WhichAI / AI Debates / The AI bubble question

The AI bubble question

Is the AI build-out a bubble or the biggest infrastructure cycle ever? · updated 2026-07-30

Big Tech capex is heading past $600B a year while measured AI revenue is an order of magnitude smaller. Whether that gap is a bubble or a rational bet on future demand is the sharpest disagreement in finance right now, and serious people sit on both sides.

>$600B in 2026US Big Tech capex, about 23% of revenue, double pre-ChatGPT intensity
$765B → $1.6TGoldman Sachs modeled annual AI capex, 2026 to 2031
~$600B vs ~tens of Bannual AI revenue needed to justify the spend vs what is measurable today

The concern

The concern: revenue is not catching up, GPU fleets depreciate fast, valuations price in perfection, and AI capex is itself propping up GDP, so a slowdown would hit twice.

The counter-view

The counter-view: unlike 2000, the spenders are the most profitable companies in history using cash, not debt; compute is a scarce productive asset; and infrastructure overbuild historically leaves useful capacity even when investors lose.

Where it stands (July 2026): capex keeps accelerating, analysts keep publishing duelling frameworks, and nobody has the counterfactual. Not investment advice.

Sources: Goldman Sachs, Tracking Trillions · Fidelity, five bubble signs · Boom, Bubble, or Buildout? (arXiv multi-method study)