Microsoft Earnings Analysis: You Can’t Spell ROI With AI ⭐

Microsoft delivered another blockbuster financial quarter, but investors remain unconvinced about the return on investment (ROI) on Microsoft’s escalating AI spending. That’s pretty much all they asked about in the post-earnings conference call.

In case you missed it, Microsoft reported a net income of $38.5 billion on revenues of $81.3 billion in the quarter ending December 31 yesterday. Those figures represent gains of 60 percent and 17 percent, respectively, year-over-year (YOY), which is rather incredible for a company of this size and maturity.

Microsoft has three core business units, but it insists on touting its imaginary Microsoft Cloud “business” as some kind of a metric, with CEO Satya Nadella noting that it surpassed $50 billion in revenue for the first time in Microsoft’s second fiscal quarter of 2026. Which is as relevant as saying that your imaginary friend just won a race against real people.

Microsoft’s three real business units broke down like so in the quarter:

Productivity and Business Processes. This is Microsoft 365, so Office plus Windows commercial. Revenues increased 16 percent YOY to $34.1 billion.

Intelligent Cloud. This is Azure. Revenues were up 29 percent YOY to $32.9 billion.

More Personal Computing. This is Windows consumer (mostly from PC makers) and Xbox. Revenues declined 3 percent YOY to $14.3 billion.

Microsoft’s post-earnings conference call provided some further details, including a big Windows 11 milestone and a few other hard numbers. But the bulk of it was Microsoft executives, primarily CFO Amy Hood, the person who really runs this company, trying unsuccessfully to calm investor fears that Microsoft’s AI spending is running off the rails.

Let’s dive in.

? Microsoft’s massive AI spend

In its quarterly results, Microsoft noted that its capital expenditures in the quarter—which translates directly to the cost of its AI infrastructure build-out—were $37.5 billion. This massive number is 66 percent higher than the $15.8 billion it spent in the same quarter one year ago. But that’s no surprise. After all, Microsoft had been escalating this spend for two years, and it at one point pledged to spend $80 billion in fiscal 2025 on AI infrastructure.

Here’s the problem. The number keeps going up. Microsoft actually spent $84.6 billion on AI in fiscal 2025, and it’s now spent $72.4 billion on AI in the first two quarters of fiscal 2026 alone. If that spending just stays flat, Microsoft will spend $145 billion on AI in 2026, almost double the spend in 2025. But it won’t stay flat over the year, despite a quarter-over-quarter decrease this quarter “due to the normal variability from cloud infrastructure buildouts and the timing of delivery of finance leases.” It will keep going up.

And as expected, analysts asked about the cost of AI in every way imaginable during the post-earnings conference call. Also as expected, Microsoft aggressively defended the spending while trying to get those analysts to think differently about how it is managing these enormous sums of money.

The discussion spanned every single question in the Q&A segment of the call, but it basically boils down to this: How can Microsoft justify the rising costs of this AI buildout when Azure revenues have slowed and there’s no possible return on investment (ROI) in sight?

Amy Hood’s response to that was about as terse as it could be.

“The risk isn’t there,” she said.

But what about the fact that you are extending the average lifespan of the hardware you buy beyond the norms you previously established?

This was, she said, “long-term decision-making.” About “investing in all the layers of the stack.” And besides, the hardware is “already already sold for the entirety of their useful life.”

What?

Hood said that Microsoft has commitments to utilize the hardware it’s already purchased and for the entire duration of its lifetime.

Um. Right. With the understanding that I am not a financial expert, that doesn’t add up.

Microsoft is spending actual money on hardware: Capital expenditures, which we can safely assume are roughly 100 percent tied to AI infrastructure buildout, were $37.5 billion, a gain of 66 percent YOY and up from the $34.9 billion it spent in the previous quarter.

But Microsoft also has financial commitments from customers to use that hardware. Commitments are not actual money. These are promises, like an item on a wish list. And so the risk, to return to literally the first question Microsoft was asked about, is real. Microsoft is betting that its customers will actually spend that money. But you can’t book a promise.

And the costs keep growing.

As you may recall, Microsoft had said in its previous fiscal year that it would spend about $80 billion on this buildout and then exceeded that amount. At this pace, Microsoft is on track to spend $140 billion or more on AI in the current fiscal year. Woof.

In the good news department, Microsoft’s cap-ex spending in the quarter was less than its profits. That wasn’t the case in the previous two quarters, and I suspect it was a one-off. But that’s what it reported. Perhaps another number is more relevant though: Microsoft’s cash flow in the quarter was $35.8 billion. And that number is lower than Microsoft’s AI spending. Or not. Free cash flow, meaning the cash flow after capital expenditures, was $5.9 billion. Ah, numbers.

Moving on.

Mr. Nadella made an odd claim right at the start of his prepared comments opening the call.

“We have built an AI business that is larger than some of our biggest franchises that took decades to build,” he said, obviously referencing Windows, Office, and Server. And then he never explained how he calculated the size of this business, which Microsoft never admits loses money overall every single quarter.

What Nadella did provide was a series of platitudes that amount to nothing material, vague information about things like “optimizing tokens per watt per dollar, which comes down to increasing utilization and decreasing TCO using silicon, systems, and software.” Um. Yes. A few other tidbits, all thrown out without any context:

• “You can think of agents as the new apps,” he said, with no further discussion or explanation beyond the vague expectation that there needs to be a store of some kind of agents.
• Microsoft offers “the broadest selection of models of any hyperscaler,” which includes GPT-5.2 and Claude 4.5. But left unsaid, this is in part because Copilot is a massive failure, Microsoft’s first-party models are nascent, and customers demand these other more powerful and/or more popular models.
• Or, as Nadella claims, Copilot is “gaining momentum.” Daily users of the Copilot app (not clear which one) increased nearly three times year-over-year. And “Copilot seat adds” were up 160 percent YOY. What happens when you multiply by zero again?
• Over 80 percent of the Fortune 500 have active agents built using Copilot Studio and Agent Builder, Microsoft’s low-code/no-code tools.
• “AI experiences are intent-driven and are beginning to work at task-scope.” That’s a fascinating admission in some ways, given that Microsoft started talking up AI agents in 2024.

There were a few Copilot hard numbers.

• There are now 15 million paid Microsoft 365 Copilot seats. That’s a small percentage (less than 3.5 percent) of the overall 450 million Microsoft 365 seats. But also just a small number. (We don’t know how many people pay for ChatGPT, but estimates range from 10 million to 35 million out of 800 million users overall.)
• There are now over 4.7 million paid GitHub Copilot subscribers, up 75 YOY. This is unrelated to other Copilot-branded offerings, and what we might think of as the one “good” Copilot.

Microsoft’s CFO offered a few details about the financial side of AI ahead of Q2. She reminded listeners that Microsoft accounts for its investment in OpenAI under the equity method.

“As a result of OpenAI’s recapitalization, we now record gains or losses based on our share of the change in their net assets on their balance sheet as opposed to our share of their operating profits or losses from their income statement,” she noted. “Therefore, we recorded a gain [that] drove other income and expense to $10 billion in our results. When adjusted for the OpenAI impact, other income and expense was slightly negative, and lower than expected, driven by net losses on investments.”

Two-thirds of Microsoft’s capex spending of $37.5 billion, or $29.9 billion, was tied to short-lived assets like CPUs and GPUs. Because demand exceeds supply (her words), Microsoft is trying to balance the competing needs of Azure, first-party AI solutions, and R&D. The other third of the capex spending is for longer-term assets that will provide value for “the next 15 years and beyond.” These are financial leases and other real estate, basically the datacenters themselves. Microsoft spent $29.9 billion on property, plant and equipment costs in cash and $6.7 billion on finance leases.

? Productivity and Business Processes

A few points here, as we didn’t get a few relevant numbers, like Microsoft 365 consumer subscriptions. But paid Microsoft 365 commercial seats grew 6 percent year-over-year to over 450 million, Hood said on the call.

We were told that Microsoft 365 commercial cloud revenues were up 17 percent YOY, while seats grew 6 percent driven by small and medium business and frontline worker offerings. And Microsoft 365 commercial products revenue was up 13 percent YOY driven by Windows commercial and Office 2024 (!). That last bit is very interesting to me because it suggests a small part of the user base is resisting AI and all the churn. Hood referenced this later, noting that “Office 2024 transactional purchasing” was higher-than-expected and she’s not sure if that will “normalize” (go back down) in the current quarter. That’s something to watch out for.

Microsoft 365 consumer cloud revenues were up 29 percent, with subscriber growth up 6 percent (but again, with no subscriber count).

The total cost of revenue for this business unit was up 10 percent YOY thanks to it shouldering part of the AI buildout costs.

?️ Intelligent Cloud

Azure and other cloud services revenues were up 39 percent YOY. As you may recall, Microsoft actually provided a hard figure for Azure revenues in one quarter last year. But it’s not enough to calculate real world Azure revenues in this quarter. For that, we have to go forward a year from that previous announcement.

Server revenues were up 2 percent YOY. Enterprise and partner services revenues were up 8 percent YOY. And the cost of revenue was up 44 percent in the quarter. One wonders if that is tied in some way to the AI spending. Just kidding. Obviously.

? More Personal Computing

The part of Microsoft I care about the most is doing terribly. Overall revenue declined 3 percent, “driven by gaming,” Microsoft noted soberly. But the cost of revenue declined 8 percent thanks to declining hardware sales. They should consider ridding themselves of that boat anchor. Cough.

Windows

Windows reached a big milestone in the quarter, Mr. Nadella said up top: One billion Windows 11 users, up over 45 percent year-over-year. He also claimed that Windows, Edge, and Bing all experienced “share gains” this quarter.

But most of the news was bad, given the holiday quarter and the end of support (sort of) for Windows 10.

Windows revenues from PC makers and Surface were up just 1 percent YOY, its smallest gain all year: This business saw revenue gains of 6, 3, 3, and 4 percent in the previous four quarters sequentially. Microsoft curiously noted that the Windows 10 end of support milestone delivered 5 percent growth in revenues from PC makers (so, excluding Surface), but I consider that a bust. It’s just a likely that PC makers were buying out components ahead of expected price increases this year. And yes, Surface did indeed suffer from a “decline,” meaning a YOY revenue decline, but Microsoft didn’t give us a number.

Xbox and gaming

The news for Xbox and gaming was even worse, and dramatically so given the holiday quarter.

Xbox content and services revenue declined 5 percent YOY, which is bad. But Xbox hardware revenue declined an astonishing 32 percent YOY, and that is a disaster. I don’t understand how hardware can decline that much for so many quarters in a row. It has to bottom out at some point. Please?

Nadella said that Microsoft was “committed to delivering great games across Xbox, PC, cloud, and every other device,” and that it “saw record PC players, and paid streaming hours on Xbox.”

But “impairment charges in our gaming business” contributed to Microsoft’s overall operating expenses going up 5 percent YOY. Gaming revenue decreased 9 percent YOY. And Xbox content and services revenue decreased 5 percent, “below expectations [and] driven by first-party content with impact across the platform.”

I keep saying and writing this, but I feel like there is a solid business here and that that business is a game publisher, not a hardware maker. And that it’s time to make some tough decisions about hardware. There are milestones Microsoft needs to hit first, of course, like bringing the entire library of backward compatible Xbox titles to PC. But time is running out on Xbox as a business. The term Xbox ROG Ally didn’t come up once in the call or anywhere in Microsoft’s documentation for the quarter.

Looking ahead

The future of More Personal Computing will resemble its past. Which, yeah, is not great.

Hood said that Windows revenues from PC makers and Surface would decline YOY “in the low teens”.

“Growth rates will be impacted as the benefit from Windows 10 end of support normalizes and as elevated inventory levels come down through the quarter,” she said. “Therefore, Windows OEM [PC maker] revenue should decline roughly 10 percent. The range of potential outcomes remains wider than normal, in part due to the potential impact on the PC market from increased memory pricing.”

As for gaming, Xbox content and services revenues should decline “in the mid-single digits against a prior year comparable that benefited from strong content performance, partially offset by growth in Xbox Game Pass.” And, yes, hardware revenue should decline year-over-year. Again.