One Ribbon Is Easy to Cut
Imagine the opening of a new national AI data centre. Government officials cut a red ribbon. Cameras flash. The speeches promise technological sovereignty.
Across town, the owner of a small engineering firm tries to cut something less ceremonial: a contract with its AI provider.
But years of drawings, decisions, maintenance history and corrections are tied to that provider's systems. The firm can stop paying. It cannot easily take its working context with it.
One ribbon was easy to cut. The other isn't.
I understand why countries want their own compute. Security, science, industry and resilience require serious infrastructure. I am not arguing against data centres.
But independence at the national level is not automatically independence for the people using them.
And there is a financial complication. Giant facilities need utilisation and long-term revenue. As smaller models and local machines become more capable, what benefits users may conflict with the economics of existing infrastructure. That is an incentive problem, not proof of a conspiracy.
I use local AI and powerful cloud models. I need both. What I reject is the idea that access to the second should make my own history dependent on its owner's permission.
If public money helps build AI infrastructure, the public deserves more than a new paywall. Otherwise taxpayers may help finance the gates and then pay again to pass through them.
Ask what happens when a school, a clinic or a workshop wants to change providers. Can it export usable history? Switch models? Keep essential work running with a local fallback?
In my World Intelligence research, the distinction is deliberate: models are replaceable sources of capability. The long-term history, permissions and obligations of a person or accountable institution are a different layer.
The test of sovereign AI is not whether a government can cut a ribbon in front of its own data centre.
It is whether its citizens can cut a contract without cutting away years of their own work.