Ontology and sovereignty: who defines what your data means

Ever since the word became fashionable, everyone talks about ontologies. It gets thrown into any AI meeting without almost anyone stopping to explain what it means. And yet, underneath that awkward word sits one of the most important decisions your company will make about artificial intelligence: who defines what your data means.

An ontology, in plain terms, is the map of what things are in your business and how they relate. What a customer is, what an order is, what an incident is, and how they connect to each other. It is not the database: it is the layer that gives it meaning, the one that lets a system (or an AI agent) understand that 'customer' in your CRM and 'account' in your ERP are the same thing. Without that map, AI shuffles isolated records without knowing what they mean.

The three layers you don't want to hand over

Over the past two weeks, two of the most influential executives in tech have each pointed, in their own way, at the same frontier. Alex Karp, of Palantir, said you have to own your weights and your compute: we covered it in controlling your weights is controlling your fate. Satya Nadella, of Microsoft, added that you also have to own what AI learns by using you, your learning loop, in you pay for AI twice. The ontology is the third layer of that same stack, and perhaps the deepest one: meaning.

Why meaning is the hardest to get back

You can switch compute providers in an afternoon. Switching models, with some work, too. But if your business ontology, the map of what your data means, lives inside a third party's platform, getting it back is another matter. That map is built over years, with every decision about what you call things and how you relate them. It is, literally, the knowledge of how your company works turned into structure.

And here is the risk almost nobody looks in the eye: if that layer is managed by your vendor, over time the vendor understands your operation better than you do. You rent the intelligence; they learn your business. It is the same asymmetry Nadella described, one floor down, in the layer where meaning lives.

The European dimension

For a Spanish or European company there is one more layer, and it is jurisdictional. Under GDPR and NIS2, protecting the data is not enough: where it is processed, who can access it and under which law all matter. If your meaning layer, your ontologies and your embeddings live on the infrastructure of a provider subject to the US CLOUD Act, you have solved the technical problem and created a regulatory one. AI sovereignty is not a slogan: it is the concrete answer to where your data's meaning lives and which law can reach it.

• The vocabulary: how your company names things, the basis of the ontology.
• The relationships: how customers, orders and incidents connect, the part an agent needs in order to act and not just look things up.
• The embeddings and indexes: the representation of your data that makes semantic search work.
• The rules: which actions an agent can run on those objects, the 'kinetic' half of the ontology Palantir talks about.

All of that should live inside your boundary, not anyone else's. We operate dedicated GPU infrastructure in a datacenter in Madrid, ISO 27001 and ENS certified, precisely so that layer (the meaning, the learning and the weights) never leaves your jurisdiction. You can use the best models without handing over the map of what your business means. We develop it in private inference.