The word sovereign is everywhere.
Every week there’s another product, another company, another announcement promising “sovereign AI.” In many cases, it simply means an open model running in a particular country or inside a private cloud.
Those are meaningful advances. They solve real problems.
But they’re only part of the story.
We use the word sovereign at Backboard because we believe it describes an important shift in enterprise computing. At the same time, I worry that the term is beginning to lose its meaning through overuse. If sovereign AI comes to mean “my preferred deployment option,” then we’ve reduced an important idea to a marketing slogan.
I think sovereignty is something much bigger.
To me, sovereignty is the ability for an organization to make every meaningful decision about how AI operates.
It is the freedom to choose your own models. To improve those models. To decide where they run. To determine where your data lives. To control how your systems are governed. To move between cloud, private cloud, on premises, and on device without rebuilding your architecture or asking another company for permission.
Infrastructure Alone Doesn't Make You Sovereign
Sovereignty is ownership.
For too long, the conversation has focused almost entirely on infrastructure. Infrastructure matters. It always has. But infrastructure alone does not make an organization sovereign.
If your model can never be adapted to your business, are you sovereign?
If your reasoning is locked behind a third party API, are you sovereign?
If your deployment options are dictated by a vendor roadmap, are you sovereign?
If changing providers requires rebuilding your entire AI stack, are you sovereign?
I would argue the answer is no.
The Model Is Becoming Enterprise IP
The model itself is becoming part of an organization’s intellectual property.
Until recently, most enterprises accepted that foundation models were fixed. Your knowledge lived outside the model in prompts, retrieval systems, and vector databases. That made sense when post-training was prohibitively expensive.
That assumption is changing.
As post-training becomes faster and more affordable, organizations will increasingly teach models how they think. Their policies, terminology, workflows, institutional knowledge, and domain expertise will become part of the model itself. The model will no longer be just something an organization consumes. It will become something it owns and continually improves.
That changes the definition of sovereignty.
Sovereignty Means Controlling Every Layer
A sovereign organization should be free to choose between open and proprietary models. It should be free to post-train those models with its own knowledge. It should be free to quantize and optimize them for its own hardware. It should be free to deploy them in the cloud, inside a private cloud, on premises, or directly onto employee devices. And it should be free to change any of those decisions as technology evolves.
No single layer defines sovereignty.
Not the model.
Not the infrastructure.
Not the deployment.
Not the data.
Real sovereignty exists only when an organization controls all of them.
The Future of Enterprise AI Is Choice
This is the future we believe enterprises are moving toward.
Not because regulation demands it.
Not because security teams insist on it.
But because organizations increasingly recognize that AI is becoming core infrastructure. The organizations that thrive will be those that own the capabilities that differentiate them, while retaining the freedom to evolve every layer of their stack as the technology changes.
The future of enterprise AI is not just open.
It is not just local.
It is not just private.
It is sovereign in the truest sense of the word: the freedom to decide.
Rob Imbeault
SHARE