In a LinkedIn post I asked a question: if access to the most advanced AI models were restricted by the countries or companies controlling them, what alternatives would we have in Europe?
It is a scenario to consider when designing a dependency, not a prediction that every service will shut down. A model integrated into daily work stops being just a website to consult: it becomes part of the business infrastructure.
Access does not mean autonomy
Being able to use an API today does not mean controlling its terms tomorrow. Pricing, availability, and capabilities can affect the products built on it. The issue is the ability to keep working when a dependency changes.
Europe is not starting from zero: Mistral’s catalog includes open-weight models alongside commercial offerings. This is a concrete example, not proof that every need is already covered. Autonomy also means having the expertise, infrastructure, and people to operate alternatives.
Mistral’s official model catalog
What this means for software builders
For an individual project, I would start with practical questions: which functions depend on the model? What happens if it does not respond? Can we export the data and test an alternative without rebuilding the whole product?
Not every SME needs to train a model or buy dedicated hardware. But it can understand its dependencies, prepare an alternative way of working, and decide where to retain room to manoeuvre.
My concern is that convenient access can make us overlook the work needed to preserve choice. Building our own capabilities takes time, but it is what makes an alternative genuinely usable.
Adapted from one of my LinkedIn posts and added to the blog on 10 September 2026. The date shown refers to publication on this website.