It’s become more of a highlight because of the reliance on big tech across the globe. UK and European companies are ultimately dependent on American technology in a lot of cases. Given some of the geopolitical risks we’ve seen recently, there’s more concern about what could happen if access to certain technology was cut off at any given moment.
If a company is relying on AI technology, and that company is important to the sovereignty of a state, that presents a massive risk. A defence company in the UK is an obvious example.
So the control element starts with infrastructure: the technology supply chain, the estate a business runs on and how it operates day to day. The second element is information. Where is that information being held? How secure is it? Could it get into the wrong hands? In some cases, information being held in allied states may be more acceptable than elsewhere, but it’s still part of the sovereignty question.
The question is also who sovereignty is actually for. Is it government? Is it big tech? Is it citizens? Is it open source communities? Whoever controls the compute and the models probably has more power than the organisation that owns the data. So I think infrastructure matters first, and then it moves into information and data as well.
Why Has Sovereign AI Become a Bigger Conversation?
I think this has probably crept up over time. With the whole move to cloud-based computing, companies have looked at technology hosting mainly through the lens of efficiency, cost and scalability. Those decisions have been led by that, rather than by sovereignty.
With the shift towards big tech, I think we’re now at a point where that balance has maybe shifted a bit too far. You can see certain governments across Europe, France for example, looking at that as a risk, and that has kicked others into the same conversation. That’s an important part of the sovereignty conversation for me.
AI adds another dimension to that dependency. Organisations are now considering foundation models, AI APIs, inference costs, GPU availability, data movement, platforms and the processes built around them.
Many of these capabilities are concentrated among a relatively small number of global technology providers, creating a monopoly of sorts. This isn’t inherently a problem - global technology companies have played an enormous role in accelerating AI development and making advanced capabilities accessible to businesses.
In the UK, DeepMind is a useful example. It was a very successful AI startup, and it was sold to Google because the company wanted to push forward, access more compute power and innovate more quickly. The reasons are understandable, but it also means less control exists in the UK over part of our own technology advancement.
There are positives. DeepMind still has a major presence in King’s Cross, so there has been some effort not to move the whole business out to San Francisco or somewhere else. But it still raises the bigger capability question - are we set up to build these companies again? If the AI and technology companies we build in the UK are acquired by foreign businesses once they become successful, do we end up losing capability as well?
I don’t think every country needs its own ChatGPT, but there’s something in spreading risk and having the ability to make purchasing decisions with sovereignty in mind. At the moment, I’m not sure that’s always a level playing field.
The real risk comes when an organisation becomes so dependent on a particular provider, model or technology stack that changing direction becomes prohibitively difficult.
That is a sovereignty problem.
Does Sovereign AI Mean Building Everything Yourself?
The challenge is that the frontier models are just so far ahead right now. If an organisation decides it’s going to build from the ground up, it’s starting the race five miles behind.
There’s a view that companies building and owning their own AI capability is valuable. In some cases, it is. But if that becomes the exclusive mindset, it could also put organisations behind. For Europe, the risk isn’t necessarily that America has better models. It’s that we’re still too dependent on American distribution.
Most enterprise companies are on some kind of AI transformation journey at the minute. Their workflows, particularly around AI APIs, platforms and processes, are often reliant on that distribution.
Switching now to building your own isn’t a flick of a switch, it requires enormous amounts of capital, compute, specialist expertise and infrastructure. It means changing processes, building pipelines and taking the organisation on a journey. If people are used to using one type of technology, moving them away from it can be disruptive.
What Does AI Dependency Look Like in Practice?
I don’t think the meaning of sovereignty is necessarily about owning every bit of technology or creating your own type of technology. It’s more about agency, having the choice and not having a single point of failure.
How fluid can the business be if it does need to change? If it had to switch technology for whatever reason, what would the business impact be? How interchangeable is the technology at the minute, how much choice does the business have and does that choice actually exist?
The market connects to so much. You have enterprise procurement, the cloud, AI models, chips and geopolitics, so there’s a whole supply chain around it all. It can become a big can of worms quite quickly.
But for me, it’s about having the agency rather than trying to build everything to be contained. If everything is contained, there may be less space to share learnings and innovate.
This is where sovereignty becomes less about ownership and more about optionality.
A business doesn't necessarily need ten AI models.
But it should understand what would happen if its primary model, cloud provider or AI platform became unavailable, commercially unattractive or strategically unsuitable
Why Do Some Businesses Get More Value From AI Than Others?
It’s very much a governance question. As with any new technology, people and process need to be considered, and governance sits around that.
Having the right structures and governance frameworks around AI usage can set businesses up for longer-term success. There’ll still be a lot of trial and error around what the optimal governance framework looks like for an enterprise business, and it’ll be different each time. But it’s a really important consideration when businesses are on the AI journey.
It also comes back to trust. How is AI being used across the organisation, and how are employees empowered to use it? There’s a fine line between not having so much rule and regulation that people can’t innovate, while still having the structures and frameworks to manage risk, trust, control and spend.
Where Do AI Programmes Start to Fall Apart?
I think things typically miss the mark when there’s a lack of planning and execution, and when the work isn’t linked back to business value.
A lot of organisations struggle with legacy technical debt. They also struggle to understand their current systems before they over-invest in AI. We’re all in an AI race now, and every company appreciates that mistakes will be made and that there’ll be a trial-and-error process to go through.
But it’s about investing in the right way at the right time, while still being conscious of the fundamentals and the foundations. Addressing both of those things in tandem is where we’re seeing success.
The final thing is the journey companies need to go on around behaviours and ways of working. That links back into governance as well: how and when the technology should be used effectively. It’s less about buying the latest frontier model and trying to embed it. It’s more about looking at people, process and technology more broadly.
What Is the Price of Sovereign AI?
The other point is the price of AI sovereignty. There’s a cost. Enterprises are discovering that AI sovereignty has a price tag.
Companies have spent years optimising for the cloud, for efficiency in the cloud and for efficiency around price. They consolidated onto hyperscalers because it was more economically rational. That was the strategy.
Now we have AI in the equation, which brings forward inference cost, GPU availability, model APIs, data platforms and data movement. Those become strategic dependencies as well. It’s a little bit of an uncomfortable position to be in because the current architecture of an organisation doesn’t necessarily set it up for sovereignty, but sovereignty is becoming a more prevalent topic of conversation.
I also think less is more. Companies don’t necessarily need to have 50 different AI models or 50 different things, but they do need the agency to rely on more than just one or two that come out of the US.
There are positive signals across Europe, in terms of investment in startups and some smart people in the AI scene. But the real problem with AI is that there aren’t many people who know how to build true AI models. That will change over the next decade, but right now it’s quite a small community that knows how to build the amazing things we’re seeing from the likes of OpenAI and Anthropic.
What Does Exit by Design Mean for AI?
Exit by design looks beyond execution. Is that organisation going to be set up for long-term success from an IP perspective? Has it gained knowledge through the transformation, within the people in that organisation, to see through long-lasting change?
Is the organisation in a stronger place, having gone through the learnings and challenges that will present themselves through any transformation? Exit by design is about leaving a customer in a position where they can almost stand on their own two feet more independently.
That means not being over-reliant on any particular technology or vendor. It means having the right operating models, structures, talent capabilities and technology, so the organisation isn’t completely dependent, and its ecosystem gives it a bit more agency.
Private companies are building structures now that look a little bit more like national infrastructure. If there are only a handful of companies out there that control the frontier models, the question is how organisations use that effectively while still maintaining control.
Planning a Sovereign AI Programme?
For organisations working through these questions now, the first step isn’t to decide whether everything needs to be built internally. It’s to understand where control matters, where dependency already exists and where the business would struggle to change direction if it needed to.
At DataReign, we believe AI transformation should leave organisations stronger, not more dependent.
We help organisations modernise their data foundations, build the governance and capability needed to use AI effectively, and create technology environments that their own teams can understand, operate and develop over time.
That means moving quickly around real business problems while keeping ownership, knowledge transfer and exit by design at the centre of the programme.
Sovereign AI isn't about controlling everything. It's about retaining the agency to choose what comes next.
