top of page
Search

What the Microsoft–OpenAI Non-Merger Teaches Regulators

  • Writer: Aişe Gül Akkoyun
    Aişe Gül Akkoyun
  • May 15
  • 3 min read

The UK's competition authority spent fifteen months on a partnership it ultimately said wasn't a merger at all.


Background

The UK Competition and Markets Authority closed its merger inquiry into the Microsoft–OpenAI partnership on 5 March 2025, concluding that there was no basis to review the arrangement under the UK's merger control regime, and published a full decision on 15 April 2025. The inquiry had examined whether the terms of Microsoft's investment and commercial arrangement with OpenAI gave Microsoft de facto control over the company, and whether any benefits from the partnership were outweighed by a lessening of competition in the AI and foundation-model market. The CMA's conclusion was narrow but specific: Microsoft holds significant material influence over OpenAI, but does not control its commercial policy in the sense the merger-control regime requires.


The end of a five-case cycle

The Microsoft–OpenAI decision closed out a cycle of five separate merger inquiries the CMA had opened since December 2023 into partnerships between AI developers and major technology companies — a run of investigations that collectively tested whether the UK's existing merger-control tools, built around traditional acquisition and control concepts, could meaningfully capture the looser, investment-plus-commercial-agreement structures that have become the dominant way large technology companies gain influence over frontier AI labs without an outright acquisition.


Why "no jurisdiction" is itself a significant finding

A regulator spending fifteen months investigating an arrangement before concluding it lacks jurisdiction to act is not a neutral, cost-free outcome. It signals, at minimum, that the existing merger-control framework's concept of "control" — built for a world of share purchases and board seats — struggles to classify a relationship defined instead by exclusive compute access, revenue-sharing arrangements, and a large minority investment stopping deliberately short of the formal control threshold. Whether that is a gap regulators should close through new legislative tools, or a signal that these partnerships genuinely do not raise the competitive concerns that trigger merger review, is precisely the question the CMA's own conclusion leaves open.


What would be worth measuring

  • How the structure of major AI-developer/big-tech partnerships (investment size, revenue share, exclusivity terms) compares across the five inquiries the CMA opened since December 2023, and whether they cluster around a similar "just under the control threshold" design

  • Whether other competition authorities examining comparable AI partnerships (the European Commission, the US FTC) have reached similar "no jurisdiction" conclusions, or diverged

  • Whether the absence of merger-control jurisdiction over these partnerships has been compensated for by other regulatory tools — data-access conditions, AI-specific market-conduct rules — or left genuinely unaddressed

  • Whether deal structures in this space have measurably shifted toward staying under whatever threshold triggered the CMA's original scrutiny, once that threshold became publicly known through the inquiry's conclusion


The open question

If the current wave of AI-lab partnerships is deliberately structured to sit just outside traditional merger-control definitions of control, then competition authorities face a choice similar to the one investment-treaty drafters face with denial-of-benefits and abuse-of-process doctrines elsewhere in this archive: adapt existing legal tests to catch functionally equivalent arrangements, or accept that the current framework was built for a different shape of transaction and needs new statutory tools rather than a broader reading of old ones.

Related on DLS

Sources

Comments

bottom of page