Who Controls the Future of Generative AI? Infrastructural Power and the Contracts Behind Big Tech–Startup Partnerships
The development of Generative AI has been largely driven by strategic partnerships between large technology companies and startups. A growing debate asks whether these alliances amount to Big Tech control over the trajectory of AI innovation, and legal scholarship has responded through two frames: novel corporate structures designed to insulate startups from market pressure, and antitrust scrutiny of the partnerships themselves. This Article argues that both frames provide only a partial account of how control operates. Drawing on case studies of the three flagship partnerships—OpenAI–Microsoft, Anthropic–Amazon, and Mistral–Microsoft—the article shows that control over Generative AI is exercised less through ownership or merger than through a combination of infrastructural power, non-equity investments, and contractual arrangements. Infrastructural power operates through conditional access to the complementary assets required to innovate at scale—compute, cloud, data, and distribution. Non-equity investments give incumbents indirect channels of influence and contractual arrangements often create technological dependence and give incumbents discretion over the continuation and terms of access to complementary assets. The Article advances a two-part framework in response. First, undermining structural dominance requires tools that disseminate access to complementary assets beyond case-by-case antitrust enforcement, including regulatory antimonopoly instruments and public AI infrastructure. Second, because collaboration will persist even where dependency is reduced—owing both to learning, experimentation, and reputational signaling, and to the fact that antitrust law will not always prevent or unwind such arrangements—the Article develops a framework of residual innovation governance, centered on contractual design, and of intermediated innovation governance, for settings of asymmetric bargaining power. The framework offers courts, agencies, and policymakers an analytical lens for evaluating AI partnerships that neither presumes collaboration is benign nor treats it as unlawful.
Author: Pablo Marcello Baquero
Volume 28, Issue 1