AI Sovereignty Requires Managed Interdependence and Strategic Use of Smaller Models

The article contends that renting access to proprietary frontier models does not constitute sovereignty, as providers control the rules, pricing, and service availability. Instead, sovereignty is defined as the practical ability to choose, adapt, govern, and replace AI systems. Citing the 2026 Stanford AI Index, the author notes that the performance gap between leading US and Chinese models has narrowed to low single digits (e.g., Claude Opus 4.6 leading Dola-Seed-2.0 Preview by 2.7%), creating a more plural landscape. The piece advocates for a portfolio approach where smaller, specialized models are used for routine or sensitive tasks due to their ease of governance, lower cost, and local hosting capabilities. It emphasizes that open weights provide leverage and options, not automatic sovereignty, and calls for investments in Australian-hosted inference, independent evaluation, and exit strategies to ensure institutions can govern AI without surrendering control.