Generation 1 of the UK's sovereign compute strategy involved buying US chips, running US frontier models on them, and deploying them via US cloud infrastructure, with companies funded by the US developing products and services on them. I think it's dangerous to assume that we will retain priority access.
The market for inference compute and intelligence at the edge will continue to grow exponentially, driven by agentic AI (AIs talking to other AIs), physical AI (AIs deployed in robots), and edge AI (AIs running outside of existing broadband and energy infrastructure).
The level of diversification and expansion in what machine intelligence can do for us is subject largely to the imagination of entrepreneurs and small business owners and their access to compute.
Access to compute, simplistically, comes down to 2 things: datacenter racks and the energy to power them.
We are buying our compute from a country who doesn't believe that it has an energy crisis. The pressure for chip manufacturers to deliver energy efficiency runs a very distant second to delivering performance and growth.
Generation 1 of the UK's sovereign compute strategy sees us adding energy infrastructure as if we were the US. This is the right call for now but it is not a long term solution for a small island.
An outline of a solution, therefore, can be seen. Build for the next gigantic market (inference compute) where established chip companies are not the de facto alternative, aggressively fund and support British start-ups, and develop within the constraints of British infrastructure.
The Government has already made an impressive set of moves to address this: ARIA's Scaling Inference Lab, AI Growth Zones, the AI Research Resource, the AI Growth Lab, the Sovereign AI Venture Fund: adding up to billions of investment
There are well-funded companies at almost every layer of the stack: chips (Fractile, Olix, Vaire) / datacenter infrastructure (Era4) / frontier models (Cursive, Ineffable) and of course a host of successful AI startups and product companies.
Thanks to the work of pioneers like Matt Clifford and Alex Depledge, builder mindset and support for entrepreneurs is baked into the Government's culture and vision.
For those of us who have been around a while and bounced off previous Government initiatives, it's easy to be cynical. I am not cynical about this opportunity. I think it is different this time, because the delta between outcomes for the UK's economy, defence capabilities and the quality of public services is so high.
Through door A - massive growth in the economy through UK tech companies capturing market share of inference compute, radically improved and more capital efficient public services, more tax to pay for those services, delivered within the UK's energy and infrastructure footprint.
Through door B - yet more massive growth in US tech companies, UK companies crippled by limited access to compute, energy-hungry datacenters creating political and community anger.
A very reasonable response to this analysis is to point at NVIDIA's market cap, their acquisition of Groq, and sneer at the UK's measly billions going into this initiative. What that misses is that there is a nuanced and sensible viewpoint, articulated by SovAI fund chair James Wise:
What AI sovereignty means continues to be the focus of global debate. Some argue that Sovereign AI means countries should work together to “pause” AI development. Others argue each country must own each part of the AI infrastructure stack themselves, from electron to token.
In my view, these options are technologically untenable, and economic and cultural dead-ends. Instead, the UK must continue to adopt and contribute to the best infrastructure and tools available to us globally, supporting and shaping the development of AI tools for everyone.
Sustained, ambitious market-making investment creates the conditions for innovation and industrialisation, to some degree regardless of outcome. And in a world of ever greater uncertainty, having a viable sovereign inference stack could become invaluable.