Every so often the machine that runs the economy starts making a noise, and someone has to explain it in plain words — not after the fact, in the soothing tense of a bank's outlook letter, but while it's still happening and there's still time to ask a question. That is what this family of series has been doing since the start of 2026. The Stack is the newest of them, and it takes apart the loudest machine of the moment: the trillion-dollar buildout of artificial intelligence.
/ Start HereThe January 2026 Baseline
Every investigation needs a baseline — the official version of events, stated plainly, before you start pulling on the threads. For the year 2026, the establishment handed us one, and it's worth looking at closely, because how it was delivered turns out to be part of the story.
Each November, The Economist publishes The World Ahead, its annual attempt to forecast the coming year — the closest thing the global professional class has to a shared script. The 2026 edition led with a warning that should sound familiar by the end of this series: rich countries are living beyond their means, and the risk of a bond-market crisis is rising.1 That is the baseline. Calm, sourced, authoritative, and — in the way of all such documents — carefully arranged so that naming the danger and doing nothing about it feel like the same responsible act.
Here is the part worth pausing on. That forecast was offered to readers not only as a magazine, but as one of Google's first NotebookLM “featured notebooks” — an AI you can ask questions of, that answers in a friendly synthesized voice, draws you a mind map, and reads the whole thing back as an audio overview, all grounded in The Economist's own pages.2 Sit with the loop for a second. The establishment's forecast of the year is now narrated to you by the exact technology whose economics this series spends seven parts taking apart. The map is being read aloud, in a reassuring voice, by the thing being mapped.
None of this is a knock on The Economist, which is very good at what it does, or on the tool, which is genuinely useful. A baseline isn't the enemy. It's the starting line. The trouble is that a summary — human or machine — is built to tell you what is expected, in a voice calibrated to keep you calm. It is not built to ask who benefits, who pays, and who is left holding the bag when the forecast lands. Those are the questions this family of series exists to ask, in the register a forecast can't use. The baseline tells you a bond-market crisis is possible. This series follows the money to the people who will absorb it.
/ The FamilyCritical Thinking, Applied to the Year as It Happens
The Stack doesn't stand alone. It's the latest entry in a running project: take a current event the establishment has already summarized, and apply to it the thing a summary can't — sustained, sourced, skeptical critical thinking. Every part in the family follows the same discipline. Read the actual filing, not the press release. Separate a fee from real ownership. Steelman the optimistic case as hard as the pessimistic one, then follow the evidence wherever it goes, even when it embarrasses the argument. Date every claim, because facts have a shelf life. And never reach for a grand theory to explain what a bank statement explains better.
The Investigation in Order
One argument, in the order the year forced it
Five investigations, written as the events arrived. Read top to bottom and you watch the argument grow — and watch what each dispatch couldn’t yet see become the subject of the next.
- 01
Read the economy as five gauges on one feedback loop, and called the bond-market repricing before it landed — but not yet the hand on the valve.
- 02The Domestic MachineMar 2026
Mapped the enforcement architecture and set the honest-reckoning self-audit this family now inherits as a rule — still reading enforcement as a domestic story, not a capital one.
- 03No Kings: The Chalice OverflowsApr 2026
Traced what happens when every institution collapses its judgment into one question — loyal, or disloyal — without yet pricing the assets marked against the code that broke.
- 04The ComputedollarMay 2026
Named the forced absorber: the pension as buyer of last resort in the AI capital cycle — reading compute as finance, before it was clear how much was physics.
- 05The StackJul 2026you are here
Took the compute economy apart floor by floor — down to the power bill and the eleven days of gas under the whole supply — and left the questions open, on purpose.
The Education Machine (Apr 2026) — the same rigor turned on mythology and merit rather than collapse. It sits beside the arc, not inside it, and reads on its own.
A note on register: these are written for readers who don't work on a trading desk. The plumbing — the filings, the index rules, the chip-licensing regime — is meant to be readable on its own merits, with no condescension and no jargon left undefined. Where an argument turns on a number, the number is sourced; where it turns on a guess, the guess is labeled a guess. The idea is that you come away understanding the machine, not just impressed that someone else does.
/ The SeriesThe Seven Layers
The AI economy is a layer cake, and a dollar climbs through all of it on the way to your screen. Each part takes one floor: who's on it, how they make money, and what they quietly hand off to the floor below. Read them in order, or start anywhere — each stands on its own. The last part follows a single dollar up the whole stack and asks who, in the end, is left paying for it.
Somebody builds the building. The people who pour the AI boom take a capped fee and hold no equity in it — while the owner keeps the upside and the risk that the building is obsolete before it opens, and the ratepayer next door covers the power bill.
A mark, not a sale. A quarter of the richest chip company's profit isn't cash from a customer — it's a number it wrote down about stakes it owns in the labs that buy its chips. And an island with eleven days of gas sits under the whole supply.
The promise, not the payment. Roughly two trillion dollars of “backlog” — contracts booked against customers who haven't earned the money to pay yet — and a rating agency quietly saying so out loud.
Everyone's revenue is someone else's capex. The circular ring where the tech giants' profits are partly just marked-up bets on the labs — and the open-weight challengers threatening to make the whole thing cheap.
The $200 seat that costs $8,000 to serve. The tools that run the models — and the moment the subscriptions priced below cost started getting clawed back, plan by plan.
Own the data, rent the model. When a small model trained on a company's own data beats the frontier at that company's task — and the difference between a real moat and a story about a moat priced for perfection.
Follow one dollar up all six floors and it collects six subsidies — each real, each paid by someone who holds no equity in the boom. What happens if the demand at the top is smaller than promised, who wins if AI becomes a commodity, and why the company everyone calls the loser might be reading the tape right. Left, on purpose, with the questions open.
“The upside is a bet on the future. The cost is local, physical, and now.”
/ Where It LandsReading the Receipt
The Stack ends where its sister series began. Part 7 merges back into The Computedollar and, in particular, into The Forced Absorber — a response to J.P. Morgan's 2026 Eye on the Market reports that shows how the AI bubble's downside was engineered, on a published calendar, to land on the passive index money in ordinary people's retirements. Nearly every number in this series traces back to those same two reports by the bank's own strategist. The machine, it turns out, was documented all along. This series just says the sentence the documentation was built to avoid.
Sources for the baseline. [1] The Economist, “The World Ahead 2026” (published November 2025): its launch framing warned that with rich countries living beyond their means, the risk of a bond-market crisis is growing. [2] Google NotebookLM “featured notebooks,” launched 2025 with initial publisher partners including The Economist (via The World Ahead) and The Atlantic — interactive, source-grounded notebooks offering audio overviews and mind maps (Google Labs; TechCrunch; Press Gazette, 2025–2026). Details and delivery are described from public reporting; readers are encouraged to verify against the originals.