Compare AI-driven power concentration to a historical case where one nation’s technological advantage enabled it to exert asymmetric economic or political pressure on others. What made that leverage possible, and what (if anything) limited it?
A close historical analogue for AI-driven concentration might be Britain’s command of the global submarine telegraph network from the late nineteenth century into the First World War. Like AI (now or soon), telegraphy was a general-purpose technology with implications for diplomacy, finance, commerce, and war planning. By building the “All-Red Line”, an empire-spanning cable system that preferentially landed on British territory, Britain was able to leverage its technological lead into a position at the centre of the world’s information ecosystem.
That centrality gave Britain the ability to decide whose messages travelled cheaply, securely, and quickly, and whose did not. They were able to exploit this advantage when conflict arrived in August 1914, cutting Germany’s undersea cables and forcing German long-distance traffic onto routes Britain could monitor, censor, or deny. This informational dominance in turn shaped battlefield coordination, intelligence collection, even the informational environment in neutral countries whose news and diplomatic traffic now transited British-controlled infrastructure. Notably, the coercive instruments wasn’t just superior ships or guns, but the ability to grant or withhold access to a critical network.
Several features (shared by AI) made that leverage possible. First, cables were capital-intensive, with strong network effects: once one network achieved global scale, everyone was incentivised to route through it, further entrenching the hub. Second, the infrastructure relied on scarce complements (landing rights, relay stations, cable-laying, repair capacity, etc.) that Britain controlled, via both industrial capability and imperial geography. Third, governance was embedded in the network itself, rather than by formal treaties: “rules” were implemented by routing, prices, and censorship decisions, so the coercion was flexible and deniable.
Of course, the limit to this leverage was substitution and diffusion. Wireless radio reduced dependence on cable chokepoints (even if it introduced new vulnerabilities), rival powers and firms built alternative routes, and the political legitimacy of information control eroded as other states recognised the strategic risk of relying on a foreign hub. Britain’s advantage lasted as long as alternatives were costly or inferior, and the benefits of integration outweighed the sovereignty risk for others.
As I noted, frontier AI concentration has the same structural components: massive fixed costs (here, compute, talent, data, infrastructure), steep scale economies, and network effects that make the first few frontier ecosystems disproportionately attractive. If frontier capabilities remain concentrated in the US or China, other states risk becoming “model-taker” economies, whose productivity and state capacity hinge on continued access to foreign compute and models, even as this access can be throttled, conditioned on regulatory or geopolitical alignment, or withdrawn at moments of friction. The analogue to cable-cutting, clearly, is denial of service, export controls on critical inputs, standards-setting that locks in dependencies, and contractual governance of foundational digital infrastructure.
The historical lesson is not that dependence is always bad (telegraph integration created enormous surplus) but that losses of sovereignty become especially salient when a general-purpose technically is both indispensable and centrally controlled. The policy implication is equally clear: countries wanting to preserve economic sovereignty must reduce single-hub dependence (multi-sourcing and portability), invest in credible domestic or allied “reserve capacity” (compute and talent), and possibly push for international governance that treats frontier AI access and standards as critical infrastructure, rather than ordinary trade. This entails mapping relevant chokepoints, quantifying substitutability, and designing redundancy before crisis conditions force improvisation. Britain’s cable era shows how quickly a technological lead can harden into durable, asymmetric influence, and how that influence weakens once substitutes, redundancy, and shared rules emerge.