Sovereign AI Is a Distraction, Not a Strategy
Cory Doctorow, who gave us the word 'enshittification' and has been gleefully poking holes in tech industry narratives ever since, has waded into the sovereign AI debate. His verdict: the whole concept is nonsense.
The prompt came from Australian economist John Quiggin, who wrote about Trump's systematic dismantling of American soft power across diplomacy, finance, military alliances, and technology. Quiggin's concern was that other countries have grown dangerously dependent on US-controlled AI infrastructure, and that relying on increasingly erratic American institutions for critical technology is a bad bet.
Quiggin himself is at least clear-eyed about the economics. He argued that second-movers like DeepSeek and France's Mistral, running roughly a year behind the frontier labs, can replicate most of what those labs produce at a fraction of the cost and without needing a small power grid to do it. The tortoise beats the hare, eventually.
That's a reasonable position, but Doctorow sidesteps it entirely. His argument is more fundamental: why does any of this matter if AI isn't actually critical infrastructure in the first place?
He points out that if Trump switched off access to every AI chatbot tomorrow, most businesses outside the US would barely notice. But if Microsoft killed Office 365 or John Deere remotely bricked its tractors, the economic damage would be immediate and very real. Those are the actual dependencies worth worrying about.
Doctorow's own framing is characteristically blunt. AI, in his view, is 'the money-losingest technology in human history,' one that has consistently underdelivered against what its loudest supporters promised. Fixating on AI sovereignty while ignoring concrete vulnerabilities in everyday software, telecoms, and e-commerce infrastructure is, he argues, just misguided prioritisation.
He's got a point, even if you think he's underselling where AI is heading. The more pressing problem is that US tech companies have made it abundantly clear they will not push back against Washington when it counts. The moment political winds shift, so does their compliance. Europe and other regions betting on American goodwill to protect their access to critical software platforms are taking a significant risk that has nothing to do with large language models.
Building out domestic alternatives for the boring, load-bearing software that actually runs economies is unglamorous work. It also happens to be more urgently necessary than debating who controls the next chatbot.