Title page 1

Contents 2

1. Introduction 3

2. The AI duopoly and the N-2 problem 5

2.1. Two full-stack sovereignty strategies 5

2.2. Why AI dependence exceeds monetary dependence 6

3. AI sovereignty: two literatures and a synthesis 7

3.1. The hard-sovereignty literature: sovereignty as productive autonomy 7

3.2. The soft-sovereignty literature: sovereignty as governance capacity 8

3.3. Programmable sovereignty: a bridging definition 9

3.4. Seven mechanisms 10

4. Measuring hard and programmable sovereignty 11

4.1. State of play 11

5. Coordinated specialisation: from individual insufficiency to coalition sufficiency 16

6. Institutional precedents and their limits 19

6.1. EuroHPC: pooled infrastructure under shared governance 19

6.2. India Stack: programmable architecture and its limits 20

6.3. The ASEAN Digital Economy Framework: regional governance as a partial building block 20

7. Policy implications 21

7.1. Lead with governance, not production 21

7.2. Anchor governance with a minimal hard fallback 22

7.3. Organise around demonstrated comparative advantage 22

7.4. Put governance in place before infrastructure 22

7.5. Make programmable sovereignty empirically measurable 22

8. Conclusion 23

References 24

Appendix: Coalition feasibility, supporting data 27

Table 1. Hard AI sovereignty (who builds and owns the stack) 14

Table 2. Programmable sovereignty (who can govern AI without building it) 15

Figure 1. Numbers of AI related patents 11

Figure 2. Share of global AI GPU-cluster performance (% of total), 2026 12

Figure 3. Critical mineral processing concentration, 2020 and 2024 12

Figure 4. Total installed electricity capacity 13

Table A1. Principal coalition sovereign wealth funds (2025) 27

Table A2. Cumulative AI investment by selected coalition members to 2024 28