The AI marathon’s biggest threat suggests the finish line is nowhere in sight

Gavekal Research economist Will Denyer argues the AI-driven capex boom in the U.S. remains intact because returns on invested capital still exceed the weighted average cost of capital (a wide “Wicksellian spread”), supporting continued investment in data-center hardware. Much of the spending benefits Asian suppliers (TSM, SK Hynix, Samsung, SoftBank), limiting direct GDP gains in the U.S., but the indirect wealth, productivity and power-sector impacts are meaningful. Key risks are a narrowing spread from higher inflation/higher rates (and sustained high oil prices) and potential electricity shortfalls, prompting a recommendation to favor U.S. grid, natural gas and renewable plays. The note also flags large pools of deployable capital — $7.6 trillion in money-market funds and up to $2.6 trillion in additional bank lending — and forecasts AI capex of around $1 trillion by 2027.