The build that outlives the bubble

Speculation that prices itself wrong can still build the future. The 2025–2026 AI capital cycle is testing that paradox at industrial scale. The interesting question is no longer whether the AI market is a bubble. It is what survives the correction.

Caballero's July 15 working paper at MIT and NBER names this third possibility: a belief-supported overvaluation can install enough durable capital that the economy settles into a higher-capital state even after prices revert. The ordinary reading hears "AI bubble" and asks whether the market is right. The pattern underneath asks a different question: what is actually getting built, and will it still be there when the music stops? The build is the real outcome, separable from the price that financed it.

Caballero's mechanism is portable. A temporary overvaluation, a funding glut flush enough to absorb the loss, and a labor-like capital good whose marginal product does not collapse as it accumulates. If enough installs before belief corrects, the economy is structurally stuck in the new steady state with a permanently lower real rate. Capital owners finance the surge and absorb the correction in paper losses. Workers reach higher wages in the destination, but with a smaller slice of national income.

That is the readable frame: a capital-deployment policy question, not a market call. Every AI capex headline is now legible as either progress toward the high-capital state or a transition that ran too short. The reader's pension, job, and electricity bill all run through which one it turns out to be.