The Federal Trade Commission’s case against Meta for monopolizing personal social media through its acquisitions of Instagram and WhatsApp serves as a warning of allowing Big Tech companies to acquire nascent competitors in the artificial intelligence market through quasi-mergers that dodge government scrutiny. Based on new research, Alexandros Kazimirov argues that antitrust agencies can look at a combination of circumstantial evidence, including market product proximity, price premiums and product discontinuation, to help adjust their approach to keep AI markets contestable, rather than trying to restore contestability ten years from now.
A new wave of emerging companies developing large language models (LLMs) has unleashed fierce competition in artificial intelligence markets. These firms have significant innovative capabilities that threaten the technological dominance of incumbent tech companies. To protect themselves, the incumbents have responded by partnering with leading product developers like OpenAI and Anthropic and subsuming smaller startups through quasi-mergers.
A quasi-merger combines an acquihire, in which one firm acquires another for its employees and their expertise rather than its product, with a licensing deal. Like an acquihire, some or all of the startup’s employees depart to join the incumbent. Unlike an acquihire, however, the emerging company survives the acquisition. The incumbents then use license agreements not because they are interested in the technology per se, but as means to cash out the startup’s venture capital investors, with part of the licensing fee serving as a lifeline to the surviving entity. This new transactional device has the potential to be both cooptive and synergistic. The hiring of software engineers can improve the incumbent’s products and enhance its innovation processes, but it can also preempt an emergent from becoming a force of disruption, resembling a killer acquisition. This is where the anticompetitive concerns arise.
From March to August of 2024, three AI startups—Character, Inflection and Adept—concluded agreements with Google, Microsoft and Amazon, respectively. The terms of these agreements followed a similar pattern: the incumbents took a non-exclusive license of the startups’ intellectual property, hired most or all of their employees, and forced them to compete further downstream. Before the quasi-merger, these emergents were developing LLMs and end-user applications, but they all ended up confining themselves in the application market after the quasi-merger. Character and Inflection would focus on developing user interfaces for LLMs, and Adept on agentic solutions. Meanwhile, a few months after these transactions, the incumbents released their own LLMs after assigning the hired software engineers to lead internal AI units. Perhaps the startups needed to readjust their purpose in a changing market. Perhaps they were purposely displaced.
Were these acquisitions anticompetitive? To successfully challenge one of these transactions under Section 7 of the Clayton Act, the government would have to establish market definition, a reasonable probability of harm, and causation. The question about market definition is inherently tied to the purpose of the technology. Specifically, are LLMs meant to replace search engines, or amplify them?
LLMs and search engines are partly interchangeable and partly complementary. If a user is seeking access to information, the query lies within the search engine’s domain. In contrast, a user prompt seeking to contextualize information or synthesize content likely lies within the LLM’s domain. Overlap in more open-ended questions obscures domain demarcation. This obscurity illustrates the interrelatedness of the two systems and the inherent arbitrariness in making static judgments about an evolving technology. Furthermore, reviewing transactions in isolation, whereby provable causation is shown by a specific act, sets a high threshold that cannot be met even with ample circumstantial evidence. And in nascent competition, even the relative probability of a quasi-merger being harmful to competition is difficult to assess.
Consequently, some adjustments are needed. Rather than scrutinizing each acquisition in isolation, a better approach may be to look at a series of similar transactions, simultaneously. Rather than trying to overcome the failing firm defense by speculating between two counterfactuals, the antitrust agencies can conduct contextual comparisons based on circumstantial evidence like market product proximity, exclusive licensing agreements, price premiums and product discontinuation. By following this approach, it becomes apparent that Google’s deal is on the higher end of the anticompetitive spectrum of probabilities, Amazon’s deal is on the lower end, and Microsoft’s deal is likely situated in the middle. And rather than maintaining agency inertia, the antitrust agencies can prioritize enforcement where relative risk is the highest, while keeping an eye on the viability of post-merger startups.
Redressing potential harms from quasi-mergers
Redressability of quasi-mergers presents two problems: one of form and one of substance. In terms of form, since the agreement involves two companies, which continue nominally independent of each other after the quasi-merger, the issue is whether a quasi-merger can be construed as an acquisition. One approach is to treat the agreement as a sale of (all or substantially all) assets, where the assets in question are software engineers. Another approach can be adapting the de facto merger doctrine to scrutinize an agreement that attempts to circumvent merger review by the antitrust enforcers.
In terms of substance, the fundamental problem is that injunctive relief cannot, and should not, be used to restrict employee mobility. In an important sense, that is the whole point of quasi-mergers—they are structured to shield the incumbent from antitrust enforcers. However, the lawyer behind this transactional device may have missed that if beauty lies in complexity, effectiveness comes with simplicity. In an attempt to shield an incumbent from government scrutiny, the structure of a quasi-merger is misaligned with acquisition incentives. It withholds the cherished “exit” acclaim from founders. It deflates the expectation of outsized returns for venture capital investors. It splits the startup’s employees into two groups, abandoning one in a lingering entity with an uncertain future.
In the cases of Character, Inflection and Adept, a quasi-merger also indicates the incumbent’s desire to transition from third-party product integration to in-house product development. But when this transition is not followed by detachment from existing partnerships, the government can aptly enforce it. If quasi-mergers thin the herd by displacing laggards, while partnerships coopt frontrunners from displacing incumbents, then the agencies should focus on targeting cloud purchase commitments for transactions that resemble vertical acquisitions, and governance or product agreements for transactions resembling horizontal acquisitions, to separate the incumbents from those leading frontrunners. Put simply, the incumbent’s dilemma should be the following: for every developer acquired through a quasi-merger, the incumbent should proportionately relinquish control over emerging companies, whose products it has previously relied on (e.g. Anthropic, OpenAI, etc.).
A less intrusive option is also available. Silicon Valley operates with a non-adversarial decorum. A gentlemen’s agreement between the founders and all other employees of the startup, which precludes partial desertions to incumbents without a viable long-term plan for the surviving entity or all investors and employees exiting together, could put in writing what is already the customary practice. Furthermore, it can function as a trust-building instrument between founders and their employees, without inhibiting the optionalities of the startup. And while the agreement’s non-binding character does not create new rights or obligations, founders who disregard the interests of their employees and continue to participate in such transactions will likely find it difficult to attract supporters in their next venture.
In the end, while quasi-mergers are intentionally designed to shield incumbents from government scrutiny, this challenge is not insurmountable. Antitrust enforcers can look beyond formalities and implement remedies that extend beyond transaction-specific solutions. And founders can be peer pressured to start looking out for their own.
Author Disclosure: The author reports no conflicts of interest. You can read our disclosure policy here.
Articles represent the opinions of their writers, not necessarily those of the University of Chicago, the Booth School of Business, or its faculty.
