by Teodora Groza, CodeX Collaborator (TRACK AI and Computational Antitrust)

OpenAI’s ties with Microsoft keep making headlines. The partnership that began with a $1 billion investment in 2019 has seen several phases and triggered the attention of multiple antitrust watchdogs worldwide. The initial stages of the two firms’ collaboration raised concerns over whether OpenAI’s independence was impaired by its relationship with Microsoft, which used to be the firm’s exclusive cloud provider. Nonetheless, recent developments have all but erased those worries. Ever since June 2024, the relationship between the two firms has gradually loosened. Last week, OpenAI announced a $12 billion investment in CoreWeave, a cloud service provider whose main customer to date is Microsoft.

The evolution of the OpenAI—Microsoft partnership is worth unpacking for two reasons. First, the dynamics of their relationship is relevant beyond the confines of the governance structures of the two companies. This partnership is not an isolated island of collaboration between Big Tech players and up-and-coming AI firms, but an epitome of a trend that characterizes the entire AI landscape. From Anthropic to Inflection AI, there is hardly any AI firm known to the large public which has not received investments from Alphabet, Amazon or Microsoft. In 2023, two-thirds of the funds raised by AI start-ups came from Big Tech players, suggesting that the latter may be taking over the role that had been traditionally fulfilled by venture capitalists.

Second, these Big Tech—Small(er) AI partnerships have been primarily scrutinized by antitrust authorities through the framework of merger control, which enables the review of deals where one market player acquires control (in the EU) or decisive influence (in the US) over another. Nonetheless, the dynamics of the OpenAI—Microsoft relationship require us to question whether these collaborations were ever aimed at the acquisition of control or decisive influence. To understand what else may be at stake, we need to first dive deeper into the evolution of this partnership.

A Partnership in Phases

The OpenAI—Microsoft saga began in July 2019, when Microsoft invested $1 billion to support “building beneficial AGI.” From the outset, Microsoft became OpenAI’s exclusive cloud provider, an agreement that was supposed to be a win-win: OpenAI would get access to computing power while “working hard […] to further extend Microsoft Azure’s capabilities in large-scale AI systems.” Between 2019 and 2023, Microsoft is reported to have invested an additional $2 billion, although this sum was never confirmed by either party.

Fast forward to January 2023. Microsoft announced “the third phase” of their “long-term partnership with OpenAI through a multiyear, multibillion dollar investment” of up to $10 billion. This sizeable investment also involved Microsoft deploying OpenAI’s models in Microsoft’s products, as exemplified by Microsoft 365 Copilot.

While Microsoft was not explicitly afforded governance rights over OpenAI, the realities of this “extended” partnership seemed to indicate otherwise. Following OpenAI’s leadership debacle in late November 2023, Microsoft initially agreed to hire Sam Altman alongside other employees and then pushed OpenAI’s board to reinstate him as CEO. It is worth asking to what extent the move was triggered by Microsoft itself or rather by Satya Nadella. The truth is, given the cult of personalities that mark the leadership of both OpenAI and Microsoft, it is often hard to tell when new developments are prompted by OpenAI and Microsoft, and when they are the actions of Sam Altman and Satya Nadella. Bracketing these observations, what we know for a fact is that after Altman took back his position and most of the board was dissolved, Microsoft obtained a non-voting observer seat.

It was at this moment that antitrust agencies began to worry. The UK’s Competition and Markets Authority (“CMA”) opened an investigation in early December 2023 to assess whether the partnership between the two companies effectively functions as a merger. The formal decision came out less than two weeks ago, rejecting that the partnership gives rise to a merger. The full text is not yet available, but the summary notes that “[i]n view of Microsoft’s potentially important role in securing Sam Altman’s re-appointment, the CMA believed there was a reasonable chance that an investigation would reveal that Microsoft had increased its control over OpenAI’s commercial policy.” Additionally, according to the CMA, “Microsoft has acknowledged during the course of [the] investigation that it has held the ability to materially influence OpenAI’s policy since 2019.”

Despite the fact that the investigation was opened in late 2023, the decision builds heavily on the post-2023 evolution of the partnership. A series of factors are notable. First, in July 2024 Microsoft dropped its observer seat on OpenAI’s board. Second, OpenAI’s reliance on Microsoft for computing power has been gradually reduced. As of June 2024, OpenAI has partnered with Oracle to be able to tap into more computing capacity. Rather than an end to the exclusivity clause with Microsoft, this collaboration was allowed as an exception. The agreement was presented as a three-way deal between Microsoft, OpenAI and OCI targeted at enabling OpenAI to “use the Microsoft Azure AI platform on Oracle’s infrastructure.” Nonetheless, commentators have argued that the lack of sufficient compute had given rise to tensions between OpenAI and Microsoft, and that the latter had to be pressured by its investors to accept the deal.

This was only the beginning of OpenAI loosening its ties with Microsoft. On January 21st, 2025, Microsoft announced yet another “next phase” of their partnership with OpenAI. Notably, the new iteration “includes changes to the exclusivity on new [compute] capacity, moving to a model where Microsoft has a right of first refusal.” Consequently, OpenAI can seek computing power elsewhere, provided that it consults with Microsoft first for additional capacity. On the same day, OpenAI announced its participation in the Stargate Project, “a new company which intends to invest $500 billion over the next four years building new AI infrastructure for OpenAI in the United States.” Whereas the announcement highlighted that it “builds on the existing OpenAI partnership with Microsoft,” Stargate involves a broader collaboration with Oracle and NVIDIA.

A Partnership Beyond Control

Given the gradual decrease of Microsoft’s influence over OpenAI, the CMA was right to decide that the partnership does not give rise to a merger. However, this does not mean that all competition concerns have been dispelled. But before we elaborate on that, it is important to clarify why the merger control framework was an intuitive first choice for competition authorities worried about the impact of the partnership on the emerging AI landscape.

The initial dependence of OpenAI on Microsoft’s compute, which represents one of the key inputs required to develop and deploy AI, raised justified concerns about the former’s ability to deviate from the latter’s interests. Furthermore, despite the absence of formal governance rights, the partnership involves close collaboration between the two companies. In addition to revenue sharing agreements that flow both ways, senior executives from both companies meet regularly, and their employees collaborate on “jointly building AI supercomputing technologies.”

Given these dynamics, reliance on merger control was justified for two reasons. On the one hand, at a first sight, the biggest implications of this partnership seemed to be on the bilateral relationship between the two companies. Faced with a scenario where a Big Tech player provides key inputs and intervenes in the internal governance of what was at the time a small start-up, the key question appeared to be whether the big player controls the small one. On the other hand, the merger investigation allowed the CMA to obtain more information about the granularities of the deal, which fed into its final decision.

The CMA was not alone in its choice. The German Competition Authority had opened and closed an investigation in the Microsoft-OpenAI partnership in late 2023. The European Commission had similarly announced in June 2024 that Microsoft had not acquired control on a lasting basis over OpenAI, and that consequently, the partnership did not qualify as a merger.

Observations from the Microsoft-Open AI partnership thus far warrant a closer look at other burgeoning relationships in this rather tight-knit ecosystem. To begin, OpenAI’s recent investment into CoreWeave, a cloud service provider, shifts the power dynamics of the Microsoft-OpenAI deal. Microsoft is CoreWeave’s biggest client. Pursuant to this deal, OpenAI will “not only have access to the same cloud, but it will also have an ownership stake in the company that runs it.” This means not only does OpenAI have its own independent investment strategies, but also that they potentially will have a significant voice in the governance of one of Microsoft’s key suppliers.

Additionally, Microsoft’s investment in OpenAI is not exclusive, either. Looking at the investments of Alphabet, Amazon, Apple, Meta, Microsoft, and NVIDIA, the CMA had identified last year “an interconnected web of 90 partnerships” with AI developers. As the CMA rightly points out, we should remain “vigilant against the possibility that incumbent firms may try to use partnerships and investments to quash competitive threats.”

The biggest concerns these partnerships raise, however, are systemic rather than bilateral. Merger control should neither be our default nor primary tool to address these issues. In a report published in January this year, the FTC notes that the effects of these partnerships are felt beyond the collaborating parties. The partnerships may lead to cloud service providers “limiting access to inputs such as computing resources for AI developers other than [their] partners” and “could affect the availability of AI developer engineering talent.” Furthermore, the partnerships “provide partners access to sensitive technical and business information that may be unavailable to others.”

From Bilateral Relationships to Market Structures

These dynamics showcase an important blind spot of existing antitrust policy. There is no meaningful existing tool to address inter-firm collaborations that are neither mergers (or joint ventures) nor cartels, and it is not obvious how such a tool should look like. What we are dealing with here is not an isolated series of bilateral deals, but a new market structure where collaboration, rather than competition, is becoming the norm. But, this is not necessarily a bad thing.

Collaborative market structures have become a feature of highly innovative industries in the last decades of the past century. Looking at the biotech industry in the 1990s, Walter Powell identified a sharp increase in alliances between biotech companies. Moreover, Powell identified a “liability of disconnectedness,” meaning that firms that were less connected grew at slower rates due to their inability to benefit from the know-how of their peers. Of the eight biotech firms with the highest market value in 1994, four were also featured in Powell’s index of the top eleven firms with the most collaborative R&D activities. One of these companies was involved in more than 100 small-scale collaborations.

Industrial organization economics teaches us that collaborative market structures come with important cost savings. They can reduce entry barriers by allowing firms to tap into the assets of existing market players. In the case of generative AI, partnerships enable AI start-ups to tap into the computing power of hyperscalers. Moreover, collaboration leads to know-how spillovers, which can speed up the rate of innovation. From an antitrust perspective, allowing firms to collaborate can be an effective merger control strategy: if companies can tap into each other’s assets without risking antitrust scrutiny, their incentives to merge are reduced.

The question is not whether collaboration should be allowed, but what type of collaboration. If the web of partnerships was open to any AI start-up seeking to benefit from the computing power of existing Big Tech players and if these deals were not exclusive, we would have little to worry about. Put simply, it is not the existence of these collaborations that should raise antitrust concerns, but their design.