On GDP

When a company builds a data centre, the building of it counts in GDP. Not the output of the data centre — the building of it. Gross domestic product is the sum of consumption, investment, government spending, and net exports, and investment means gross fixed capital formation, which means the concrete, the steel, the electrical work, the cooling plant and the servers. All of it lands in this year's output figures at the moment it is installed, and it lands there whether or not a single useful computation is ever performed in the building.

This has a consequence which is genuinely strange when you first see it. An economy can post excellent growth figures entirely on the strength of building infrastructure that turns out to be worthless. Growth measures activity, not whether the activity was worth undertaking, and building the wrong thing is activity.

So there are two completely separate questions about the AI boom and they get conflated constantly. The first is: what is this spending doing to measured output right now? The answer is: raising it, substantially and measurably, with the size of the effect depending on how much of the equipment was imported, which turns out to matter enormously. The second question is: is the spending going to produce a return? The answer to that is unknown, and — this is the part people find hard — the answer to the second question has almost no bearing on the answer to the first for several years, and possibly for a decade.

from Everywhere But the Statistics: The Macroeconomics of Artificial Intelligence, Told Through Its Bottlenecks (2026)