The Tony Blair Institute presents a compelling route to AI sovereignty through open ecosystems. A wider architectural choice changes the strategic equation: middle powers can reduce how much critical national and commercial activity depends on frontier models, while building valuable forms of intelligence around their own knowledge, institutions and strengths.

The AI sovereignty debate is often framed as a contest over who controls the largest models, the compute needed to train and run them, and the infrastructure beneath them.

The Tony Blair Institute’s paper, Open Source: How Middle Powers Can Build Influence in the Age of AI, offers a more productive path. It argues that middle powers do not need to match the scale of American and Chinese frontier laboratories to capture the economic and strategic value of AI. Their opportunity lies downstream: in the ecosystems that adapt, govern and apply AI across industry, science and public services.

This moves the goal from owning a national foundation model to possessing sovereign capability and the practical ability to use AI on national terms, retain meaningful choices and create domestic value.

TBI’s argument invites a deeper strategic choice.

Open ecosystems give countries greater choice among foundation models, providers and tools.

Emerging architectures such as Neuro-Symbolic AI empower a choice as to where foundation models are not needed at all and whether different, sovereign, governed forms of intelligence are better suited for critical decisions.

AI is larger than foundation models

Foundation models are remarkable general-purpose systems. They bring powerful capabilities to language, discovery, synthesis, content generation and human interaction. They will be an important part of national and commercial AI estates.

Their success has also created a foundation-model default: an assumption that most intelligent activity will eventually pass through an LLM or related general-purpose model.

That default concentrates dependency on a small number of model families, compute platforms, cloud providers and technical standards. It places growing volumes of operational activity inside foundation models whose outputs can be difficult to explain and reproduce consistently, and whose behaviour cannot be governed with the precision consequential decisions demand.

The vast majority of sustainable value generated by the use of AI will occur when AI reaches deeply into material processes and consequential decisions across public services, healthcare, energy, finance, infrastructure, business and security. These environments require more than the fluent or plausible outputs foundation models produce. They require explicit policy, stable constraints, causal understanding, traceable evidence and clear accountability.

Architectural choice changes the strategic equation

Dependency on AI has two dimensions.

Quantitative dependency concerns how much national and commercial activity relies on foundation models.

Qualitative dependency concerns the importance of that activity: whether foundation models support peripheral tasks or sit inside the critical path of decisions affecting citizens, infrastructure, capital and security.

Sovereignty also depends on control of data, infrastructure, supply chains, skills, licences and operational capability. Architecture matters because it determines how those dependencies combine, how deeply they reach into critical activity and how readily they can be changed.

Open-model ecosystems reduce dependence on any single supplier.

Architectural diversity reduces both the scale and strategic importance of that underlying dependence.

Foundation models should remain the choice where their distinctive strengths create value - interpreting language, exploring information, generating content and supporting natural interaction.

Consequential reasoning should sit within architectures designed for trust, consistency, policy fidelity, evidence and institutional control.

This separation creates true resilience. True resilience occurs when intelligence can be deployed in architectures that are unaffected by foundation-model provider change, withdrawal or new model introduction. Where institutional knowledge, policies and controls governing a critical process are detached from and independent of foundation models.

Consequential reasoning belongs in intelligence architectures that operate under the full authority and control of the institutions accountable for outcomes. Domain knowledge, policy, causal understanding and decision logic must remain assets those institutions own and on their full control.

Moving decision intelligence to capable sovereign architectures greatly reduces and switching costs of foundation models and hence the dependance on those models. This materially shifts the power dynamics and risks of AI.

Hybrid Intelligence as an example

Hybrid Intelligence is UMNAI’s full-stack neuro-symbolic architecture. It is a leading example of an alternative architectural choice for intelligence applied to consequential decisions and processes.

Hybrid Intelligence combines learned intelligence, symbolic reasoning, causal structure, explicit human knowledge and deterministic governance. Learned components identify patterns and anticipate what happens next. Symbolic reasoning applies explicit knowledge and logic. Causal structures make explicit how interventions and events are expected to influence outcomes, allowing those assumptions to be tested against evidence and domain expertise. A deterministic core imposes defined policies and guardrails. An unbroken evidence chain connects the information entering the system to the decision produced.

The architecture learns and operates entirely within the institution’s environment. Data, knowledge, policies, controls and decision records remain inside its chosen operational boundary, with no egress to external model providers or other third parties.

Human expertise, policies, constraints and causal assumptions persist in the architecture. Institutions have full and direct control over them. Models are retrained or evolved without surrendering the wider knowledge and decision structure. Every material decision remains reproducible and examinable in the context in which it was made.

Foundation models can still contribute where they add value. They may interpret a request, extract information or explain an outcome. Hybrid Intelligence retains authority over the reasoning, decision and permitted action.

In lending, for example, an LLM may interpret unstructured evidence or communicate a decision. The decision to extend credit or intervene in a vulnerable customer journey remains within the governed architecture.

This places foundation models outside the authoritative decision path without excluding them from the wider system. Institutions gain the benefits of foundation models offloading the dependencies, risks and exposure.

A different advantage for middle powers

The frontier-model race rewards capital, compute, data and engineering scale.

Architectural diversity depreciates that advantage and brings other national strengths into play.

Middle powers often possess deep scientific capability, mature legal and regulatory systems, trusted institutions and substantial expertise in complex industries and public services. Hybrid Intelligence provides a sovereign infrastructure to that expertise, policy and causal understanding into explicit, governed and reusable intelligence assets.

This creates a different field of competition. Countries can build specialised decision systems grounded in their own knowledge and standards, operate them within sovereign boundaries and export them to markets with comparable needs.

Invest in architectural choice

Middle powers have strategic choices.

They can concentrate their efforts on competing with or adapting to foundation models developed elsewhere. They can build a more diverse intelligence economy in which foundation models work alongside causal, symbolic, neuro-symbolic and other specialised architectures.

Realising the second path requires deliberate support and investment.

Governments, research funders and public institutions can:

• fund research and commercial development across emerging AI architectures;
• incorporate architectural diversity in national AI strategies, testbeds and growth programmes;
• use procurement to reward policy fidelity, evidence, reproducibility, modularity and local operational control;
• treat institutional knowledge, regulation and causal expertise as sovereign assets alongside data and compute;
• create benchmarks for governed decisions;
• back domestic companies turning these architectures into deployable and exportable applications.

This is an industrial opportunity as well as a sovereignty agenda. Early investment will help middle powers build companies, intellectual property, skills and standards in areas where the frontier-model race won’t reach.

TBI is right that middle powers can gain influence without winning yesterday’s race.

Sovereign AI begins with architectural choice.