Big Tech's AI Capex Is Now a Macro Variable

Combined hyperscaler AI capex for 2026 is on track to exceed $400 billion. That changes how we think about every sector that touches power, cooling and silicon.

Dan Gallagher · WSJ Heard on the Street · April 23, 2026

A line item that moves the GDP print

Gallagher's column compiles the latest hyperscaler capex guidance — Microsoft, Google, Meta, Amazon, Oracle — and notes that the combined 2026 number is on track to exceed $400 billion, larger than the entire annual capex of the U.S. utility sector. AI capex is now large enough that it shows up in macro statistics, which is a transition in kind, not just degree.

How we are positioned

We are overweight the immediate beneficiaries (NVDA, AVGO, TXN added today as our preferred analog/power-management proxy), the second-derivative beneficiaries (Power & Infrastructure: ETN, VST, GEV), and the inference-cost-curve story (GOOGL's new in-house chip, MSFT's Azure premium pricing). Underweight the names whose moats depend on training-cost stability (smaller foundation-model labs and their public proxies).

What This Means For Your Book

ADD TXN today as analog/power-management proxy on the AI capex curve. Maintain NVDA, AVGO, GOOGL, MSFT overweights and Power & Infra (ETN, VST, GEV).

AI capex is now a macro variable. Position the second-derivative as well as the chips themselves.

Want to talk through how this affects your plan?

15-minute Q1 review — no pressure, just an update on positioning.

Schedule a Call