Sovereign AI is 'nonsense,' says Cory Doctorow: build sovereign apps instead

Cory Doctorow, the writer who coined ‘enshittification,’ has a blunt message for nations scrambling to build their own AI models in the name of digital sovereignty: you’re solving the wrong problem.

The real vulnerability isn’t that countries rely on American or Chinese large language models. It’s that they lack control over the applications and infrastructure those models plug into. ‘Sovereign AI is nonsense,’ Doctorow said, arguing that what governments actually need are sovereign apps and sovereign datacenters — software and hardware stacks they can audit, modify, and ultimately govern themselves. Without that, a domestically trained model is just another dependency.

The argument lands at a particularly chaotic moment. As the Trump administration dismantles elements of the US-led world order — pulling back from alliances, threatening tariffs, and unsettling decades of trade policy — the assumption that American technology will remain reliably accessible is suddenly in question. Doctorow’s point is that even a fully homegrown AI model won’t protect a country if it’s running on someone else’s cloud, inside someone else’s application layer, subject to someone else’s terms of service.

It’s a reframing that cuts against the grain of current AI industrial policy. Nations from India to France are pouring billions into domestic model development, treating it as a strategic imperative. Doctorow is telling them they’re building the roof before the foundation. Control the runtime, he argues, and the model matters less. It’s the kind of structural critique that sounds obvious once you hear it — and deeply inconvenient if you’ve already written the check.