Key points
- Microsoft (MSFT) held its capex outlook near $190 billion and rose 7.6% after hours. Meta (META) raised its and fell 8.0%.
- Azure grew 43%, the fastest since early 2022, and crossed $100 billion in annual revenue for the first time.
- Commercial backlog reached $678 billion, up 84% from a year ago. Microsoft signed more than $130 billion of new data center leases in the quarter.
- Meta beat on revenue at $60.8 billion but missed badly on earnings, $6.18 against a $7.14 estimate.
- The chip names did not follow: Nvidia (NVDA) up 0.7%, Micron (MU) down 0.6%, Taiwan Semiconductor (TSM) down 0.3%.
Two of the biggest buyers of AI hardware reported within an hour of each other Wednesday night. One held its spending flat and went up 7.6%. The other raised its spending and got taken out back.
For a year the market has treated capex announcements as the scoreboard. Bigger number, bigger commitment, bigger stock. That stopped working tonight, and I think it stopped working for good.
The reason is not that Microsoft spent less. A capex raise is a promise. Cloud revenue growing 43% against $678 billion of signed commitments is a receipt, and a receipt is worth more to the companies selling the hardware than another promise would have been.
What Microsoft actually reported
Microsoft (MSFT) put up $90.0 billion of revenue for its fiscal fourth quarter, up 18%, against roughly $87.7 billion expected. Azure and other cloud services grew 43%, the fastest rate since early 2022, where the street had about 40%.
| Metric | FQ4 2026 | Change |
|---|---|---|
| Revenue | $90.0B | +18% |
| Operating income | $40.6B | +18% |
| Net income | $35.8B | +31% |
| Microsoft Cloud revenue | $59.3B | +27% |
| Azure and other cloud | over $100B annual | +43% |
| Commercial backlog | $678B | +84% |
| Operating cash flow | $55.4B | n/a |
Azure passed $100 billion of annual revenue for the first time in the fiscal year that ended in June. Microsoft 365 Copilot is above 30 million paid seats.
And the capex line did not move. Microsoft kept its calendar 2026 capital spending outlook at about $190 billion, the same number Amy Hood gave in April. Fourth quarter capital expenditures and finance leases came to $41 billion, up 69%, and the company signed more than $130 billion of new data center leases during the quarter, taking uncommenced lease commitments to $329.1 billion.
"We are advancing the frontier on the cost-to-outcome curve," said Satya Nadella, Microsoft's chief executive.
What Meta reported an hour later
Meta Platforms (META) did the opposite on both counts. Revenue came in at $60.8 billion against $60.2 billion expected, up 28%, with advertising at $59.3 billion. Earnings were $6.18 a share against a $7.14 estimate, which is not a small miss.
Then it raised full-year capital spending guidance to a range of $135 billion to $145 billion, lifting the low end by $10 billion in a single quarter. The stock fell 8.0% after hours, to $540.02 from a $587.00 close.
Microsoft raised nothing and gained 7.6%, to $422.30 from a $392.40 close. That is the entire lesson in two data points. Capex is not the variable. Whether the revenue shows up is the variable.
Why this is the better kind of news for chip stocks
Here is my actual read, and it is a real opinion rather than a hedge. The chip supply chain does not need Microsoft to promise more money. It needs proof that the money already committed is attached to customers who have signed something.
The bear case on AI hardware all summer has been that hyperscaler spending is faith-based, that the compute is being built ahead of demand that may never arrive, and that when the buyers blink the orders vanish. Azure at 43% and $678 billion of signed commitments is the direct answer to that. The money already being spent is underwritten by contracts, not by a slide deck.
That matters more for Nvidia (NVDA), Broadcom (AVGO), Taiwan Semiconductor (TSM) and Micron (MU) than another headline capex number would have. A raise tells you a customer intends to spend. A backlog tells you someone has already agreed to pay for the output.
Let me be clear about what a flat outlook does not mean, because this is where people are going to get it wrong. It does not mean Microsoft is buying less silicon. Fourth quarter capital expenditures and finance leases came in at $41 billion, up 69% from a year ago. The company added more than $130 billion of new data center leases in the same three months and its uncommenced lease commitments now sit at $329.1 billion, the largest quarterly jump it has reported.
The forward number stayed near $190 billion because $190 billion is already an enormous number, not because the buildout is decelerating. Sustaining spending at a record level while proving the revenue behind it is a better outcome for the supply chain than promising a bigger number with nothing underneath it. Flat guidance on a record base is not a warning sign.
Put it next to Alphabet (GOOGL) raising its own capex to $205 billion last week, its third increase this year, and the demand side of this trade looks considerably better tonight than it did at 4 p.m.
The part of my own thesis that does not work
Two honest problems, and anyone selling you this story without them is selling you something.
First, an unchanged outlook means the street does not get an upward revision to plug into its models. The $190 billion is already in everyone’s numbers. If you were waiting on a raise to justify higher estimates for the suppliers, you did not get one tonight. The bullish read here is about durability and risk coming down, not about the order book getting bigger, and only one of those shows up as an earnings revision.
Second, and this is the bigger one, the chip selloff this month was never really about demand. July cost Sandisk 55.3% and Micron 36.1%, and the reasons were supply reasons. CXMT went public in China on a 466% debut, and SK Hynix (SKHY) guided 2026 capital spending 50% higher, to at least $31 billion. More DRAM is coming. Azure growing 43% does nothing about that. Proving demand is real does not fix a glut, and I am not going to pretend it does.
There is also a wrinkle in the earnings themselves. Discrete items added $0.27 to Microsoft's diluted earnings against prior guidance, most of it a $3.2 billion gain on its Anthropic investment. So the profit line got help that has nothing to do with selling cloud. The revenue line and the backlog did not get that help, which is exactly why I am pointing at revenue and not at the earnings beat.
What I am watching
The chip names did not move on any of this. Nvidia rose 0.7% after hours. Micron fell 0.6%, Broadcom fell 0.3% and Taiwan Semiconductor fell 0.3%. The two largest buyers of AI silicon just told the market the demand behind the buildout is contracted, and the suppliers closed the day between 36% and 55% cheaper than they were four weeks ago.
Either the market decides that gap is wrong, or it decides the supply problem outweighs the demand proof. I have owned Microsoft for a while, and I think the revenue number is the most important thing that happened this week. I am also long SK Hynix (SKHY), and I am not adding to it while the industry is announcing a supply expansion. Holding a position and buying more of it are two different decisions, and this week only answered the demand half of the question.
Disclosure: I am long Microsoft (MSFT), which I have held for a while, and long SK Hynix (SKHY). Microsoft figures are from the company's fiscal fourth quarter 2026 results released July 29, 2026, and Meta figures from its second quarter 2026 release the same day. After-hours prices are as of 6:41 p.m. ET on July 29, 2026, and will have moved by the time you read this. This is my own opinion and general market commentary, not investment advice.
