John O’Connell was recently published in Kiplinger, the third in his series on AI concentration risk, laying out four risks that could break the assumption holding up the AI buildout: that today’s spending gets paid back by tomorrow’s revenue. The four are demand risk, with the gap between AI spending and revenue running around $600 billion and widening; capital spending risk, since the hyperscalers’ budgets are quarterly guidance, not contracts; infrastructure bottleneck risk, where power, not chips, is now the binding constraint; and return-on-investment risk, with capex-to-revenue divergence already exceeding the late-1990s telecom buildout that preceded a multi-year correction. His advice is to look through every fund for exposure to the supply-chain layers rather than the familiar ticker, measure overlap across funds rather than within one, and stress-test each holding against a scenario where hyperscaler capex growth slows from 60% to 15%.