AI Access Is Not Enough: Middle Powers Need Strategic Reserves
Researcher: Kasia Jakimowicz
Mentored by: Prof. Robert Trager, Charles Martinet
From the project UK and international AI digital sovereignty, Q1 2026 AI Safety Fellowship
Summary
Middle powers increasingly depend on frontier AI systems they neither own nor control. Policy debate has equated AI sovereignty with the physical location of compute, but hardware sited domestically can still be governed from abroad through vendor licensing, remote management, and legal instruments such as the US CLOUD Act. Countries including the UK, France and Canada hold real AI capability and still cannot independently develop frontier models. As governments wire frontier AI into critical infrastructure, a geopolitical disruption — the paper draws the parallel with Russia throttling gas supplies to Europe in 2022 — could sever essential services with no reliable fallback. The paper proposes AI strategic reserves: dependency audits that identify where frontier AI underpins critical functions and what failure would look like; break-glass capabilities, including contingency access arrangements and locally held fallback models for essential services; and reserve capacity pooled among allied nations under agreed emergency protocols. Reserves address the immediate vulnerability while longer-term sovereignty initiatives mature.