AI Sovereignty Is Mostly a Negotiation Over Chokepoints
Carnegie gives middle powers four AI futures. In nearly all of them, national leverage comes from controlling something the frontier labs cannot casually route around.
AI-generated conceptual editorial illustration. It is not documentary evidence of any country’s infrastructure or model access.
Two variables, four worlds
Carnegie’s Alasdair Phillips-Robins, Teddy Tawil and Sam Winter-Levy organize the AI race around two uncertainties: how fast frontier systems improve, and how large the “frontier premium” remains over cheaper or open alternatives.[1]
Fast progress plus a large gap produces a chokepoint world, where a few firms and states can ration model access. Fast progress with a smaller gap favors rapid adoption. Slower progress and a large gap produces tiered access. Slower progress with a smaller gap gives countries more room to build or adapt systems on their own terms.[1]
The matrix matters because governments are currently shopping for “sovereign AI” as if the phrase itself were infrastructure. The useful policy depends on which world arrives.
Leverage beats the commemorative model launch
The authors’ most durable recommendation is to secure bargaining power in the supply chain. That can mean advanced semiconductor capacity, lithography, energy, data-center sites, talent, security relationships or another resource the leading labs actually need.[1]
Countries without a true chokepoint can still pursue adoption, local-language systems, pooled research or compute hosting. But a national model that depends on foreign chips, foreign cloud and foreign updates is sovereignty with a very attentive landlord.
The paper is careful to call its four futures scenarios. They are not probability estimates. The “frontier premium” could also change quickly as model techniques diffuse, regulation shifts or cheap systems catch up on commercially useful tasks.[1]
Our read
The AI sovereignty debate often begins with branding and ends with a procurement ceremony. Carnegie starts in a better place: what can your country withhold, trade or make costly to replace?
That is a harsher question, but it is the geopolitical one. Access promises are policy until the leading state changes its policy. Compute contracts are commercial until export controls arrive. A regional data-center boom creates leverage only if it is scarce enough to matter and the host government is willing to use it.
Middle powers should hedge across scenarios: build adoption capacity, protect talent, map dependencies and invest where they can create real interdependence. The objective is not a flag on a chatbot. It is enough leverage that the next cutoff requires a negotiation.
Caveat: The four-quadrant framework is a disciplined thought experiment rather than an empirical forecast. It simplifies several linked variables, and the countries shown in its investment map are illustrative. Carnegie’s technology program approaches the problem through governance and U.S.-allied strategy; policy outcomes will also depend on domestic politics, firm behavior and technical changes the scenarios cannot settle.[1]
FOLLOW THE SOURCES.
Checked Oct 30, 2026. Numbered references point to the evidence behind factual passages. Commentary is our own.
Texas and Georgia elected governors who campaigned to slow or pause new data centers. The backlash over power, water and local control just graduated from hearings to executive offices.