In short

AI policy researcher Hamish Low argues that computing power alone can't deliver AI sovereignty. Building chips and data centres leaves nations as customers dependent on US-controlled supply chains (illustrated by Graphcore's collapse). Instead, countries like the UK should pursue strategic interdependence, making the US dependent on them by dominating 'AI middleware' (sector fine-tuning, RL environments, assurance), backed by aggressive industrial policy, talent visas and a 2–5 year leverage horizon.

Every government now wants “AI sovereignty,” and most of them think it means buying chips and building data centres. This episode is a sustained argument that they’re wrong.

Hamish Low’s case is that hardware is the one layer where smaller nations can never win independence, you can spend a fortune on compute and still be a customer of TSMC, ASML, and US export-control policy. Real leverage comes from making yourself indispensable instead: dominating the “AI middleware” layer of sector-specific fine-tuning, evaluation, and assurance, where a country like the UK actually has an edge and where things move at software speed.

It’s a refreshingly concrete conversation about industrial strategy, including a clear-eyed look at why national-champion bets like Mistral and Cohere mostly disappointed, and what a Chinese-style multi-entrant approach would look like instead.

Key takeaways

  • 'AI sovereignty' has become a vague catchall; the real question is how a nation fits into a US/China-dominated landscape.
  • Competing on hardware is false sovereignty: builders remain customers dependent on TSMC, ASML and US export controls.
  • The UK should pursue interdependence, becoming indispensable to the US, rather than unattainable independence.
  • 'AI middleware' (sector fine-tuning, RL environments, assurance) matches the UK's talent and moves at software speed.
  • National-champion bets like Mistral and Cohere mostly failed; Chinese-style multi-entrant sector policy is a better model.

Listen to the full episode

This is a summary. Listen to the full conversation on your platform of choice, or read the companion essay on Substack.

Read on Substack → Watch on YouTube Listen on Spotify

Frequently asked questions

Can a country achieve AI sovereignty by building its own compute?
No. Building chips and data centres still leaves nations dependent on US-controlled supply chains (TSMC, ASML, export controls), so they remain customers without real leverage.
What is AI middleware and why does it matter for sovereignty?
It's the software and service layer between foundation models and enterprises, sector fine-tuning, RL environments, assurance, where a country's talent can create strategic value foreign giants can't replicate per-sector.
What should the UK do instead of building AI chips?
Pursue strategic interdependence by dominating middleware and assurance, deploy aggressive Chinese-style industrial policy and talent visas, and focus on leverage within a 2–5 year horizon.

People & ideas in this piece

Hamish LowJeffrey DingDavid EdgertonASMLTSMCNvidiaGraphcoreMistralCohereAI middlewareInterdependence vs independence

Topics: AI & Geopolitics , The Political Economy of AI