How Much Money Do Machines Spend on Discretionary Goods?

Citrini Research published a speculative macro memo written from June 2028, looking back on a cascading economic crisis triggered by AI doing exactly what the bulls promised it would do. It's fiction — clearly labelled as such — but the mechanism it describes is not.

The premise: what if AI capability keeps improving on the trajectory everyone is betting on, companies keep cutting headcount because the maths is obvious, margins expand, earnings beat, stocks rally — and then the 70% of GDP that depends on humans having money to spend quietly falls apart? Each company's decision is rational. The collective outcome is a deflationary spiral with no natural brake.

The best concept in the piece is "Ghost GDP" — output that shows up in the national accounts but never circulates through the real economy. A GPU cluster in North Dakota doing the work of ten thousand white-collar workers in Manhattan registers as productivity growth. It does not register at the restaurant downstairs, the mortgage lender, or the tax base.

The weakest part is the timeline. They compress five-to-ten years of disruption into two, which is how you turn a plausible risk into implausible science fiction. AI agents routing around Mastercard via Solana stablecoins by early 2027? The payments industry moves at the speed of regulation and contract law, not demo day. And the political response is written as if Washington just sits there looking confused while the economy drops 38%. Every actual crisis of this magnitude has produced aggressive intervention well before the third act.

But the core question is sharp, and it's the one almost nobody in the AI bull camp has a good answer for: if AI makes human intelligence abundant and cheap, and the entire economy — labour market, mortgage market, tax code, consumer spending — is built on the assumption that human intelligence is scarce and expensive, what exactly is the adjustment mechanism? "New jobs will appear" is the answer that's been correct for two centuries. It's also an answer that assumes the new jobs require something machines can't do. That assumption is load-bearing, and it's worth checking how much weight it can actually hold.

The piece buries its best line early: "We probably could have figured this out sooner if we just asked how much money machines spend on discretionary goods."

The answer, of course, is zero. Everything else in the piece is just tracing the consequences.