Context: Five months ago, the UK Competition & Markets Authority (CMA) simultaneously announced three investigations under merger law of AI-related partnerships (Amazon-Anthropic, Microsoft-Mistral, Microsoft-Inflection) (April 24, 2024 ai fray article).
What’s new: After clearing Microsoft-Mistral in May (PDF) and Microsoft-Inflection earlier this month (PDF), the CMA has now also terminated, without launching an in-depth investigation, Amazon’s partnership Anthropic (PDF). Unlike the two Microsoft partnerships, this one was a multi-billion dollar deal.
Direct impact: All three partnerships were cleared on the fast track, which means that only a major change of the situation, or another acquisition of shares, could result in another investigation.
Wider ramifications: The CMA, like some other competition enforcement agencies, is trying to expand the scope of the merger review regime as it is the legal framework that gives such regulators more leverage than unilateral conduct, cartel or other non-merger antitrust investigations. That’s why the reasons underlying those clearance decisions warrant a quick look, despite the appearance of “much ado about three nothing burgers.”
There’s this saying that “if all you have is a hammer, everything looks like a nail.” Competition authorities like the CMA have a lot more than just a hammer. There’s a diverse toolbox. In the technology sector, unilateral conduct investigations (abuse of dominant market position, or “monopoly” in U.S. antitrust terminology) are the most common category of investigations, while other sectors such as automotive are prone to the creation of cartels. The European Commission’s Directorate-General for Competition (DG COMP) even managed to take action against a tax issue (Apple-Ireland) on a “state aid” basis and prevailed in the EU’s top court.
But merger control is the most powerful hammer those agencies have. That has been the case for a while, but now there is an expansive trend: they try to see a merger in commercial partnerships that the predecessors of today’s agency leaders would not have (mis)categorized as mergers in their wildest dreams. That is not what lawmakers intended when merger rules were put in place in the UK and other major jurisdictions. It’s obviously key that merger rules are not circumvented, and where that is the case, the CMA can resort to a “material influence” theory. It should do so within reason, however. Otherwise it’s not that companies are circumventing merger rules, but about regulators disregarding the clear boundary of merger review rules.
The result of each of the three investigations is zero, and at the outset of those inquiries ai fray expressed doubts about all three of them.
Amazon described its shares in Anthropic as a “minority” position. It’s worth noting that Google is also a major Anthropic partner. The CMA’s summary of the Amazon-Anthropic decision describes it as “[a] $4 billion investment by Amazon in Anthropic, with $1.25 billion invested in September 2023 and $2.75 billion in March 2024, which is convertible into non-voting equity in certain circumstances.” That makes it, in financial terms, a far larger deal than Microsoft-Mistral and Microsoft-Inflection.
This decision came out as the last one of the three, but the reasoning provided by the CMA would actually have made it a very straightforward one:
“In particular, the CMA found that Anthropic’s UK turnover does not exceed £70 million in the UK, nor do the Parties, on the basis of the available evidence, together account for a 25% or more share of supply of any description of goods or services in the UK.”
The redundancy (“UK turnover … in the UK”) is in the original.
Microsoft-Mistral (where those thresholds are mentioned only in a footnote, but not actually applied) and Microsoft-Inflection (where there isn’t even a mention of the threshold) don’t meet those UK revenue thresholds either. If it was just about thresholds, the CMA could not only have finished those investigations far sooner: it could simply not have launched them in the first place. That would have conserved agency and party resources. So why didn’t they do that? And is it a good or bad sign for one deal or the other?
Instead of dropping all three investigations on the basis on which Amazon was let off the hook today, the CMA looked at each deal in different ways:
- The Microsoft-Mistral agreement “in the absence of other factors conferring influence, [is] unlikely to enable Microsoft to influence materially the commercial policy of Mistral.” There is no distribution dependency either. As a result, the CMA found that Microsoft and Mistral have not ceased to be independent entities (which was obvious, though the CMA sometimes makes statements that suggest those AI partnerships are opaque).
- In the Microsoft-Inflection case, the CMA viewed the combination of the hiring of an alleged majority of Inflection’s employees (founders included) and a license deal as a merger. But it found that Inflection simply wasn’t significant enough in the chatbot and foundation model markets to raise any concerns over a substantial lessening of competition.
Microsoft-Inflection is does not appear likely to lead to a follow-on investment that would change the CMA’s perspective.
Unless Microsoft and Mistral amend their agreement or enter into any additional ones, that partnership is not going to be reinvestigated either. Unless and until Mistral reaches UK merger thresholds, the CMA would lack jurisdiction.
Those three non-merger investigations under merger law made no sense in the first place. Is this just a prelude to a CMA investigation under “merger” law of Microsoft’s OpenAI partnership? The CMA still has a webpage for that one, but it’s been more than nine months since the first and last news. There have been media reports, however, according to which the CMA might launch a formal investigation of Microsoft-OpenAI. Did they launch three other AI “merger” investigations just so they could later say they cleared most of those partnerships rather quickly, hoping to come across as more reasonable that way? Time may tell.