Big Tech-Small AI Partnerships

The interaction between corporate governance and competition policy has never been more topical. Last week, Microsoft announced that it would drop its non-voting observer member on the board of OpenAI, the artificial intelligence (‘AI’) firm made world-famous by its ChatGPT bot. Since 2019, Microsoft invested a total of $13 billion into OpenAI. As of 2023, it has also become its exclusive cloud provider. The collaboration was not initially meant to involve any governance rights. But OpenAI’s governance debacle in 2023, which caused the ousting of its CEO Sam Altman, led the Big Tech player to take a non-voting observer role on the entity’s board.

 In the same vein, Apple—which also has a partnership with OpenAI—had recently disclosed that it also planned to take an observer seat on OpenAI’s board. After Microsoft’s withdrawal, rumours are that Apple won’t join the board after all.

Both developments are allegedly due to intensifying scrutiny from antitrust authorities, which have pushed OpenAI to adopt a ‘new approach’ to collaborating with its partners. These events are noteworthy because of the peculiar nature of OpenAI’s relationship with the two companies. The OpenAI-Microsoft deal is the poster child of a broader web of partnerships that link Big Tech firms with up-and-coming AI companies. These Big Tech-Small AI collaborations appear to be a win-win. Big Tech companies provide capital, cloud access, computing power, and an easy way to reach customers—all much-needed assets for small AI companies seeking to sustain cutting-edge research. In return, the Big Tech players typically get to embed state-of-the-art AI applications into their pre-existing ecosystems, as well as profit rights and sometimes exclusive licences.

These Big Tech-Small AI partnerships involve a degree of collaboration which goes beyond traditional market interactions. As a result, antitrust agencies are increasingly concerned that what is presented as a ‘partnership’ is in fact a merger in disguise. From the Competition and Markets Authority to the German Bundeskartellamt, the European Commission and, most recently, the Federal Trade Commission, antitrust investigations have mushroomed.

To be caught by merger control rules at the EU level, a deal needs to confer control of one firm over another, with control being understood as ‘the ability to exercise decisive influence.’ In the US, Section 7 of the Clayton Act catches any acquisition of stock or assets whose effect ‘may be to substantially lessen competition or to tend to create a monopoly.’ In contrast with the EU, no finding of control is necessary to trigger a merger investigation.

These inter-jurisdictional differences reveal that it is unclear at which level of collaboration between two entities a merger probe is warranted.

Fifty Shades of Control

In corporate law, an entity is usually considered as controlling another when the former holds a majority stake in the latter’s capital, or at least a significant minority stake giving it de facto control over the latter. The Delaware Chancery Court, for instance, performs an ‘adequate control’ test which ‘requires the court to undertake an analysis of whether, despite owning a minority of shares, the alleged controller wields “such formidable voting and managerial power that, as a practical matter, it is no differently situated than if it had majority voting control.”’ De facto control by a minority blockholder is usually found when the company’s capital is widely dispersed, making it possible for that blockholder to effectively control it in a similar fashion to a majority shareholder.

European jurisdictions have a similar approach. In France, for instance, control exists when an entity legally holds the majority of voting rights, when it can factually determine the decisions taken at general meetings, and/or when it can name or revoke a majority of the company’s board members. Control is also presumed to exist when the biggest shareholder of the company holds at least 40% of the company’s voting rights.

While the EU merger control framework operates with a comparable distinction between de jure and de facto control, its definition of control is much broader than in corporate law. Besides de jure control, which is confined to situations where a shareholder legally holds the majority of voting rights, de facto control can mean widely different things. First, there are situations where a minority shareholder can nonetheless have the majority of voting rights in general meetings due to special circumstances, most notably low attendance rates at shareholder meetings. Second, control can also be conferred through contractual relationships which enable one firm to impose its vision on another firm’s governing bodies. This latter configuration is the most relevant to Big Tech-Small AI partnerships.

In a line of merger decisions, the Commission identified de facto control in situations where a minority shareholder could impose its views on the majority due to the existence of a relationship of quasi economic dependence. For example, in the Sollac/BAMESA decision, the Commission found that a minority shareholder owning 30% of the share capital was in a position of control because it was a supplier of key inputs and it represented a major player in its industry, thus having much stronger overall economic power over the company than any other shareholder.

As opposed to the clarity of de jure cases, de facto control situations are heavily fact-specific, and their identification involves a solid degree of administrative and judicial discretion. Matters are equally complicated in the US. The reach of the Clayton Act extends beyond acquisitions of stock to acquisitions of tangible and intangible assets. Case law has shown that pretty much anything can count as an ‘asset’ since the language of the Clayton Act ‘was deliberately couched in general and flexible terms.’ For instance, the granting of an exclusive distribution license over a TV series coupled with a commission of 27,5% from the proceeds was considered as an acquisition of ‘assets’ and examined as a merger.

Assessing Control in the Context of Big Tech-Small AI Partnerships

The law is open-ended, but is it open-ended enough to catch OpenAI’s collaborations? The EU Commission recently concluded that for now, Microsoft does not have controlling influence over OpenAI, hence their partnership does not amount to a merger. While we do not have much information about the Commission’s assessment, we can attempt to reverse-engineer it.

The Commission may have assessed OpenAI’s economic dependence on Microsoft. Indeed, Microsoft is a supplier of key inputs to OpenAI due to acting as its exclusive cloud provider. It also represents a major player in several markets. Furthermore, many commentators claimed that the tech giant played a key role in reinstating Sam Altman as CEO, showing its ability to strongly influence, if not outright determine, some key decisions of OpenAI.

But things are not so simple. In past economic dependence cases, minority shareholders usually held significant shareholdings. This does not apply to the Microsoft-OpenAI scenario. OpenAI has a ‘weird,’ dual structure consisting of a nonprofit and a capped-profit arm. The nonprofit arm controls the for-profit arm through its board. In the for-profit arm, Microsoft holds a 49% stake, which was decoupled from any governance rights until the appointment of the non-voting observer on the nonprofit arm’s board.

The question, then, is to what extent a non-voting observer member can be considered as controlling the board. One can argue that OpenAI’s reliance on Microsoft’s cloud and computing power means that Microsoft could use the observer as a platform to impose its will on OpenAI’s board. US case law determined in the past that board observers could ‘significantly influence’ board discussions. Contrary to US corporate law, however, the EU merger control standard requires decisive influence. In this context, claiming that one non-voting member could control the board is a stretch.

We do not know whether the non-voting board member was a decisive factor in the Commission’s initiation of the merger probe. If it was, it is doubtful that OpenAI’s new approach to structuring its partnerships would do much to calm the antitrust watchdog. The entity vouched to ‘host regular stakeholder meetings to share progress [...] and ensure stronger collaboration’ as well as to ‘continue to receive feedback and advice from these key stakeholders.’ Under the EU merger rules, the right to obtain access to ‘competitively significant information’ and ‘to exert influence’ are material factors which point to the existence of de facto control. Based on this, the elimination of non-voting observers might not suffice to dispel worries. In an official speech, Margarethe Vestager declared that ‘the story is not over,’ keeping the door open for potential re-assessments of this and other Big Tech-Small AI partnerships under the merger rules.

In the US, the story has only just begun. The FTC launched an inquiry into these partnerships in late January, requesting information on, inter alia, ‘governance or oversight rights’ and ‘the topic of regular meetings.’ So far, the agency has only issued a request for information, and the outcome of any potential merger scrutiny is up for grabs. Scrapping the observer seats will eliminate any formal governance rights, but the FTC would still assess the unfoldment and implications of the regular meetings OpenAI announced. Importantly, given that no proof of control is necessary, the US is for once a more fertile ground than the EU for finding a potential antitrust violation.

A Fine Line

The fact that OpenAI’s partnerships can be plausibly considered as mergers by the EU and US antitrust watchdogs testifies to the malleability of the merger control frameworks in both jurisdictions. It also shows that firms willing to collaborate without merging are navigating a fine line.

The overall configuration of the Big Tech-Small AI partnerships is unusual, sitting at the unclear boundary between market contracts and vertical integration. Some voices claim that these partnerships were carefully engineered to avoid antitrust scrutiny. Yet there might be more to the choice since structuring the deals as market-based relationships comes with certain efficiencies. Most importantly, it preserves the high-powered incentives associated with market transactions, safeguarding—and potentially even bolstering—the innovation incentives of the AI players. This doesn’t mean that these ‘new modes of organizing’ should not be scrutinized by antitrust authorities at all. But stretching the merger control framework to capture them might be too much of a stretch.

Teodora Groza is a PhD student in competition law at Sciences Po Paris and the Editor-in-Chief of the Stanford Computational Antitrust Journal.

Paul Oudin is a DPhil in Law candidate at the University of Oxford and a lawyer at Vermeille & Co.