How do you run AI on real data, in real operations, without handing control to someone else? The short answer is: own your data, keep your architecture and models portable, and never depend on something you can’t replace. It sounds straightforward until you map out everything your AI actually depends on.
At Nion, that question now comes up in almost every conversation about AI. And it usually starts with dependency. What happens if a supplier fails, changes direction or gets acquired? What happens if the model behind your agents is no longer the right one?
That is what sovereignty is really about. Not a single compliance box, but the freedom to keep running, adapting and growing on your own terms.
Why AI sovereignty has climbed the agenda
We’re not the only ones seeing this. In a recent interview with CIO Analytics, Per Jacobson, Partner Manager at our partner IBM, shared findings from a global study of more than 2,000 CEOs across 30 geographies and 23 industries. The shift is striking. AI is no longer seen mainly as a way to cut costs. Instead, leaders now expect it to drive growth, with some seeing it contribute up to half of top line revenue growth.
However, bigger ambitions mean bigger exposure. According to Jacobson, sovereignty has risen sharply up the agenda since late spring, exactly as companies move from anonymised test data to live operations. In other words, this is no longer an IT decision; it belongs with the executive team.
Four dimensions, not one
The most common mistake we see is treating sovereignty as a data question only. IBM’s framework splits it into four, and it matches what we see in delivery.
Operational: Can you run the business without relying on one critical supplier, and take operations back if needed?
Data: Do you control where your data lives, who can access it and how it is used?
Technical: Is your architecture built so that leaving a vendor is a project, not a crisis?
AI: Do you know which models your agents run on, and can you switch when you need to?
Together, they come down to two things: flexibility and control.
Our checklist before AI goes live
Building on IBM’s recommendations for CIOs, these are the questions we ask with every client before an AI solution moves into production:
- Which processes must keep running if a key supplier or the connection goes down?
- Which regulations actually apply to this use case, whether GDPR, NIST or ISO 27001, and how will we keep up as they change?
- What is the fallback if a critical partner can no longer deliver, and do we need to be able to step in ourselves?
- Could we switch models or vendors without rebuilding the solution?
If any of these are hard to answer, that is where to start. It’s also why security and control are built into how we work, and why Nion is ISO 27001 certified.
What this means for you
AI is moving into the core of how businesses grow, and that raises the stakes. Owning your data, keeping your architecture and models portable and having a plan B for every critical partner isn’t extra work on top of AI. It’s what makes AI worth scaling.
Talk to us about where your organisation stands today.
Insights from IBM are based on Per Jacobson’s interview with CIO Analytics.