The AI Sovereignty Emergency
For a couple of years we lived through something that turns out to have been an anomaly. A frontier model would launch, we would all get excited, and within days we could actually use it. Everyone in, no queue, no list.
The pattern now is different. A big model lands, the benchmarks are extraordinary, and then comes the second sentence: you can't use it yet. You are not on the list. Fable went through it. GPT-5.6 sat in a slow, account-by-account preview under the banner of security before general release. The gap between "isn't this amazing" and "here's what you're actually allowed to touch" is widening,.
Security is the stated reason, and there's real substance to it. These models are becoming national security assets, and governments are nervous about rival states getting frontier capability. Mustafa Suleyman made the case years ago in The Coming Wave; that unbounded, exponentially improving intelligence needs the same kind of control we put around nuclear proliferation. Take that seriously and gated access starts to look less like gatekeeping and more like grown-ups in the room.
But security isn't the whole story, and this is the part nobody in Britain is talking about.
The problem isn't the frontier. It's the fence.
Think about energy. People understand energy sovereignty because they feel it. At the pump, on the heating bill. We know it matters whether we generate power at home or depend on it arriving from far-flung corners of the world that can turn the tap when it suits them.
There is no equivalent conversation about AI sovereignty. And there should be, because the shape of the risk is identical. We are already dependent on this infrastructure in ways most people don't understand, and access is decided elsewhere.
Run the trend forward. US labs keep their current pace. Europe stays behind. The UK builds effectively nothing of its own. There comes a point where European and British organisations aren't debating regulation and bans. Instead they're asking for access to whatever the frontier is prepared to let them have, a year late and a version behind. When you build your economic model on intelligence you don't control, someone else owns the switch.
This becomes a competitiveness problem for UK and European enterprise on a globalised stage, and it's the kind of thing that becomes obvious only once it's too late to fix.
So what does Britain actually do?
The honest answer starts with a hard admission. Britain cannot lead the frontier. The compute, the infrastructure and the capital required are on a scale we cannot touch. Even inside the EU, the best on offer today is Mistral, and it sits leagues behind. Realistically, playing this game means resigning ourselves to third place and choosing a partner: the US, China, or the EU. None of those is a comfortable choice right now.
But there's a more useful place to look, and it isn't the model at all.
When the cloud arrived, the virtual machine was only the beginning. The value accreted in everything that grew around it. The scalable databases, blob storage, search, event streaming, data lakes. The VM was the forebear of the cloud. In the same way, the LLM is the forebear of what comes next: a whole layer of infrastructure and connectivity built on a foundation of intelligence. Call it Cloud 2.0.
That layer isn't a pile of SaaS apps. It's the services this space still doesn't have. Think data exchange, privacy management, memory, personalisation. All of the things that turn a raw model into something an organisation can actually run on safely. That is real, buildable, and it plays directly to what this country is genuinely good at.
It is not a strategy of owning the brain. It is a strategy of owning what the brain plugs into.
The conversation we're not having
None of this settles the sovereignty question. If the infrastructure you build on can be gated, delayed or switched off, an application-layer strategy carries a dependency risk of its own.
We don't have neat answers. We're not sure anyone does. What we're sure of is that this is a large, largely invisible question sitting just around the corner. It's a debate that is politically untouchable and therefore isn't happening at all.
It should be.