Goldman says Big Tech's AI splurge could bite back

The AI party isn't free

Goldman Sachs just reminded everyone that hyperscalers don't get to live in the cloud rent-free forever. Amazon, Microsoft, Alphabet, Meta, and Oracle have been pouring money into AI infrastructure, and Goldman thinks the math is starting to look less like a gold rush and more like a very expensive gym membership.

According to the bank, those companies could spend a jaw-dropping $770 billion on capex in 2026. That’s roughly equal to all the cash flow from operations those names generate combined, which is the financial version of emptying the pantry to build a bigger kitchen.

Why investors should care

The issue isn’t just the upfront spend. Once the data centers, chips, and servers are in place, depreciation kicks in like a recurring subscription you forgot to cancel. Goldman says depreciation and amortization could rise from 7% of hyperscaler revenue in 2022 to 12% by 2027, which is a fancy way of saying margins could get squeezed even if AI demand stays hot.

And there’s a second shoe dropping: buybacks are getting less generous, debt is climbing, and share counts are starting to drift higher. That’s a nasty combo if you own these stocks for their old-school “asset-light, cash machine” vibe.

The bigger picture

The silver lining? This only matters if the AI buildout doesn’t eventually pay for itself. Goldman’s point is basically: the long-term winners will be the ones who turn all this capex into actual monetization, not just prettier earnings call slides.

Big picture: AI can still be the engine. But right now, the scoreboard also includes a very real utility bill.