Amy Hood had a sentence ready for the question every analyst asks now. “We still have demand that exceeds supply,” Microsoft’s finance chief said on the call, after a quarter in which the company’s cloud business grew faster than Microsoft itself had forecast.
The numbers were strong almost everywhere. Revenue rose 19% to $92.28 billion in the quarter ended September 30, against $90.69 billion expected by analysts surveyed by LSEG. Operating income climbed 21% to $45.92 billion. Adjusted earnings of $4.89 a share beat the $4.73 consensus. Shares rose 4% after the close.
Azure did the heavy lifting. Revenue in the cloud unit rose 48% at constant currency to $33.2 billion, ahead of the 44% to 45% Microsoft guided to in July and well ahead of the 45.1% analysts expected. It was the first quarter in which Microsoft disclosed Azure in dollars rather than as a growth rate, and the figure makes it the company’s largest single business. Microsoft Cloud, the broader bundle, brought in $63.9 billion, up 30%.
Then came the line that moved other stocks. Capital spending including finance leases was $45.2 billion, up about 30% from a year earlier but roughly $5.5 billion below the $50.67 billion analysts had projected. Hood attributed the gap to third-party capacity that had been scheduled to arrive in the quarter and did not.
Read one way, that is a company pulling back. Read the way Microsoft described it, it is a delivery problem, not a decision. The company has been renting capacity from outside providers while its own buildings come online, and a late shipment from a partner shows up as underspending rather than as lost revenue.
The distinction matters because the market has begun to treat hyperscaler capex as a proxy for the health of the entire AI trade. Alphabet’s spending plans and Oracle’s debt load have both been repriced this month. When Microsoft spends less than expected, the first reaction is to ask who did not get paid.
Some of them are listed. CoreWeave and Nebius, two of the rented-capacity suppliers that have grown alongside Microsoft’s buildout, fell in after-hours trading on the read-through, along with IREN. Microsoft signed a contract with Nebius worth close to $20 billion that runs into the 2030s, and it is among CoreWeave’s largest customers.
The quality of the spending is changing as well. About 75% of Microsoft’s capital expenditure is now tied to short-lived assets such as Intel processors and Nvidia graphics chips, up from roughly 66% in the previous three quarters and about half a year ago. That mix shift favors revenue that arrives sooner and depreciates faster, and it tells you what management thinks the constraint is.
Cash generation held up better than at its peers. Free cash flow fell 27% to $18.63 billion, which was still more than double what analysts had modeled. Alphabet, reporting the same afternoon, produced negative free cash flow of $9.07 billion. The two companies are running similar-sized building programs and getting opposite results on the cash line, largely because Microsoft’s depreciation schedule and lease treatment differ from Google’s.
The quarter also carried a disclosure that reframes the OpenAI relationship. Microsoft booked $10 billion of revenue from its commercial arrangement with the AI lab, about 11% of total revenue. That is a large single counterparty for a company that sells to everyone, and it makes the terms of the partnership a line item investors will now track every three months.
Guidance pointed up. Hood said she is aiming at more than $50 billion of capital spending in the December quarter, with a further step up in the second half of the fiscal year as CPU capacity arrives. Microsoft told investors to expect $96 billion to $97.1 billion in revenue for the current quarter, above the $95.60 billion consensus, and Hood guided Azure to 49% to 50% growth at constant currency against 45.1% expected. For calendar 2027 she put capital spending at about $170 billion, with a step up in the second half of the fiscal year as CPU capacity, including more of Microsoft’s own silicon, comes online.
One complication sits inside the comparisons. On September 2, Microsoft narrowed what counts as Azure, moving GitHub’s cloud services and Security Copilot into a different reporting line. Prior-year figures were restated, so the 48% is clean on the new definition, but the business being measured is not quite the one investors were measuring a year ago.
What the quarter actually settled is narrower than the headline suggests. Azure demand is not the problem, and price is not the problem. The problem is physical: buildings, power, and the third-party capacity that Microsoft rents while it waits for both.
Hood’s answer to that is more money and less patience. The $170 billion figure for next year compares with $45.2 billion in the quarter just reported, and it assumes the supply chain delivers on schedule. The delayed third-party capacity is the test case. If it lands in the December quarter, the miss was timing. If it does not, Microsoft will be spending against demand it cannot serve, which is a more expensive problem than a light capex number.
