Can Britain Build Its Own AI Future? DeepMind, Mistral and Sovereignty

A 25-year-old founder just raised $312 million to build a British AI chip that deliberately avoids the one component — high-bandwidth memory — that constrains every rival, including Nvidia. Meanwhile the EU opens a 10-billion-euro sovereign compute programme built openly on American processors, Washington's export controls turn out to have a hole big enough for Chinese tech giants to rent Nvidia's banned chips through data centres in Thailand, and China's own chipmakers close in on 90% of its domestic market. Last episode ended on an asterisk — sovereign, except for the silicon. This episode asks whether that asterisk can ever actually be removed, or whether every strategy on the table is really just a different way of living with it.

A thirty-person London startup has signed an NHS-adjacent agreement for Lumen Sovereign, Britain’s proposed sovereign frontier AI model. But the memorandum names no customer, creates no deployment obligation, and carries no financial value. The episode asks whether sovereignty means where servers sit or who can operate, inspect, and alter the system. From Cosine and Google DeepMind to France’s Mistral and China’s constrained chip supply, every claimed sovereign system still depends on infrastructure controlled somewhere else. So who is actually in control?

Demis Hassabis steps back from day-to-day leadership at Google DeepMind as operational command moves to California. At the same time, Britain’s Sovereign AI Fund reaches eleven backed companies, unveils a £1.1 billion hardware and compute plan, and backs Lumen Sovereign, an attempt to build a fully sovereign British frontier model. Can Britain turn invention into lasting power, or will its most important AI decisions continue to be made elsewhere? France’s Mistral offers a revealing alternative.