Are Big Tech’s Quasi-Mergers With AI Startups Anticompetitive?

By: Alexandros Kazimirov (ProMarket)

In this piece, author Alexandros Kazimirov (ProMarket) explores how the recent rise of companies developing large language models (LLMs) has triggered aggressive competitive responses from incumbent tech giants. These incumbents, seeking to preserve their dominance, have engaged in quasi-mergers—transactions that blend employee acquihires with IP licensing deals. Startups often survive these deals nominally, but lose most of their key personnel to the acquiring firms. The licensing component is used to buy out investors and sustain the shell of the startup. While these deals can enhance the incumbents’ innovation capacity, they may also suppress potential disruptors in a manner resembling so-called killer acquisitions, raising anticompetitive concerns.

Between March and August 2024, three notable quasi-mergers occurred: Google with Character, Microsoft with Inflection, and Amazon with Adept. All three followed a familiar pattern—non-exclusive licensing of IP, absorption of startup employees, and a shift in the startups’ roles from core LLM development to downstream applications. For instance, Character and Inflection pivoted to LLM user interfaces, while Adept focused on agentic tools. Meanwhile, the incumbents quickly launched their own LLMs, led by the very engineers they had hired. This pattern raises the question of whether the startups were strategically redirected—or deliberately sidelined.

To challenge these deals under Section 7 of the Clayton Act, authorities would need to define relevant markets, demonstrate probable harm, and prove causation—a difficult task in the fluid AI landscape. LLMs and search engines overlap and complement one another, making static market boundaries hard to draw. The uncertainty around technology evolution and startup potential complicates any attempt to isolate harm. Kazimirov argues that a better enforcement strategy would assess these deals collectively, weighing circumstantial evidence like licensing terms, pricing, and market behavior. Under this framework, Google’s transaction appears most anticompetitive, Amazon’s the least, and Microsoft’s somewhere in between—suggesting a more strategic, risk-weighted approach to antitrust enforcement.

CONTINUE READING…