Infrastructure
Compute, power and planning: the physical limits of British AI
Each generation of frontier model has required more computing power. That turns AI strategy into energy strategy—an area where Britain faces high prices and long connection queues.
Why compute has become strategic
Training and running the most capable models requires large clusters of specialised chips. Access to that compute shapes who can do frontier research, who can evaluate models independently, and who can serve AI at scale to businesses and public services.
The energy constraint
Large data centres draw substantial and continuous power. In Britain, industrial electricity costs are high by international standards and new grid connections can take years. These factors weigh on investment decisions as much as tax or talent.
Policy responses
The government's AI Opportunities Action Plan proposed AI Growth Zones to speed planning and power access for data-centre sites, alongside expanded public research compute. Delivery will depend on grid reform, local consent and coordination between departments that have not traditionally worked together.
A realistic ambition
Britain is unlikely to match the largest American or Chinese buildouts. It does not need to. A credible aim is sufficient sovereign compute for research, safety evaluation and critical public uses, combined with trusted access to international capacity for everything else.
Editorial note: British Superintelligence is an independent publication and does not represent the UK Government. This analysis distinguishes current evidence from prospective scenarios.