AI industry needs $3 trillion（約480兆円） in revenue to justify $1.5 trillion（約240兆円） infrastructure spend

The crux of this news lies in a widening gap between investment and return. Three years ago, Sequoia's Cahn estimated that $200 billion（約32兆円） in revenue would be needed to justify AI infrastructure spending. Today, after three years of hyperscaling, that figure has exploded to $3 trillion（約480兆円）—reflecting both the unprecedented scale of compute buildout and rising costs of memory and specialized chips. The body itself acknowledges this is probably an underestimate, as "the required revenue per GW of CapEx has sharply increased due to bottleneck dynamics and rising costs of construction."

The vulnerability centers on 2028, when hyperscalers have publicly committed to realizing cash-flow payback from their accumulated chip purchases. Cahn's calculation and Slok's analysis both hinge on whether organizations and users will generate enough demand—and willingness to pay—to support that return. Yet the article identifies a countervailing trend: organizations are adopting cheaper, open-weight models (often from Chinese vendors) rather than relying on premium frontier labs, and token efficiency improvements mean users may not increase their token consumption proportionally even as AI usage spreads. If demand falls short or price compression accelerates, the consequence reaches beyond the AI sector itself. Slok explicitly warns that a miss on these cash-flow goals could "risk tipping the economy into recession and the S&P 500 into a correction," because "so much riding on so few names."