Operations | Monitoring | ITSM | DevOps | Cloud

AI factories don't sell floor space. They produce tokens, the currency of intelligence, and every jump from Hopper to Vera Rubin packs in up to 4x the density. Training wants one giant site. Inference wants to be as close to you as physics allows. So the panel maps out what comes next: edge data centres of 5 to 50MW, handing work between your watch, your phone and the cloud at the lowest possible latency.

Everyone's racing to build gigawatt AI factories by 2035. What if the faster route is already plugged in? Swarms of agents need far more compute than a chat prompt ever did, and the scaling laws say demand isn't slowing. But in the UK, anything over 15MW can mean waiting until 2032 or later for a grid connection. So the panel makes a different case: smaller sites on existing power and fibre, linked into one national campus, live in less time than a single mega-build.

Why can't the UK make its own AI chips? David Sullivan argues that the biggest gap isn't chips, compute or electricity. It's trust. If people won't let AI near their finances, the best model in the world goes unused, and that's where the UK has a real chance to lead. Civo makes complexity a thing of the past.

Open source models are closing in on the frontier. But is running them the same as being sovereign? David Sullivan explains why true sovereignty means owning the whole model lifecycle, and why that's where prototypes stall before production. James Faure shows how distilling small, task-specific models can slash costs. Dr Rosemary Francis argues it's ridiculous to use the most expensive model for every task, and that your harness and engineering process matter more than which model sits at the top of the leaderboard.