The numbers behind the AI build-out have quietly crossed a threshold that should give every investor pause. Goldman Sachs strategists, in a note published this week, just put a hard figure on what the market has been treating as a vague "spending a lot" story.
Goldman Puts a Number on AI Capex Boom
The figure is striking enough that Microsoft Corp.
Here's how Goldman laid it out:
"The scale of spending has increased considerably. Last quarter, consensus capital expenditure estimates for the largest cloud infrastructure companies jumped by $130 billion, reaching $670 billion for 2026. That's equivalent to more than 90% of their expected cash flows this year, according to our strategists."
Read that again. In a single quarter, the consensus expectation for what the largest cloud companies will spend on AI infrastructure jumped by $130 billion - not the total, the upward revision.
That's roughly the size of Switzerland's annual federal budget, added to one cohort's capex line in three months.
The more important number is the second one: 90%.
When capex consumes more than 90% of a group's expected operating cash flow, the math of self-funding breaks. There's almost no internal cash left for buybacks, dividends, or - more importantly - the next leg of capex.
Goldman didn't name individual companies, and that's worth being precise about: the 90% applies to the cohort, not any single name. But Microsoft and Amazon are two of its largest members, and the directional pressure applies to both.
CapEx Steals the Show
The numbers each company reported on April 29 underscore why.
Microsoft posted Q3 capex of $31.9 billion, up 84% year-over-year, and CFO Amy Hood guided full-calendar-year 2026 capex to roughly $190 billion - a 61% increase over 2025 that includes about $25 billion of incremental cost from higher component pricing.
Microsoft stock fell more than 3% in after-hours trading despite a clean beat, with investors focused squarely on capex.
Amazon's print was even starker. Q1 cash capex hit $43.2 billion - a 107% year-over-year jump - and CEO Andy Jassy reaffirmed a full-year 2026 plan of approximately $200 billion.
The cost is showing up in cash flow: Amazon's trailing twelve-month free cash flow has compressed to just $1.2 billion, down from $25.9 billion a year earlier.
And the trend Goldman flagged keeps pointing up. Alphabet also raised 2026 capex guidance to between $180 billion and $190 billion.
CFO Anat Ashkenazi went further, telling investors that 2027 capex will "significantly increase" compared to 2026, citing "unprecedented internal and external demand for AI compute resources."
The Takeaway
The AI thesis is not broken, but the funding model beneath it is shifting.
With nine out of every ten dollars of operating cash flow going straight back into capex - and 2027 already telegraphed to be bigger - buybacks become harder to defend, bond issuance becomes routine and sensitivity to credit spreads goes up.
The era of self-funded hyperscaler AI is, on Goldman's numbers, ending.
