In-depth reporting and analytical commentary on artificial intelligence regulation. No legal advice.

UK competition authority CMA launches triad of AI investigations: Microsoft-Mistral, Microsoft-Inflection, Amazon-Anthropic

Context: Recently, the UK Competition & Markets Authority (CMA) published a paper on AI Foundation Models (April 14, 2024 ai fray article). The agency expressed concern over a large number of partnerships between major technology corporations and AI startups, though there is a diagram in that paper according to which those AI companies appear to be in a privileged position to choose their partners, and many have more than one Big Tech partner.

What’s new: Today the CMA has followed up to its research with the launch of three parallel investigations:

  • a merger inquiry into Microsoft’s minority investment in French AI startup Mistral;
  • a merger inquiry into Amazon’s investment in Anthropic; and
  • an inquiry for the purpose of determining whether “Microsoft Corporation’s hiring of certain former employees of Inflection AI and its entry into associated arrangements with Inflection AI” should give rise to yet another merger review.

Direct impact: The two “mergers” that the CMA has already decided to investigate are minority investments, meaning they can fall under merger rules only in exceptional circumstances involving an acquisition of a controlling interest. And the third one, Microsoft-Inflection, appears even further from a merger. It is unclear why the CMA stretches the envelope of merger law.

Wider ramifications: The CMA’s (and some other regulators’) agenda may ultimately be to ask lawmakers for new competition rules that would have scope for reviews of partnerships that do not meet traditional merger control criteria. Another possibility is that the CMA exercises its investigatory powers in hopes of finding out anything that could give rise to non-merger antitrust investigations.

The CMA’s choice of those three targets raises a number of interesting questions.

It is, of course, only a coincidence that the CMA is launching two AI-related investigations of Microsoft partnerships about a year after making the shocking decision to block Microsoft’s acquisition of game maker Activision Blizzard, a deal that was ultimately cleared but after a certain restructuring. But there is an interesting parallel: the CMA’s sole remaining theory of harm in the Activision case related to cloud gaming, which represents approximately 1% of the worldwide gaming market, and now the same agency is investigating Microsoft’s Mistral investment, with Microsoft apparently owning not even one percent of that company. It appears that when Microsoft is involved, the CMA treats 1% (or sub-1% numbers) like 51% or 100%.

It’s easy to see that Microsoft’s share in Mistral is small and must be lightyears away from a controlling interest. Microsoft invested only $16 million, but according to a Bloomberg article, Mistral has a multi-billion-dollar valuation and in December received hundreds of millions from Salesforce, NVIDIA, and venture fund Andreessen Horowitz. A more recent Financial Times article says Mistral is looking to raise another $500 million at a $5 billion valuation.

At this year’s Mobile World Congress, Microsoft’s president outlined his company’s commitment to “AI Access” (February 26, 2024 Microsoft corporate blog post). It appears that Microsoft’s strategy is about forming multiple partnerships and providing resources to different AI startups. From a traditional competition perspective, that would have been deemed a good thing. The CMA appears to be unconvinced, at least until it has conducted a couple of investigations.

The only true M&A transaction concerning AI that the CMA really should have been concerned about at the time was Google’s 100% purchase of the UK’s AI crown jewel, DeepMind. But that deal was waved through. They may realize it was not a good decision for the UK, but they can’t correct an old mistake by making some new ones.

The investigation of Microsoft’s “arrangements” with Inflection is even harder to understand. Unless there is something that the world outside the CMA doesn’t know, there are precisely two “arrangements” at issue:

  • Microsoft hired Inflection co-founder Mustafa Suleyman and other employees (largely U.S.-based, though Microsoft is actually investing in AI jobs in the UK) (March 19, 2024 Microsoft corporate blog post). Microsoft stresses that it negotiated with each individual separately.
  • Plus there is a non-exclusive intellectual property license.

What the CMA is doing by launching an inquiry (instead of concluding right away that this is anything but a merger) flies in the face of everything that is known about merger rules. Hiring people and entering into IP license agreements is not an acquisition of a controlling interest. The CMA’s concern about this even runs counter to a trend in the U.S., where the Federal Trade Commission is promoting employees’ freedom of choice, such as by banning noncompete clauses in employment contracts (April 23, 2024 FTC press release).

Now, compared to Microsoft’s ownership of less than a percent of Mistral (for $16 million) and apparently 0.0 percent of Inflection, Amazon’s partnership with Anthropic at least involves a substantial amount of money: $4 billion according to Amazon’s announcement. But even in that case, it’s apparently just a “minority ownership position” (according to Amazon).

Not only Amazon but also Google has a partnership with Anthropic (February 3, 2023 Anthropic press release). For whatever reason, the CMA decided to focus on Amazon and Microsoft rather than Google.

In all three cases, it is hard to see what the CMA will accomplish. Is it just a fishing expedition, attempting to find evidence of anticompetitive dealings? Or is it about making a political statement?

In a first step, the CMA is asking stakeholders for comment. It will evaluate the submissions, form its own opinion, and then decide what to do. For now, the situation is puzzling. It is true that major technology companies are now at the forefront of the AI revolution, partly through their own developments and offerings (just yesterday, Microsoft announced its Phi-3 small language models) and partly through partnerships. Competition regulators are right to keep an eye on whether the relevant markets remain contestable. But turning merger law on its head is highly unlikely to be helpful in any way.