Early last month, I spent a few minutes editorializing around the historical relationship between capex booms and US monetary policy.

Long story short, spending bonanzas tend to beget higher Fed funds eventually. Rate hikes on “long and variable lags,” to make the obvious joke.

The relationship’s intuitive. Sort of. If business (and particularly business spending) is booming, the neutral rate’s probably higher, at least in the near- to medium-term.

If failing to cut when the neutral rate moves lower is to countenance “passive” tightening, then failing to raise rates when neutral’s higher is to passively ease.

In the latter scenario, you’re courting inflation. Unless of course the prevailing capex boom’s in the service of standing up a technology with the potential to bring about structural disinflation. In that case, you can argue for keeping rates unchanged. Or at least for not raising them aggressively. That’s basically Kevin Warsh’s argument right now.

With all of that in mind, the figure below’s worth highlighting.

That’s from BMO’s Ian Lyngen and Vail Hartman, and it shows you the extent to which the US capex boom’s running away from Fed funds.

“The AI buildout has been responsible for the bulk of economic momentum thus far in 2026 [and] non-residential business fixed investment is currently contributing +1.15% to the Atlanta Fed’s Q2 2026 GDP estimate,” Lyngen and Hartman wrote, adding that “beyond the positive contribution to growth, there is an active debate regarding [the] impact on inflation and the neutral rate.”

This conversation became more urgent — or at least more salient — in the wake of Apple’s price hikes. As discussed here, the oil shock may be over, but the “iShock” could be just beginning.

All that capex is pushing up component costs, and while the rate of spending may slow, overall outlays will be enormous for the foreseeable future.

The figure on the left, above, from Goldman, shows you the latest estimates for hyper-scaler capex growth alongside the revision trajectory by year (on the right).

“Hyper-scalers’ 2026 capex budgets are mostly set while 2027 budgets are in early planning stages,” the bank’s Ben Snider remarked. “The pattern of capex estimate[s] during the last few years suggests revisions to 2027 [are] more likely as we near the turn of the calendar.”

Goldman’s company analysts see upside to consensus 2027 hyper-scaler capex forecasts. They were right about that this year, and I suspect they’ll be right about it again, which is to say the big spenders are more likely to overshoot than undershoot when it comes to AI investment. That, in turn, could mean AI-related inflation sticks around longer than expected.

“Inflationary angst has rotated toward the AI buildout as energy prices have retraced to pre-war levels,” BMO’s US rates team went on, in the same Thursday note cited above. “Even if AI ultimately proves disinflationary over the long-term, the build-out phase has been adding to computer software inflation among other categories that are exposed to the infrastructure surge.”