AI sovereignty needs open infrastructure, not just a local chatbot

AI sovereignty is not achieved by swapping one foreign chatbot for another, argues an O’Reilly Radar essay published on 22 September. Its central point is sharper: governments and institutions need open infrastructure across the AI stack if they want genuine control and the ability to change providers.

Watch Desk analysis

What happened

The essay, AI Sovereignty: Bargaining with Big Tech and the Promise of Full Stack Open Source AI, argues that closed commercial models and even open-weight models can leave institutions dependent on US-based infrastructure and chips. Its proposed answer is a federated system of open source AI covering data protocols, inference engines and agent harnesses.

The useful distinction is between owning or accessing a model and controlling the machinery around it. A government may run a locally hosted model while still relying on overseas hardware, cloud services, software layers or data plumbing. That is local deployment, but it is not necessarily sovereignty.

Why it matters

The argument lands as countries court domestic AI providers and pour money into data centres. Building a national model can look impressive while quietly reproducing the dependencies it was meant to escape. The essay says institutions should sustain open tools at every layer, giving them more choice over providers and making it easier to replace a component when prices, policies or access change.

That is a concrete agenda, not merely a flag-waving slogan. It also comes with a large bill. Open infrastructure needs long-term funding, skilled maintainers, compatible standards and enough users to avoid becoming a beautifully independent ghost town.

Our read

This is a worthwhile policy argument because it moves the sovereignty debate away from model branding and towards the plumbing that makes AI useful. The hard question is not whether a country can announce its own model, but whether its institutions can operate, inspect and replace the surrounding stack.

The essay does not provide a costed implementation plan or prove that federated open infrastructure will outperform commercial alternatives. It does, however, offer a practical test for sovereignty claims: ask which layers remain controlled elsewhere. “Made here” is doing rather a lot of work if the server rack, chips and critical software all live somewhere else.

What to watch

• Infrastructure ownership:
  whether new national AI programmes control more than model weights.
• Open tooling:
  whether governments fund maintainers of inference, data and agent infrastructure.
• Interoperability:
  whether systems can switch providers without expensive rewrites.
• The procurement test:
  whether institutions choose genuine portability over a locally branded lock-in. The debate now has a useful fault line: should AI sovereignty prioritise domestic control of the whole stack, even at higher cost, or accept commercial dependencies in exchange for speed and capability?

Discussion spark: Should AI sovereignty mean controlling the entire technical stack, even if that costs more and slows deployment, or is dependable access to commercial systems enough?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.