Microsoft trimmed 2026 capex $5 billion on chip delays

Two of the largest buyers of AI compute spent less than expected this quarter, and both pointed at 2027 for where the missing capacity goes. The market read the shortfall as a change of mind.

Microsoft cut its 2026 capital expenditure forecast to $170 billion from $175 billion and said chip supply delays would push some capacity into the first half of 2027, CNBC reported. The same quarter had Azure sales growing 48%. Those two facts do not combine into a retreat. A $5 billion trim alongside that growth rate is a company that cannot take delivery fast enough, by its own stated reason.

Alphabet arrived at a similar place by a different route. Third quarter capital expenditure came in at $51.36 billion against the $55.09 billion analysts expected, a shortfall of $3.73 billion, while 2026 guidance was held unchanged and a significant increase was flagged for 2027, CNBC reported. Cloud revenue more than doubled to $30.7 billion in the same quarter. Shares fell about 10%.

A miss and a cut are not the same thing

Neither company lowered the trajectory. Microsoft moved capacity into the first half of 2027 and named the reason. Alphabet left 2026 guidance where it was and said next year goes up. In both prints the 2026 number came down relative to what was expected, and in both the deferred spending has a destination.

That distinction matters because the two kinds of capex miss look identical in a quarterly release and imply opposite things about the following year. Spending less because customers are not there shrinks the forward number. Spending less because the hardware has not arrived moves it to the right. Only one of those is visible in this quarter's revenue lines, and in both cases those lines went up.

The repricing went the other way

More than $2 trillion of Nasdaq 100 market value is set to be erased in October, with the index down over 5.8% for the month so far, as doubts deepen about the returns on AI spending, Reuters reported. That is a genuine verdict and real money, and it is not obviously reading the same thing the filings said. The returns question is open. The question of whether the buildout slowed by choice is answered differently in the two releases that triggered the move.

Limited access is the same constraint seen from outside

Google released Gemini 4 Carbon, which it says retakes the top spot on updated and harder benchmarks with gains in cybersecurity and complex software engineering, at the same introductory token prices as its previous flagship Argon, though access is limited and it does not yet replace a generally available model, VentureBeat reported. The benchmark claim is the vendor's and can wait for independent runs.

The launch shape is the part worth reading. A better model at unchanged introductory prices, restricted access, and no promotion over the generally available flagship is what rationed serving capacity looks like from the customer side. It comes from the company that also spent $3.73 billion less than expected on building that capacity out.

What settles it

• The 2027 capex guidance both companies pointed at, when it is given as a number rather than a direction.
• Whether the capacity Microsoft moved into the first half of 2027 lands in that half.
• Whether Gemini 4 Carbon becomes the generally available model, and when access opens.

Until then the ledger records what was said: a $5 billion forecast cut attributed to chip supply, a $3.73 billion consensus miss with next year flagged higher, cloud revenue up in both houses, and $2 trillion of index value gone in a month.