A Sequoia alum just put $400M into UK tech, and his first demand isn't more models or compute — it's cheaper electricity.
The Summary
- Matt Miller, ex-Sequoia partner and founder of £400M UK VC fund Evantic Capital, says Britain's AI ambitions are stalled by power costs
- His prescription: build affordable baseload power, specifically nuclear, or watch AI investment flow elsewhere
- This isn't abstract infrastructure talk — a former top-tier Silicon Valley investor is telling the UK its power grid is a competitive disadvantage
The Signal
Matt Miller left Sequoia Capital to launch Evantic, a £400M venture fund focused on UK startups. Most people who make that move talk about untapped talent or European market opportunity. Miller's talking about power plants.
That's not random. Training frontier AI models requires industrial-scale electricity. So does running inference at scale. So does keeping agent infrastructure online 24/7. The UK has strong AI research labs and engineering talent, but it's pricing itself out of the infrastructure game. Miller's argument is blunt: if you want to compete in the agent economy, you need power that doesn't bankrupt you.
"Nuclear energy isn't just about carbon targets anymore — it's about keeping AI workloads onshore."
The timing matters. We're in the middle of the great AI infrastructure buildout. Hyperscalers are locking down power purchase agreements in the U.S., building data centers next to substations, and negotiating directly with utilities. The UK, meanwhile, has some of the highest electricity costs in Europe. That's fine if you're running SaaS. It's a killer if you're training models or running millions of agent transactions per day.
Miller's not alone in this diagnosis. Microsoft, Google, and Amazon have all signed nuclear power deals in the past 18 months to secure long-term energy for AI workloads. OpenAI has publicly discussed co-locating compute with power generation. The pattern is clear: the companies building Web4 infrastructure are thinking like utilities now, not just tech firms.
Key forces converging:
- AI compute is moving from "buy GPUs" to "buy power plants"
- Baseload generation (nuclear, geothermal) beats intermittent renewables for 24/7 agent infrastructure
- Countries with cheap, reliable power will attract AI capital; those without won't
The Implication
If you're building AI products in the UK, watch where the big infrastructure money goes next. If nuclear or other baseload projects don't materialize in the next 24 months, expect more compute to migrate to regions with better power economics — parts of the U.S., France, the Nordics. For policymakers, this is a shot across the bow: the AI economy doesn't care about your research pedigree if you can't keep the lights on cheaply.
For investors, Miller's bet is instructive. He's not waiting for the UK to solve this. He's funding companies now, but his public comments are a hedge — build the infrastructure or lose the race. Smart money is starting to think like infrastructure money.