Cory Doctorow, the writer who coined the term ‘enshittification,’ has a blunt message for the wave of nations scrambling to build their own AI infrastructure: you’re solving the wrong problem.
Speaking on the shifting geopolitics of technology, Doctorow argued that ‘sovereign AI’ — the push by countries to develop domestic AI capabilities independent of American or Chinese tech giants — is ‘nonsense.’ His point isn’t that digital sovereignty is a bad goal. It’s that the current approach, fixated on hardware and model training, misses the mark entirely. With the Trump administration actively dismantling the US-led world order that once underpinned global tech supply chains, the argument has new urgency.
What nations actually need, Doctorow contends, isn’t their own version of ChatGPT. It’s ‘sovereign apps and datacenters.’ The distinction is critical. Pouring billions into GPU clusters to train frontier models from scratch is a fool’s errand for most countries — they’ll always lag behind the hyperscalers with their near-infinite capital and talent pipelines. But controlling the application layer and the physical infrastructure where data lives? That’s where real leverage exists. It’s a classic Doctorow argument: power doesn’t come from building the shiniest object, but from owning the pipes and the interfaces that people actually use.
This lands at a moment when the old assumptions about global tech cooperation are crumbling. The post-war alliance structure that allowed a relatively open flow of chips, software, and expertise is being actively unwound, replaced by a transactional, adversarial framework. In that vacuum, the instinct to build a fully domestic AI stack is understandable. But Doctorow is essentially telling policymakers they’re falling for the hype. You don’t need to beat OpenAI at training runs. You need to ensure your citizens aren’t locked into platforms controlled by a foreign power that can enshittify the experience on a whim, or cut off access entirely.
