The sovereignty conversation all too often becomes a debate about geopolitics. This means that the nuances, as well as some of the dangers, of entirely ignoring sovereignty get buried, with the UK’s technology secretary warning that if the UK depends on US technology companies, it could threaten national security.
Questions about AI sovereignty often relate to location, considering where a datacentre is, where the provider is headquartered and if the infrastructure sits or can sit in the UK.
These are reasonable questions, but they don’t tell us everything we need to know. We also need to ask who can access the data, who sets the rules and how much influence we can have over the systems we’re relying on.
For the UK, this should be the starting point for AI sovereignty. We need to establish what we want to protect and which values should guide our decisions. From there, we can identify partners who share those priorities and build infrastructure together, rather than letting this get tied up in broader politics.
The cost of narrowing our options
If the UK lets geopolitics hold us back from forming alliances, we will struggle to keep pace. The risk becomes even greater if we decide that sovereignty requires us to develop AI infrastructure independently rather than look at who we could work alongside.
This challenge extends well beyond building datacentres. Computing capacity, specialist skills and the resources needed to keep improving systems must all come together to create a truly ‘sovereign’ ecosystem. Meanwhile, the technology we’re trying to catch up with continues to develop even further every day.
It’s difficult to see how the UK could close that gap alone across every part of the infrastructure. Public funding has to serve competing demands, from defence to healthcare and social care. Expecting it to support a wholly independent AI ecosystem as well would be considerable for the UK government.
Private investment could help, but attracting it requires a convincing proposition. Why should a business commit substantial resources to the UK if it can access more established capabilities and larger markets elsewhere? Being a valuable partner in a wider AI ecosystem would give us a stronger position here.
This also matters in the boardroom. If sovereignty is interpreted too narrowly, businesses can find themselves choosing from a restricted set of suppliers before properly assessing their options. Decisions then become an exercise in meeting a location requirement, with less consideration given to resilience, security or access to capacity.
Cloud and IT leaders need the freedom to explore different architectures and partnerships. The UK should give them a clear set of objectives against which to judge those choices.
Finding partners who share priorities
Choosing trusted partners begins with a question about values. The UK needs to ask a series of questions: How do they approach privacy? Who is accountable when systems fail? Should access to AI infrastructure extend to researchers, entrepreneurs and smaller businesses?
We also need to understand the assumptions behind the technology. If a model has been developed elsewhere, whose priorities shaped its training and safeguards? Do those decisions reflect how we want to use AI? From these responses, we can start to consider which countries are best to partnership with.
Shared interests doesn’t necessarily mean that we’ll provide shared answers. Countries may agree on a commercial opportunity or cooperate against a common threat while taking different positions on privacy and social responsibility. Those differences deserve attention before we establish long-term dependencies.
This is extremely important, because when we discuss how new frontier models are trained, values impact how agents act and make choices, as well as shaping the outputs that will come from generative models.
The security risks
There are concerns about the potential security risk if we build these relationships. However, the key factor here is trust. AI governance shouldn’t just be a compliance exercise, it should be a framework to make better decisions, and this can happen if we work alongside trusted partners. The main question that you need to be able to answer is if your values align, if they do, security issues shouldn’t be a concern.
A question to ask is why can’t a shared infrastructure be considered sovereign? The issue is centred around where you draw the boundary, but this boundary doesn’t have to be so singular. We have seen several instances this year alone of huge data centre damage within countries and geographic regions. Thinking wider is more important than ever. Sovereign AI can be a community resource that we all safeguard, we should trust those in our community to play their part, just as they trust us.
If we refuse alliances, we’re assuming that the UK has the best security and privacy for sovereign AI models on our own. This would be a dangerous assumption, as it takes a team and community to bring the best to the table, and sharing makes us stronger than a one-nation approach to sovereignty.
If we choose our partners carefully, aligning our goals and values, the chance of success is much higher. We’re already sharing aspects of our security with our neighbours as well as global partners, so why should AI sovereignty be any different?
What would the first steps look like for the UK?
First of all, we need to know our own values. The government can then implement a framework to outline these, so we, as a country, can take these values and consider who our allies might be, emphasising values over geography.
The UK can be the first to invest and break ground on a project and invite our allies to build and contribute immediately. Otherwise, if we wait for others to take the reins, we will only ever be a follower. We’re in a strong position, and we can show the way by inviting others along on our journey.