We tested the most-watched leading indicator in AI — it has no lead

The most repeated causal story in AI markets goes: hyperscalers guide capex up → suppliers' revenue follows a few quarters later → so customer capex is a leading indicator you can read ahead of a supplier's print. It sounds so mechanical that almost nobody checks it. We checked it.

We assembled quarterly SEC fundamentals for 98 AI-chain companies (2018–2026), resolved supplier→customer links for the 24 suppliers with mapped customers, and asked the falsifiable version of the question: does a customer's trailing revenue/capex growth at quarter t predict the supplier's revenue acceleration over the next 1–4 quarters — beyond what the supplier's own momentum already tells you? That's 467 supplier-quarters, scored with per-quarter rank correlations (Fama-MacBeth) and Newey-West t-statistics.

The answer is no. The customer-demand pulse carries a mean rank IC of −0.07 at every horizon — statistically indistinguishable from zero, and directionally negative. The capex-only version is significantly negative at one quarter out (IC −0.12, t = −3.1): by the time a customer's reported capex growth is high, the supplier's growth is already peaking and mean-reverting. And after controlling for the supplier's own recent acceleration — the dumb "what grew keeps growing" baseline — the customer signal adds nothing.

Why is the intuition wrong? Because for AI hardware, *the supplier's revenue is the customer's capex. Nvidia books the sale in the same period Microsoft capitalizes the server — often a quarter before the capex shows up in a cash-flow statement. There is no lag to harvest, and the five customers that dominate every supplier's book (Microsoft, Amazon, Alphabet, Meta, Oracle) are the most-watched companies on earth. The classic academic result this idea leans on — Cohen & Frazzini's customer momentum — was about stock-price drift along under-followed* customer links, not fundamentals propagating through mega-caps everyone models in real time.

What the panel does show is boring and useful: growth levels mean-revert hard (high YoY growth today predicts deceleration, IC −0.32 at two quarters), and short-term acceleration persists for a quarter or two (IC +0.52 at one quarter). Neither needs a supply-chain graph.

The honest caveats: 24 mapped suppliers is a thin cross-section; we proxied "surprise" with acceleration (no consensus-estimates data); and it's one cycle, 2019–2026, with COVID inside it. None of that rescues a signal this flat. The aggregate capex cycle itself — level, growth, and whether growth is fading — remains the number everything downstream hangs on. That's why it's our headline read. But reading one company's capex to front-run its supplier's print? The market already did.