Five years ago a 50 MW campus made news. Today the serious conversations start at a gigawatt. What changed, what it means for grids and capital, and why discipline matters more as the numbers get bigger.
Not long ago, a 50 MW Data Centre was a headline. It was the size of a serious regional facility, and a developer with two of them had a business. Today the conversations that matter open at a gigawatt, and multi-gigawatt campuses are announced weekly. The unit of ambition has moved by a factor of twenty in half a decade.
What changed is simple to state. Models grew, and the compute to train them grew faster. A single GPU rack now draws in a day what a legacy rack drew in a fortnight. The hyperscalers began measuring annual capital expenditure in the hundreds of billions. Governments joined in with sovereign programmes of their own. Demand of that scale does not fit in a 50 MW building, and so the building grew to fit the demand.
It is worth pausing on the number, because the industry has started using it casually. A gigawatt of load, running around the clock, uses more electricity in a year than three million homes. It needs its own substation, often its own transmission line, and in most markets its own generation. It costs well north of ten billion dollars before a single GPU is purchased. It is, in every sense that matters to a grid operator or a finance minister, a piece of national infrastructure.
That is a very different proposition from the Data Centre of the previous era, and it demands a very different developer. A gigawatt is not a real-estate play with a power connection. It is an energy project with a building on top.
The industry has a habit, and it is getting worse. Gigawatts are announced against land nobody controls, grid capacity nobody has been offered and customers nobody has signed. The press releases are indistinguishable from the real thing, and the market has begun to price them accordingly, which is to say with growing scepticism. We call them vapourwatts. They are not harmless. They clog connection queues, inflate land prices and make it harder for real projects to be believed.
“Anyone can announce a gigawatt. The market is drowning in vapourwatts. The number that matters is the one with land, power and a date attached.”Jamie MacDonald-Murray, Chairman & CEO, Eppur
Our answer is a reporting discipline we would like to see become the industry standard. Two tiers, published. Anchor projects, where land and power are confirmed. Pipeline, where they are not yet, labelled by the stage each site has actually reached: controlled and owned, in development, in due diligence. As sites move, the labels move with them. It is a less flattering way to present a portfolio. It is also the only honest one.
Here is the uncomfortable truth about bigger numbers: mistakes scale with them. A wrong assumption about grid headroom on a 50 MW site is a delay. On a gigawatt site it is a stranded asset. An offtake counterparty that fails on a small campus is a bad quarter. On a gigawatt campus it is a restructuring. Ambition at this scale has to be matched by a platform that can carry it, and that means holding power, permits and partners in house, structuring capital to the duration of the asset, and delivering on every promise made when the meeting started.
We report 5.7 GW of pipeline across twelve markets and three continents, of which 2 GW is controlled and owned and 952 MW sits in anchor projects with land and power confirmed. Our roadmap is written in the same units: 952 MW, then 2 GW, then 5.7 GW. Every one of those numbers has a stage attached, and we will publish each change as it happens.
“The gigawatt is the new unit of ambition. It should also be the new unit of accountability.”Jamie MacDonald-Murray, Chairman & CEO, Eppur
The numbers will keep getting bigger. The developers who deserve to build them will be the ones who can show, site by site, that the number is real.
Chips get the headlines; grid connections decide the winners. Why interconnection queues, not fabrication capacity, will set the pace of the AI buildout, and what twenty years in renewables taught us about jumping them honestly.
Read the essay →Nations are concluding that intelligence is critical infrastructure, and that it should live at home. What sovereign AI programmes mean for where the next gigawatts get built, and who gets to build them.
Read the essay →