AMZN
MSFT
The AI spending spree got a receipt
Goldman Sachs just did the market a favor: it put a hard number on the AI capex boom everyone’s been hand-waving about. According to the note, consensus capital spending for the biggest cloud infrastructure names jumped by $130 billion in just one quarter, landing at $670 billion for 2026.
That’s not a typo. That means the market is now assuming these companies will pour almost everything they make back into servers, chips, data centers, and all the plumbing needed to keep the AI arms race running.
Why investors are getting twitchy
The part that should make you sit up straight is the cash-flow math. Goldman says that capex now equals more than 90% of expected operating cash flow for the cohort. Translation: the old “spend big, but still fund it internally” story is getting harder to believe.
When that much cash is going back out the door, a few things get uglier fast:
- buybacks become tougher to justify
- dividends look less flexible
- debt markets start acting like the backup battery
- credit spreads matter more than they did when these companies were printing mountains of free cash flow
Microsoft and Amazon are in the splash zone
Goldman didn’t single out any one company, which is the polite way of saying the risk is spreading across the whole hyperscaler club. But Microsoft and Amazon are two of the biggest names in that lineup, and both just reminded investors how intense the spending cycle has become.
Microsoft’s capex surged, and its guidance points to roughly $190 billion in full-year 2026 spending. Amazon is right there too, with cash capex jumping and free cash flow getting squeezed to a tiny sliver compared with a year ago. In other words: the AI thesis may still be alive, but the funding plan is getting more dramatic by the quarter.
Big picture
This isn’t a “AI is over” story. It’s a “AI is expensive, and the bill is coming due” story. If the hyperscalers keep turning nearly all their operating cash into capex, investors may need to stop treating these companies like endless cash machines and start valuing them more like heavy industrial utilities with a chatbot side hustle.
Big picture: the AI build-out may still win the race — but the price of admission is rising fast.