What happened
Carnegie Endowment's analysis examines how middle powers can survive the AI race, building a matrix on two variables — the speed of frontier progress and the 'frontier premium' (the gap between the best systems and open-source or other cheap alternatives) — that defines four possible worlds, each with its own best strategy. The paper notes that most governments and analysts advocate strategies (sovereign models, pooling resources, chokepoints, faster adoption, new technological directions) in isolation rather than across the full range of futures, and argues that finding leverage at supply-chain chokepoints is valuable in almost every technological future and most valuable where AI poses the largest risks to middle powers' economic and political standing. It cites the June 2026 U.S. move revoking foreign access to Claude Mythos and Fable and restricting OpenAI's GPT-5.6 as evidence of the emerging access-restriction regime.
Why it matters
For government and corporate strategists assessing AI dependency, export-control exposure, and access risk, this maps how access to frontier models may be curtailed and which positions remain robust across uncertain futures.
Action needed
Assess your organisation's exposure as a middle-power AI consumer or producer and identify supply-chain leverage or hedging options.