We are witnessing something genuinely novel in the history of civilisation: the systematic dumbing down of entire nations by choice. The race to build ‘sovereign AI’ — national large language models trained exclusively on local data — is not a technological arms race. It is a race to the bottom of intellectual provincialism. Every country, from France to India to Saudi Arabia, is now pouring billions into constructing models that will produce answers that flatter local pieties, reinforce national prejudices, and never, ever suggest that perhaps the French baguette is not the pinnacle of bread-making or that Indian bureaucracy might have room for improvement. This is not artificial intelligence. This is artificial stupidity, custom-made for a world that has decided that truth is less important than identity.

Let me be blunt: the idea that a country as small as Switzerland, with its eight million people and three official languages, can produce an LLM that rivals GPT-5 is laughable. But that is exactly what they are attempting. The Swiss government has pledged 100 million francs to build an ‘open, trustworthy’ model that reflects Swiss values. What does that even mean? A model that refuses to comment on banking secrecy? One that tactfully ignores the fact that the Swiss army hasn’t fired a shot in anger since 1847? The entire project is a monument to political vanity. It will produce a model that is polite, neutral, and utterly useless for any serious intellectual work. It will be the IKEA of AI: functional, bland, and completely devoid of character.

The tragedy is that this is happening everywhere. The European Union, terrified of American and Chinese dominance, has committed billions to building models that are ‘European’. This is a category error. There is no such thing as European mathematics or European logic. 2+2=4 in Brussels just as it does in Beijing. A model that deliberately excludes non-European data is not a model. It is a dogmatic catechism. It will serve up answers that confirm the EU’s bureaucratic obsession with regulation, data protection, and the innate superiority of the continental philosophy. And it will fail. Not because the engineers are incompetent, but because the premise is flawed. Intelligence does not respect borders. It thrives on heterogeneity, on the clash of ideas, on the uncomfortable realisation that your founding assumptions might be wrong.

Consider the historical parallel. In the late Roman Empire, each province began minting its own debased coinage. They did this to preserve local prestige and economic autonomy. The result was not prosperity. It was hyperinflation, trade collapse, and the eventual inability to pay the legions who were supposed to protect them from the barbarians. We are doing the same with AI. Instead of using a universal currency of knowledge, we are stamping our own tribal insignias on increasingly worthless data. The Barbarians are not at the gate. They are inside the server room, building models that will tell us exactly what we want to hear.

This is also a colossal failure of strategic thinking. The United States and China are not building national models. They are building global models trained on the sum of human knowledge. Their LLMs can discuss the nuances of Bengali poetry, the intricacies of Andean irrigation systems, and the subtleties of Edwardian etiquette. A sovereign Indian model, trained predominantly on Hindi and English sources, might be excellent at explaining the caste system but hopeless at understanding the context of the French Revolution. Why? Because in the quest for cultural authenticity, they have amputated a limb of global understanding.

I can already hear the objections: ‘But Arthur, what about data sovereignty, privacy, and cultural sensitivity?’ To which I reply: these are the concerns of bureaucrats, not thinkers. Of course, there are legitimate worries about foreign models being used to spread disinformation. But the solution is not to build a walled garden of mediocre AI. It is to demand transparency, open standards, and rigorous peer review. The answer to bad speech is more speech, not a nationalised model that quietly avoids controversial topics.

The truth is that the push for sovereign AI is a symptom of a deeper intellectual decadence. It is the digital equivalent of burning books in the name of cultural purity. If every country builds its own model, the global conversation will fragment. We will have Turkish AI insisting that the Armenian genocide never happened, Japanese AI politely ignoring the Rape of Nanking, and American AI carefully sidestepping the legacy of slavery. The result will be a world where AI does not enlighten: it stifles. It will be the ultimate victory of identity over truth.

We have been here before. In the 19th century, nationalist historians rewrote the past to serve their present. In the 20th, totalitarian states used propaganda machines to control information. Now, in the 21st, we are voluntarily building our own propaganda machines, wrapped in the rhetoric of sovereignty and privacy. It is a tragedy of intellectual cowardice. AI was supposed to be the dawn of a new Enlightenment. Instead, we are getting a thousand tribal campfires, each one casting just enough light to see our own shadows.

If we continue down this path, the only truly global intelligence will be the one that no one trusts. And the barbarians will not need to sack Rome. We will have already burned the library ourselves.