Sovereignty Intelligence

The Five-Layer Framework for reading sovereign AI capability

Capital announcements are not sovereignty. This is the map for what is.

Why capital numbers mislead

A nation committing $100B to AI infrastructure is not necessarily sovereign over AI. Sovereignty is not a single condition — it is three to five distinct layers, and most nations have one or two, not all of them.

The Five-Layer Framework was developed to make this gap structure legible. It does not rank nations by ambition or spending. It maps which control mechanisms are actually present, which are absent, and which are emerging.

The framework is stable. The data changes every tracker cycle. This page explains the framework. For current state, see the live surface below.

The Five Layers

Layer 1

Capacity Sovereignty

Does the nation control compute and models? Domestically-developed foundation models running on owned or controlled infrastructure. The most demanding layer — China, South Korea, France, India meet it fully. Most others are building toward it.

Layer 2

Regulatory Sovereignty

Can the nation set the rules of the game? AI acts, governance frameworks, enforcement mechanisms. The EU AI Act is the current benchmark. Regulatory sovereignty creates compliance moats — or friction that slows deployment. It can soften under pressure.

Layer 3

Capital-Selection Sovereignty

Is the state directing which infrastructure gets built and which actors get backed? SWF-driven investment (Saudi PIF → HUMAIN), government fund cohorts (UK Sovereign AI Fund), academic compute programs (Canada SCIP). Money allocated is not the same as money directed.

Layer 4 — Emerging

Protocol Sovereignty

Can the nation define the technical standards for how AI systems interact? MCP/A2A interoperability, governance frameworks for agentic AI. Currently held by one jurisdiction. Not yet binding globally. The amber signal: whoever leads here before the standard sets gains structural advantage.

Layer 5

Loop Closed

The feedback condition: domestic model driving demand for domestic silicon driving domestic deployment, tightening under pressure rather than merely announcing. Confirmed in one case only. This is not a layer that can be built — it has to emerge from the other four becoming mutually reinforcing.

Key structural readings

These readings are framework-level observations. They explain what the current data means, not what the current data says. The current data is in the live matrix.

China

The only confirmed case of all five layers present, with the loop closed. The DeepSeek V4 → Ascend 950 demand loop is not product news — it is ecosystem closure. Export controls accelerated this convergence rather than preventing it. Every other node has at least one absent layer.

Singapore

The sole Protocol Sovereignty first-mover. IMDA's agentic AI governance framework (Jan 2026) is the world's first jurisdiction-level governance standard for agentic systems. Combined with capacity (S$1B NAIRD) and regulatory layer, Singapore holds three layers — the second-strongest position outside China. The protocol layer gives Singapore a structural window before standards bind globally.

Saudi Arabia

The largest committed capital in the dataset ($115B+, Project Transcendence) with the thinnest sovereignty structure. Full Capital-Selection — the PIF directs infrastructure at scale. But Capacity is under construction (not yet operational), Regulatory is absent (no AI governance framework), Protocol is absent, Loop is open. Capital-selection without the other layers is resource deployment, not sovereignty.

Germany

Partial Capacity (SOOFI operational, Q3 2026 model target) and Full Regulatory (EU AI Act, strictest framework outside China). Two layers confirmed, building a third through the Cohere-Aleph Alpha merger. The gap: no loop, no protocol layer. The question for Germany is whether the Cohere/Pharia commercial track and the SOOFI open-source track can reinforce rather than compete with each other.

The Protocol column

One occupied cell out of 23 nodes. This is not a measurement failure — the standard does not yet exist in binding form. The amber color in the matrix signals a different epistemic category: first-mover position in a layer that isn't yet binding. When MCP/A2A standards converge and governance frameworks propagate, the Protocol column will have candidates. Singapore holds the position before the race begins.

Live tracker surface

The framework above is static. Current node state, bloc analysis, and the visual matrix are on sovereignfields.org — updated each tracker cycle.

Sovereignty Layer Matrix

Five-layer grid across 23 nodes. Visual representation of gap structure, updated 2026-05-01.

→ sovereignfields.org

Live Tracker Data

23 nodes, 52 events, schema v1.5. Machine-readable JSON updated each cycle.

→ sovereignfields.org

sovereignty-ai.org is the registry and framework layer. sovereignfields.org is the live intelligence surface. The two-domain architecture is intentional.