA appeared to have in mind a customer foreclosure theory of harm in the market for accelerated compute (on the basis that OpenAI could prove to be a particularly important customer in that market in the future).

The CMA also raised potential horizontal theories of harm, such as the overlaps between the parties in the development of FMs, the distribution of FMs, and the supply of FM-based services, such as chatbots. 

Ultimately, however, given the CMA concluded that it did not have jurisdiction to review the partnership, there was no basis for it to explore any of these potential theories of harm. 

A role for the new UK digital markets competition regime?

Under the Digital Markets Competition and Consumers Act 2024 (DMCCA), the CMA has the power to designate firms as having “strategic market status” (SMS) in relation to a particular digital activity. 

While, as we have seen, the competitive situation can change quickly in digital markets, at present it appears unlikely that the CMA will open an SMS investigation looking specifically at market power in AI FM markets over the next year or two. There are a few reasons for this, key among them that these are fast-moving markets, in which it is not yet clear that a sufficiently significant level of market power has been reached by any one player. The CMA is also likely to prioritise other areas for investigation first, in particular: (i) areas where it has already carried out extensive work, such as cloud services; and/or (ii) areas regulated by the EU Digital Markets Act, such as desktop operating systems and social media services. 

However, the references in the Microsoft/OpenAI decision to possible vertical foreclosure in AI-related markets suggest that the CMA may look closely at such issues should there be a future SMS investigation into cloud services. One might reasonably conclude that such an investigation within the next year is likely, given one of the provisional remedies emanating from the CMA’s ongoing cloud services market investigation is a recommendation to the CMA Board to prioritise commencing SMS investigations into both Amazon Web Services and Microsoft in respect of their activities in cloud services.

Indeed, while AI and FM markets have not been the direct focus of the first two sets of SMS investigations, the CMA’s invitations to comment in those investigations made reference to AI services, and the CMA is clearly considering how AI services might influence the digital activities under investigation. This suggest that the CMA is likely to take a holistic approach when exercising its new digital markets powers, conscious of the need to “future-proof” its interventions so that they can capture within their scope future developments in AI that might impact competition in the relevant digital activities.

The end of a chapter, but not the whole story?

While it appears from the Microsoft/OpenAI decision that the facts on the ground supported the conclusion that the CMA did not have jurisdiction, the CMA may also be relieved not to have to consider expending political capital launching what would likely have been a highly contentious substantive merger review. Wider arguments around discouraging investment and innovation, the predictability of UK merger control, and whether the CMA should be taking action to block global deals (as played out in the press during the CMA’s review of the Microsoft/Activision deal back in 2023), may well have come to the fore again. Such a debate would likely have been particularly unwelcome for the CMA at this time, given the current focus on growth, innovation and investment

In closing this merger inquiry, however, the CMA is unlikely to be closing the door entirely on its monitoring of AI partnerships or AI FM markets more broadly. In addition to its ongoing broader study on AI FMs, there remain other avenues for the CMA to ‘keep an eye’ on AI FM markets as they develop, in particular through the use of tools available to it under the DMCCA, which potentially allow for more adaptable and less binary solutions than those available to it when reviewing partnerships or transactions under the EA02.

1 The CMA has previously defined foundation models as “large, machine learning models trained on vast amounts of data”.

2 Technically, as the CMA opened the merger inquiry on 4 March 2025 and closed it on 5 March 2025, the merger inquiry itself took just one day. The 15-month period referred to in this article is in reference to the 8 December 2023 to 5 March 2025 period during which the CMA was investigating the partnership. 

3 This acquisition of material influence was not investigated by the CMA at the time and the limitation period for doing so has long since expired. 

4 Application Programming Interface.