The line Goolsbee drew is the one worth reading twice, because he put the buildout on the safer side of it and then said the border is close: firms competing for the same resources is a sector rebalance, which relative prices sort out on their own, while aggregate overheating is the thing the funds rate exists to lean against, and his phrasing was that we are "not far" from the first turning into the second [3]. On his own taxonomy, then, the policy instrument becomes relevant only after a threshold he has now publicly named [14].
The evidence he offered for it is anecdote of a very specific kind, which is firms on his Seventh District tours saying they are scaling back plans because construction labor has got too expensive and HVAC cannot be had [4]. That is capital expenditure which never becomes a number: the deferred warehouse and the postponed line expansion do not print anywhere, so the crowding out is audible on a tour and close to invisible in the aggregate investment series that the same buildout is inflating [1].
Which is why the Wells Fargo item carried in the same newsletter, that we are about to hit peak capex growth [5], is less reassuring than it sounds. Peak growth is a statement about the rate of change and not the level, and if year-over-year growth tops out while staying positive, spending keeps rising and demand for inputs keeps rising with it [6]. An electrician does not get cheaper because a comparison base flattened.
The politics are running ahead of the economics, or rather, ahead of the Fed's version of the economics. A leaked National Republican Senatorial Committee memo calls AI a "toxic brand" and describes data centers as "the anchor hanging around Husted's neck" in the Ohio Senate race against Sherrod Brown [9], Pimco's Libby Cantrill reads the leak as deliberate, aimed at Republicans who have not yet worked out the cost [10], and Bespoke Investment Group has August searches for "midterm election" at least 2.5 times any August since 2004, which is to say at least 150% above the previous high [7][8]. Trump's position, posted on Truth Social, is that communities opposing data centers want to end up "backwards and poor" [12].
This is probably wrong, but the useful version of the thesis is that crowding-out language gives the Fed something it does not currently have, namely a way to describe stubborn goods and construction inflation as a composition problem rather than an excess-demand problem, and a composition problem is one you can watch rather than act on. The counter-thesis is in the same newsletter: Fortune's own framing, citing Pantheon Macroeconomics, is that AI capex is large enough that without it there would be fewer jobs and slower GDP growth [13]. A committee looking at that chart may quietly prefer the rebalance to the alternative, and Goolsbee himself said the rebalance is what we have, not overheating [3].
What would show the thesis is wrong: if the scale-backs he heard about resolve as an ordinary supply story, with construction wages and equipment lead times normalising as capacity arrives, then the competition for resources was a queue and not a crowd-out, and the Fed never has to price it. One regional president talking to a magazine is not a committee view. It is a vocabulary being tested in public.
