Paper by Talia Caplan and Bilal A Mateen: “In 1600, Queen Elizabeth I granted the East India Company a royal charter and a monopoly over trade across much of the known world. For more than a century, the relationship was mutually advantageous: the Crown received revenue, access to rare commodities, and unparalleled geopolitical reach; the Company received protection, legitimacy, and the coercive backing of a state. By the mid-18th century, it became something very different. The Company maintained its own army, territories, and foreign policy, and increasingly operated in ways that were counter to the objectives of the Crown, resulting at times (due to its unconstrained, singular objective of maximising profits) in the deaths of millions.
Today, the world’s largest technology firms are at the precipice of a similar transition: towards becoming firms that function as corporate sovereigns whose power derives from the control of indispensable digital infrastructure and the algorithms that shape the attention economy, rather than from physical territory.2 The concentration of power is stark. A single firm (Nvidia) supplies between 80 and 90% of the chips on which advanced artificial intelligence (AI) models are trained; a handful of hyperscale providers (Microsoft Azure, Amazon Web Services, Google Cloud) operate the data centres in which those chips run; and a small number of laboratories (eg, OpenAI, Anthropic, Google DeepMind, DeepSeek), almost all American-owned, use those data centres to produce the frontier AI models on which others build.
These layers are less separate than they appear. A dense web of cross-investment now binds them, with chipmakers taking equity in the laboratories that buy their processors, and cloud providers and laboratories acquiring stakes in one another while committing to purchase each other’s services; analysts have estimated that interlocking arrangements of this type are worth more than US$800 billion. The effect is a tendency towards vertical integration achieved through finance rather than a formal merger, drawing nominally competing entities into a single, mutually dependent interest.
We expect that two structural features will characterise AI firms’ transition to pseudo-sovereign status. These firms will increasingly occupy every position in their own oversight, building the systems, funding the safety evidence (based on evidence from a preprint), staffing the advisory bodies, and drafting the standards meant to constrain them. They will also become jurisdictionally mobile in ways territorial regulators and sovereign governments are not, using digital infrastructure and complex corporate structures to obfuscate the residency of data and software in an effort to escape true oversight.
In that context, a health system that adopts a frontier AI solution for triage, imaging, or clinical decision support—such as the Horizon 1000 initiative seeks to roll out in Rwanda—does not necessarily acquire a substitutable tool but rather creates a dependency on a supply chain it neither controls nor can reproduce. The risks are especially acute in the global health context because upfront philanthropic support, especially when provided by the frontier AI companies that stand to benefit most, increases the likelihood of three specific risks…(More)”.
