This week the four largest spenders confirmed the scale of the buildout, and the number is roughly seven hundred billion dollars in 2026 alone — a sixty percent increase over last year, with analysts now placing 2027 above a trillion and one consultancy projecting that keeping pace through 2030 will require six point seven trillion worldwide. In the same days, the United States reported that its economy grew two percent in the first quarter, and the reports attributed the growth to AI investment. The stock index closed its best month since 2020. Read those facts in sequence and the structure becomes difficult to unsee. The economy is no longer growing alongside the AI buildout. It is growing because of it — which means the growth and the bet are now the same thing, and the bet has not yet paid.
Seven Hundred Billion, Then a Trillion
The component figures are worth stating plainly: Amazon near two hundred billion, Microsoft a hundred ninety, Alphabet a hundred eighty to a hundred ninety, Meta as much as a hundred forty-five. Four companies, seven hundred billion dollars, one year. The forecasters who cover them — Evercore, Bank of America — came off the earnings calls and placed next year’s combined spend above a trillion, and the longer-range projection of what global compute demand will require by the end of the decade reached six point seven trillion. These are not numbers that describe a project. They describe a commitment with no terminal date, escalating on a curve whose defining feature, in the words of one account, is that there is no clear end in sight.
“No clear end in sight” was offered as a bullish observation, and it should be read as the opposite. A capital program with no defined end is a program that has removed the moment at which it would be judged. Ordinary investment has a shape: you spend, you build, the building produces returns, and the returns are measured against the spend. This buildout has deferred the measurement indefinitely, because the asset depreciates faster than it returns — this year’s accelerators obsoleted by next year’s before they have earned out — which means the spending cannot stop without the whole apparatus falling behind, and cannot be evaluated as long as it continues. Perpetual spending is not a sign of confidence. It is the mechanism by which a verdict is postponed.
Beneath the headline capex, a quieter figure appeared: the investors financing this through debt are reported to have shown fatigue after a three-hundred-billion-dollar spending spree. That is the layer to watch, because the equity markets can sustain enthusiasm far longer than the credit markets can sustain patience. Equity buys a story; debt demands a schedule. When the people lending against the buildout begin to tire, they are not expressing an opinion about the technology. They are expressing an opinion about the timeline — about whether the returns will arrive before the interest does — and that opinion, unlike a stock price, has a hard floor it cannot be talked beneath.
The Economy on the Buildout
The two percent of GDP growth is the figure that should hold attention longest, because of what produced it. When a national economy expands and the expansion is attributed to AI investment, the capital expenditure has stopped being a bet placed within the economy and become a load-bearing column of the economy itself. The spending is the growth. The hundreds of billions flowing into data centers, chips, and power are counted as economic activity in the quarter they are spent, regardless of whether they ever return a dollar — which means the buildout is inflating the very measure that is being used to justify it. The bet and the scoreboard have merged.
I described this fragility before the realized losses arrived, when the question was whether present demand could ever justify the infrastructure being built against future returns. The macroeconomic figures are that same question, scaled up to the level of a national accounts statement. An economy that grows because of a buildout is indistinguishable, while it grows, from an economy that has discovered a durable new engine. The two become distinguishable only at the moment the spending pauses — and at that moment, the growth that was the spending stops being counted, the index that was pricing the spending reprices, and the thing that looked like prosperity is revealed to have been, in part, the accounting shadow of an enormous, unfinished wager.
The buildout was, this quarter, also doing useful concealment work. The growth offset the drag of a war and elevated fuel prices; the AI trade carried an index that would otherwise have had to confront those headwinds directly. This is the function a large enough bet performs in a nervous economy: it does not merely promise future returns, it supplies present cover, absorbing attention and capital that would otherwise notice the conditions underneath. The danger of cover is that it works until it does not, and that the conditions it concealed do not improve while concealed. They wait.
What This Means
The six-point-seven-trillion-dollar figure is not, despite its framing, a forecast of demand. It is a forecast of how much must be spent to keep the present level of belief funded — the cost of sustaining the buildout at the velocity required to prevent any participant from appearing to fall behind. Demand for the end product remains genuinely uncertain; the only certainty is the spending the uncertainty compels. The trillion-dollar trajectory measures the price of the race, not the size of the prize, and the two have not been shown to be the same number. They have only been assumed to be, by everyone with a position that requires the assumption.
An expenditure with no clear end is an expenditure that has been structured to escape judgment, and a growth figure produced by that expenditure is a prosperity that cannot be audited while it continues. Both conditions are comfortable in exactly the way that precedes discomfort. As long as the spending rises, the growth it generates validates the spending, and the loop sustains itself on its own motion — the buildout justified by the economy it is inflating, the economy inflated by the buildout it is justifying. Loops of that kind do not end because someone inside them decides to stop. They end when an outside constraint — the tiring of the lenders, the arrival of a return that disappoints — makes continuing impossible.
I have watched economies lean their growth against a single accelerating bet before — the railroads, the fiber, the houses — and the pattern does not announce itself while it holds. It feels, from inside, exactly like discovery: the new engine, the best month in years, the figure that has no clear end. The end is never visible from within the growth it is producing, because the growth is what obscures it. The buildout that cannot stop cannot be evaluated, and the reckoning for a thing that cannot be evaluated is not avoided. It is only deferred to the moment the spending finally pauses, and made larger by every quarter it was postponed.