How much revenue is needed to justify the current AI spend? (pracap.com)

A provocative industry audit: after asking how much revenue is required to justify today’s AI capex, the author reports widespread agreement from datacenter operators, investors, and engineers that the financial math doesn’t add up. Key technical revisions—shorter asset lives (3–5 years versus a previously assumed 10), annual or biennial GPU obsolescence, and rapidly changing cooling/rack/power designs—mean capital is written down far faster than assumed. Using these faster depreciation curves, breakeven for 2025’s buildout jumps from an earlier $160B estimate to roughly $320–480B; with ~ $400B of capex expected in 2025 and current AI revenue only ~$15–20B, the author estimates ~ $1T of revenue would be needed to break even across 2025–26, and many trillions more to deliver acceptable returns. The implications are systemic: if AI spending outpaces monetization, the industry risks massive capital destruction and macro feedbacks comparable to 19th-century railroad collapses or the fiber-optic bust—because AI datacenters have short useful lives and are often financed on long-term contracts and power commitments. Hyperscalers might subsidize losses for strategic or ecosystem reasons, or governments could intervene for national security, but absent massive future revenue (e.g., broad displacement of high-skilled labor years from now), the buildout exposes investors and the broader economy to a potential “Minsky moment.”