Microsoft Held Capex Flat and Meta's Cash Flow Went to Zero

Microsoft and Meta both reported on the evening of July 29. Both are spending historic sums on AI infrastructure. Both told investors demand exceeds what they can build. Microsoft's stock rose roughly 8% after hours. Meta's fell as much as 10%, settling around -7%.

The gap between those two reactions is the most useful thing either company disclosed, because it is not a gap about AI. It is a gap about cash.

What Microsoft actually said

Revenue came in at $90.01 billion for fiscal Q4, up 18% year over year against an LSEG consensus near $87.6 billion. Azure grew 43% — including on a constant-currency basis — accelerating from 40% the prior quarter and beating a FactSet consensus around 40.3%. Azure crossed $100 billion in annual revenue for the first time.

Capital expenditure, including finance leases, jumped 69% to roughly $41 billion in the quarter. That is the largest capex quarter in the company's history, and by itself it is not a positive datapoint.

The positive datapoint was the sentence that came after it: Microsoft held its capital expenditure outlook steady.

Going into the print, the buy side had convinced itself Microsoft would follow Alphabet, which a week earlier raised full-year guidance to $195–205 billion and watched its stock fall about 4% despite an otherwise exceptional quarter. Previews circulating before the call put Microsoft's FY2027 capex expectation in the $255–260 billion range — roughly a 35% step up. Microsoft declined to validate the escalation.

That is the whole trade. Azure accelerating to 43% is a demand signal, but demand signals have been abundant all year. What was scarce was any hyperscaler willing to say the spending curve has a ceiling. Microsoft, alone among the three that reported in July, implied one.

What Meta actually said

Meta's top line was arguably better. Revenue of $60.80 billion, up 28% year over year, beat expectations comfortably. In any other capex regime that is a blowout quarter.

Everything under it went the wrong direction. Net income fell 14% to $15.85 billion. Operating income fell 8%. Operating margin compressed to 31% from 43% a year earlier. Diluted EPS came in at $6.18, down from $7.14. Even Family of Apps operating income — the profitable, non-speculative core of the business — slipped to $23.4 billion from $25.0 billion.

Then the number that moved the stock. Quarterly capex of $31.08 billion, nearly double the year-ago figure, against operating cash flow of $31.9 billion. That leaves free cash flow of $784 million.

Meta averaged roughly $12 billion of quarterly free cash flow across the prior eight quarters. It just printed a rounding error.

Full-year 2026 capex guidance moved to $130–145 billion, raising the floor from $125 billion while leaving the ceiling untouched. Meta has spent $50.9 billion in the first half, which means the guidance implies an accelerating second half rather than a plateau. CFO Susan Li declined to give a 2027 number and said the company expects to remain demand constrained, adding that Meta has "more profitable uses for computing power than computing power to use."

The distinction the market drew

Both companies are supply-constrained. Both are spending at a rate that would have been incomprehensible three years ago. The difference is that Microsoft is funding its buildout out of a commercial cloud business that bills external customers on contracted terms, and Meta is funding its buildout out of an advertising business that has to keep clearing the bill every quarter with nothing contracted behind it.

When capex was 30% of operating cash flow, that distinction was academic. At 97%, it is the entire investment case.

This is the shift that has been building since June, when the Magnificent Seven shed $2.3 trillion in a single month. The market spent 2025 asking whether AI infrastructure would generate returns. It spent the first half of 2026 asking when. It is now asking a narrower and much less forgiving question: which balance sheets survive the interval between spending and returning?

Consensus has 2026 hyperscaler capex rising from roughly $384 billion to $682 billion, and 2027 landing near $878 billion — over $1.5 trillion across two years. Nobody on either call disputed those trajectories. What separated the two reactions was whether the company disclosing them still had visible slack.

Zuckerberg's tell

Asked about monetizing surplus compute, Mark Zuckerberg confirmed Meta is fielding interest: "We're getting a lot of offers for compute at a significant premium for what we paid for it." He then declined to lean into it — "It would be foolish to basically just sell all of the compute and take a short-term profit" — arguing the higher-margin path is selling intelligence rather than raw capacity.

Strategically that is defensible. Renting GPUs is a commodity business with commodity margins; Meta Compute, floated earlier this month, was always the lower-ambition version of the plan.

But read against a $784 million free cash flow quarter, the answer lands differently. Meta is holding compute it could sell at a premium, in a quarter where its cash generation went to approximately zero, on the argument that a better use will materialize later. That may well be correct. It is also, precisely, the bet the market is currently refusing to underwrite on faith.

The read

Nothing in either report suggests AI demand is softening. Azure accelerated. Meta's ad business grew 28%. Both companies say they cannot build fast enough.

What changed on July 29 is that the two reports were graded on completely different criteria, and the criterion that mattered was not growth. It was whether the spending line still has an owner-declared limit and enough cash flow underneath it to absorb a bad quarter.

Microsoft offered a ceiling and got an 8% rally on a stock still down 17.4% year to date. Meta offered a raised floor, a deferred 2027 number, and $784 million of slack — and got sold.

The buildout is not being questioned. The financing of it is.