The next bottleneck is not silicon. It is substations.

The queue is the market.

Every conversation about AI capacity starts with GPUs and ends with a transformer. The chip is the glamorous constraint: allocations, lead times, who is friends with whom in Taiwan. But fabrication capacity is a supply-chain problem, and supply chains catch up. The grid is a different kind of problem. A fab is built in years. A transmission line is built in decades, if it is built at all.

In the markets we work in, the grid queue is where ambition goes to wait. Britain’s connection queue swelled at one point to more than ten times the country’s peak demand, most of it projects that would never be built, and connection dates were being offered well into the 2030s. Continental Europe is not far behind. Northern Virginia, the densest Data Centre market on earth, now rations new load. The price of entry to the AI buildout is no longer a GPU order. It is a megawatt with a date attached.

What twenty years in renewables taught us.

Solar and wind developers lived in the interconnection queue long before Data Centres arrived in it. We spent the formative decades of our careers there: originating grid positions, negotiating connection agreements in three languages, learning which substations had headroom before the market priced it. That experience produces three convictions.

First, the grid is a relationship, not a form. A connection offer is the end of a process that starts years earlier, with network planners who need to trust that your project is real. Second, nobody jumps a queue. Reforms in Britain and elsewhere are now clearing speculative applications and reordering by readiness: land, consent, capital and a credible programme. The projects that survive are the ones that can prove they exist. Third, the fastest way through is to bring generation with you. A developer who arrives with new clean power alongside new load is solving the operator’s problem, not adding to it.

“We spent two decades in the interconnection queue before anyone in this industry knew it existed. The grid does not care how many GPUs you have ordered. It cares whether you are real.”Jamie MacDonald-Murray, Chairman & CEO, Eppur

Honest queue-jumping.

There is a dishonest version of speed, and the industry has seen plenty of it: gas turbines behind the meter with a vague promise to connect later, capacity announced on land nobody controls, gigawatts pencilled against substations that were never asked. It makes for a good press release and a bad decade.

The honest version is readiness. It means securing the land option before the grid application, not after. It means designing the campus as a controlled energy system from the first drawing: batteries that let the site flex when the grid is tight, workloads that can shift, heat that goes somewhere useful. It means additionality, contracting new renewable generation so the grid has more clean power after you arrive than it had before. Regulators can tell the difference, and increasingly the rules are written so that the difference decides who connects first.

“Every one of our connection agreements started years before the application, in a room with a network planner who needed to believe the site was real. You cannot fast-track that. You can only turn up ready.”Peter King, Co-founder, Eppur

Where this leaves the AI buildout.

The pace of AI capacity over the next decade will be set by cables, transformers and consents, not by wafers. The developers who win will be the ones holding grid positions they originated before the consensus moved, on land they control, with power partners they have known for years. That is precisely the discipline our platform is built on: 952 MW of anchor projects with land and power confirmed, 2 GW controlled and owned, every site labelled by the stage it has actually reached.

“Silicon is a supply-chain problem, and supply chains catch up. A substation is a decade of relationships, permits and patience. That is the moat, and it cannot be borrowed.”Jamie MacDonald-Murray, Chairman & CEO, Eppur

The next gigawatt will not be won in the fab. It will be won at the substation.