Somewhere in the last two years the data centre industry stopped being a property business and became a queueing business. Access to electricity, rather than the price of electricity, is now the factor that decides where a facility gets built. That is a short sentence with an enormous amount of rearranged capital behind it.
For twenty years the siting calculation was a spreadsheet. Developers compared tariffs across jurisdictions, added land cost, subtracted tax incentives, checked fibre routes and latency to the nearest exchange, and picked the cheapest column. Power was an input with a number attached, and numbers can be optimised. Nordic markets sold cold air and hydro. Ireland sold tax treatment and transatlantic cable. Poland and the Czech Republic sold land and labour at a discount to Frankfurt. Everyone was competing on the same axis, which was cost per megawatt hour, and the winner was whoever could shave it.
Artificial intelligence training loads broke the axis. When a single campus wants the consumption of a mid-sized town, the relevant question stops being what the electricity costs and becomes whether anyone can physically deliver it to that coordinate inside the decade. A cheap megawatt that does not exist is not cheap. It is unavailable. And availability is not a market variable a developer can negotiate. It is a function of grid topology that was planned in the 1970s by engineers who were not anticipating this.
There is a cafe version of this that every operator will recognise. You can buy the best beans on the continent, hire a barista with competition placings and fit out a room that photographs beautifully, and if the building has one twenty-amp socket you are selling filter coffee to a queue of forty people who wanted espresso. The constraint is never the thing on the menu. It is the thing in the wall. The entire AI buildout has just discovered the thing in the wall.
What this does to the operating mechanics is more interesting than the headline. Site acquisition has been inverted. Developers used to buy land and then arrange power. Now they buy power and then find land near it, which means the valuable asset in a transaction is frequently not the plot at all but an interconnection agreement attached to it, or proximity to a substation with spare capacity, or a decommissioned industrial site that retains a heavy-industry grid connection from a smelter or paper mill that closed in 1994. Brownfield land with an oversized transformer has become more valuable than clean greenfield with a view.
Second, the critical path has moved to manufacturing. A data centre shell can be built fast, because concrete, steel and glass have deep supply chains and competitive contractors. High-voltage transformers and switchgear do not. There are few factories, long order books and specialist labour, and no amount of construction financing accelerates a transformer that is scheduled. The industry has gone from being constrained by capital, which is abundant, to being constrained by heavy electrical equipment, which is not.
Third, and this is where it reshapes regions rather than balance sheets, the decision has shifted from the developer to the utility. A transmission operator deciding which interconnection requests to energise first is now, functionally, an economic development agency. It is allocating the single scarcest input in the digital economy, and it is doing so through a technical process designed for connecting factories. Governments across Central and Eastern Europe are working out that their grid queue is an industrial policy instrument they did not know they owned.
The result is a market where the winners will not be the operators with the cheapest power. They will be the ones who got in the queue earliest, held their place, and had the patience to wait at a substation while everyone else was still comparing tariffs.
