Opinion · July 2026
One Day, Four Companies, One Question: Is AI Capex Still Paying for Itself?
Thursday’s selloff had a lot of separate-looking causes — a Netflix guidance miss, a TSMC capex raise, an IBM earnings call, a Gemini delay report — and the market’s instinct was to treat them as four unrelated stories that happened to land the same day. I don’t think they are. Read together, I think they’re the same question asked from four different seats: is AI infrastructure spending still generating a return that justifies its size, or is the market starting to price in the possibility that it isn’t?
Four seats, one question
Taiwan Semiconductor’s decision to raise 2026 capex to $64 billion is, on its face, a supply-side confidence signal — a company willing to spend more because it believes demand justifies it. But the market’s reaction wasn’t confidence, it was concern that the spending curve keeps steepening with no visible ceiling. IBM’s admission that customers are shifting budget from software to hardware is a demand-side version of the same worry: enterprise dollars are chasing infrastructure, and something else is losing that fight. Netflix’s guidance miss, while not an AI-infrastructure story on its face, landed the same day as a reminder that even a durable, profitable, non-AI business can disappoint the market on forward growth alone — exactly the standard AI-infrastructure spending is eventually going to be held to.
Where this connects to my own thesis
This is the live version of the argument I made at length in the Physical Limits of Compute whitepaper: the real risk in this cycle isn’t a single bad print, it’s the moment the market decides capex growth and demand growth have quietly decoupled. Thursday wasn’t that moment — one session of chip-adjacent weakness against a backdrop of continued real revenue growth (Netflix’s own quarter, TSMC’s own guidance) isn’t evidence the thesis broke. But it’s the kind of day worth logging precisely because the question it raised doesn’t go away just because the tape recovers tomorrow.
Update, July 19, 2026: I went and checked
This piece asked the question. I didn’t actually answer it here — I just noted that it wasn’t going away. Since then I built The AI Capex Reality Check, a full study that pulls Alphabet’s and Meta’s own Q1 2026 SEC filings and puts a real number on the gap: Alphabet generated $0.28 of free cash flow for every $1 of that quarter’s capex; Meta generated $0.65. Neither company is self-funding its current build-out pace out of cash alone right now. That’s not the same as saying the thesis is broken — the full study lays out the bull and bear case honestly — but it's the actual answer to the question this piece only raised.