The global AI boom is built on staggering inefficiency: repurposed videogame chips, vast energy waste, and escalating costs. While the US and China race blindly ahead, philosopher of mind and the Founder and CEO at Euclyd, a Dutch company developing chips and systems for core AI datacentre compute, Bernardo Kastrup, argues that Europe has an overlooked opportunity to redesign AI hardware itself, and in doing so reclaim technological sovereignty.

As a philosopher of mind and computer engineer, AI is the one topic that connects both of my professional aspirations. And since this powerful new technology is bound to shape the future of civilization, either supporting or undermining our ways of life, the survival of European values so consistent with my own idealist views—such as personal liberty, liberal democracy, human rights, equality of opportunity, consumer protections, distribution of power, etc.—will largely depend on how we manage the ongoing technological transition. To ensure the survival of its way of life, Europe must thus have the means to control the deployment of AI in its territory, so it happens on our terms.

In this context, it is with sadness that I observe how accepted it has become, among European high-tech industry leaders, that the continent’s AI chips will continue to come from Asian fabrication plants for the foreseeable future. After all, Asia is now so many generations ahead in the development of chip fabrication technology that to think of more than just obsolete chips being made in Europe is to indulge in fantasy; or so the story goes. Many believe that our only hope to secure a palatable AI future is to develop our own homegrown AI software models to ensure their alignment with European values and regulations—that is, to ensure that AIs applied to government systems, healthcare, industry, the military, etc., reflect what Europeans consider fair and appropriate.

___

To ensure the survival of its way of life, Europe must thus have the means to control the deployment of AI in its territory, so it happens on our terms.

___

Here, I shall argue that such thinking is wrong on both counts, and that true European AI sovereignty can only be achieved through homegrown hardware—that’s the bad news. The good news is that, if our problems and strengths are correctly understood, European AI sovereignty can be achieved not in twenty or thirty years, but before this decade is out.

AI alignment is not about models

Industry leaders understand that, for Europe to catch up with the leading chip fabrication facilities, or “fabs,” in Asia, an investment of decades and tens of billions of Euros would be required. Therefore, it is understandable that many should seek refuge in software: no manufacturing is needed, and development cycles are measured in months, not decades. This way, Europe could ignore its manufacturing lag and still lead in at least one salient segment of the AI ecosystem. Sovereign models—many believe—are crucial to ensuring that our AI future is aligned with our laws and way of life.

This is baseless. By and large, the value-alignment of an AI model is not an intrinsic property of the model itself—that is, of the software algorithm—but instead an emerging outcome of its training and parameter fine-tuning. This is analogous to the human mind: our values are largely defined by what we learn through education and feedback from our social milieu, not brain anatomy. Trying to address the need for alignment with homegrown software models is like prescribing brain surgery to address improper education.