Context: Last month, the UK Competition & Markets Authority (CMA) launched inquiries into three AI partnerships (Microsoft-Mistral, Microsoft-Inflection, Amazon-Antrophic) (April 24, 2024 ai fray article).
What’s new: On Friday (May 17, 2024), the CMA “decided that Microsoft Corporation’s partnership with Mistral AI does not qualify for investigation under the merger provisions of the Enterprise Act 2002.” But the full-text decision (PDF) lays out an expansive standard under which a wide range of commercial partnerships could be subjected to review under the UK’s strict merger review regime.
Direct impact: The CMA’s wide range of theories under which commercial arrangements might be deemed a relevant merger situation creates significant legal uncertainty, particularly in light of how difficult it is to overcome CMA merger-blocking decisions through an appeal as compared to other key jurisdictions.
Wider ramifications: The CMA has not yet decided on Microsoft-Inflection (where it proceeded more cautiously than jumping straight into a merger review process) and Amazon-Anthropic. The standard pronounced last Friday may result in “merger” reviews of some AI partnerships, and will apply far beyond the field of AI.
A year ago, there was debate over whether the UK was closed for business. The CMA’s aggressive interpretation of merger laws, in conjunction with the limitations of the UK appellate process, threatened to discourage investment in the UK. The debate subsided a few months later. The low-key seven-page ruling that the CMA put out on Friday does not have the potential to reignite anything: they decided to let the Microsoft-Mistral partnership off the hook for now. But if one reads the decision carefully, the risk of future regulatory overreach is real.
The CMA found that “Microsoft’s potential shareholding in Mistral [in exchange for a €15 million investment] is less than 1%” and also saw no indication that Microsoft would be able to “influence Mistral’s policy through exercising votes at shareholders’ meetings or through otherwise influencing the board.” A merger is about two companies becoming one, and that is obviously not the case when one company owns 1% of the other and doesn’t have any minority-shareholder rights that would effectively amount to an acquisition of control.
Paragraph 7 of the CMA decision says this:
“Control is not limited to the acquisition of outright voting control but may include situations falling short of outright voting control, including material influence, de facto control and a controlling interest.”
If applied reasonably and in accordance with traditional standards, the hurdle for “material influence” would be rather high. But the CMA’s Microsoft-Mistral decision seeks to broaden the scope of that term in the next paragraph:
“The ability to exercise material influence is the lowest level of control that may give rise to a relevant merger situation. When making its assessment, the CMA focuses on the acquirer’s ability to influence materially policy relevant to the behaviour of the target entity in the marketplace. The policy of the target in this context means the management of its business, and thus includes the strategic direction of a company and its ability to define and achieve its commercial objectives.”
Then comes paragraph 9, which sounds extremely arbitrary:
“The assessment of material influence requires a case-by-case analysis of the overall relationship between the acquirer and the target, having regard to all the circumstances of the case. The variety of commercial arrangements entered into by firms makes it difficult to state categorically what will (or will not) constitute material influence.”
Put differently, companies in commercial partnerships are at the CMA’s mercy. The agency can always say that the “overall circumstances” meet the low “material influence” standard. That is disconcerting. There is no safe harbor for companies.
After ruling out the traditional form of influence through shareholder voting rights (minimal) and board representation (not the case here), the CMA discusses three other theories. While the relatively small Microsoft-Mistral deal doesn’t raise issues under any of those theories (at least not based on the facts that the CMA looked at), those parts of the decision are, collectively, a harbinger of things to come:
- The provision of cloud computing resources to an AI provider or developer (called “compute commitment” in the decision) “may in certain circumstances result in an acquisition of material influence” (paragraph 12). The CMA expresses concern over a potential lock-in, with a foundation model developer being dependent on such resources. What makes that part really puzzling is that cloud computing resources are a digital commodity. In the case at hand, the CMA is not concerned, given that it’s just about a “relatively modest proportion of Mistral’s contracted compute” and there are no approval rights or similar ways related to those cloud resources that would give Microsoft material influence. But what’s troubling is that agreements where, in addition or in lieu of a cash consideration, a company receives highly interchangeable cloud computing resources are susceptible to an merger inquiry at all.
- Starting with paragraph 14, the CMA then discusses potential “[i]influence through the distribution agreement.” In the Microsoft-Mistral case, the availability of Mistral’s foundation models through platform like Microsoft’s Azure Machine Learning Studio is considered a deal of that kind. For different reasons, one of which is the non-exclusive nature of that agreement, the CMA is not concerned in this particular case, at least for now. But again, the problem is that there could be merger reviews, with all that it entails in terms of risks and delays.
- The decision then also discusses, in only one paragraph, the “possibility of future collaboration and development opportunities.”
Lawmakers had reasons for which they gave competition watchdogs like the CMA more power and wiggle room with respect to actual mergers than in connection with commercial partnerships. The idea of merger control is that a company should not be able to avoid competition through acquisitions, or that in a market where two players are key competitive forces, they shouldn’t just have the option to merge in order to generate higher levels of profits as a result of reduced competition.
Competition authorities can also investigate and take action against commercial partnerships. But procedurally and substantively, the rules are defendant-friendlier, as they must be in order to ensure that a country is open for business. In the alternative, regulatory excess or just the fact that companies have to fear it will have a chilling effect.
If AI partnerships give companies additional distribution options (downstream benefits) and/or access to resources (upstream benefits), that will normally result in more competition and innovation. This does not mean to say that there would never be a risk of some companies seeking to exploit what they believe to be loopholes. But there are tools that antitrust agencies have to deal with that if it comes to worst.
The end of combating a hypothetical circumvention of merger rules cannot justify the means of simply seeing a merger in everything. Merger law gives regulators more power and flexibility than antitrust laws relating to horizontal or vertical agreements between independent companies. But it’s not meant to lower the standard for inquiries into run-of-the-mill commercial partnerships and to open the door to unpredictable and arbitrary decisions that harm competition and innovation on the bottom line.