Microsoft Earnings Analysis: Pay No Attention to the Capex Behind the AI ⭐️

In this free-wheeling analysis of Microsoft’s latest earnings, I look at Windows, Edge, Bing, and Xbox. You know, all those businesses that Wall Street couldn’t not care less about. But don’t worry, this hits on AI too. There’s no way to avoid that.

To recap, Microsoft yesterday reported that it earned a net income of $31.8 billion on revenues of $82.9 billion in the quarter ending March 31, 2026. This was the third quarter of Microsoft’s fiscal year 2026, and if all you care about are big numbers, then the business is doing great overall. But if you care about the truth, what you see if you dig deeper is a lot more troubling.

Same old, same old, in other words.

In this era, AI is the only thing that matters to Microsoft, investors, and Wall Street. And we will get to that; it’s impossible to ignore. But first, I will focus on what you and I care about the most, the parts of Microsoft that sell software and services to human beings. That way, we can rip off the proverbial band-aid and get the worst news out of the way first.

Windows

If there’s a theme to this earnings report, it’s the equivalent of a magician waving their hands to take your attention away from the trick they’re performing, a classic diversion tactic. The audience, in this case, is Wall Street, the analysts and investors who implicitly control Microsoft’s power by determining through their actions how impressed they are by the business and its ability to grow. And if there is anything Wall Street wants to hear about from Microsoft less than Windows, it’s a Windows business in crisis. And so Microsoft waved its hands furiously when discussing Windows in the wake of the earnings announcement.

“Monthly active Windows [PCs] surpassed 1.6 billion,” Microsoft CEO Satya Nadella said during the post-earnings conference call, not pausing to note that this is the highest-ever figure that Microsoft has ever provided for Windows usage.

This statement came right after Nadella explained that his company was “doing the foundational work required to win back fans and strengthen engagement across Windows, Xbox, Bing, and Edge.” Here, he neglected to state why it needs to reverse at least some of the enshittification that has to some degree ruined each of these offerings. But the fact that Microsoft is “prioritizing quality and serving [its] core users better” in 2026 is an implicit acknowledgment that it was not doing this in the recent past.

That’s as close as we got to a mea culpa from Nadella during this call. He said that this work had begun in Windows with “performance improvements for lower memory devices, a streamlined Windows Update experience, and [a new] focus on the core features and fundamentals that matter most to our customers.” I apologize for hammering on this, but what the f$%k was this company doing before this? That’s rhetorical. It was doing this.

One of the obvious questions we have about Microsoft’s sudden focus on Windows quality is why. That is, what happened that caused Microsoft to suddenly change course from not just ignoring Windows at a high level, but from actively undermining its own customers?

My theory is that this is about the only customers who really matter to Microsoft, the enterprise, and that the quality of Windows, especially the bloated and buggy Windows Updates and the endless cascade of Copilot marketing and new features that few seem to want, has finally triggered pushback. But the Windows commentary during this earnings call focused almost entirely on consumers. Not for the first time, Microsoft seems to realize that its position in the important consumer market isn’t just small, but antagonistic. And that if it wants any kind of growth with this product line, it’s going to have to happen there. Microsoft already owns the Fortune 500, after all.

Well, that growth isn’t happening now. And as I will get to, it won’t happen anytime soon, either. Windows revenues from PC makers (and Surface, a business so small it doesn’t warrant an explicit mention) decreased by 2 percent year-over-year (YOY) in the quarter. This indicates that the component crisis over-buying that helped this business in previous quarters has come to an end. The revenue gains from the previous four quarters, in reverse order, were 1 percent, 18 percent, 3 percent, and 3 percent.

Since that 18 percent number was an outlier—and based on that over-buying noted above—one might make the case that growth in PC maker revenues has been in the plus or minus 3 percent range for the past year or so. So a 2 percent decrease isn’t all that bad. There is a component crisis, after all. But a throwaway line from Nadella exposes a darker problem here: He said that “[revenues from PC makers] increased slightly and were ahead of expectations as [PC makers] and channel partners continued to build inventory given increasing memory prices.” Microsoft earns money here when these customers buy Windows licenses, and they apparently overbought them in the quarter because of fears that component prices will be even higher and availability will be even worse in the future. In other words, a 2 percent decrease in revenues is a win, as this number would have been much worse otherwise.

That’s not just a theory: In the current quarter, which closes out Microsoft’s fiscal year on June 30, Microsoft expects Windows revenues from PC makers to “decline in the high teens.” So we’re going to go from a 2 percent decline in FY26Q3 to an 18 percent decline in Q4. Yikes. And yes, it’s an expected 18 percent: 6 points of the decline will come from an unfavorable YOY comparison from the end of life, sort of, for Windows, 6 points will come from inventory level declines and 6 points will come from “a lower PC market” because prices will go up so much customers will stop buying. Overall, “[Windows revenues from PC makers and Surface] should decline in the mid to high teens.”

Microsoft Edge

We don’t discuss Microsoft Edge much during these earnings announcements analyses because Microsoft’s web browser doesn’t generate direct revenues of any kind, and it’s rarely discussed. But Microsoft did discuss Edge during its Q3 earnings announcement. As noted above, Edge, like Windows, is part of Microsoft’s effort to “win back fans and strengthen engagement” with consumers.

“Our Edge browser has taken share for 20 consecutive quarters,” Mr. Nadella claimed.

Really?

How? Taken share from what browser? Chrome?

Bob, let’s roll the tape. According to Statcounter, Microsoft Edge has 5.79 percent usage share across desktop and mobile. And if we look just at the desktop, where Edge performs best because it’s bundled with Windows, the dominant desktop platform, Microsoft’s web browser has 12.75 percent usage share. I assume Nadella is referring to that latter data point, since these announcements are essentially marketing and you always present the best possible news.

So let’s look at the numbers. 20 quarters is 5 years, literally, and a web browser usage share chart of that time frame shows pretty clearly that Edge did not gain share every quarter during this period; just two quarters ago, Chrome had a nice upward spike and Edge was declining. More broadly, Chrome has had over 65 percent share during almost this entire time—it fell to 61/62 percent twice in 2023 before recovering—while Edge has never exceeded 14 percent. More to the point, Edge has had higher share several times during this period than it does right now.

I have to assume, then, that Nadella is referring to Edge share on Windows only and that he is using some internal metric to make this claim. But I will also point out that a product that allegedly gained share for 20 quarters in a row doesn’t need to “win back fans and strengthen engagement.” You only make that effort when you’re failing.

Bing

Microsoft does mention Bing here and there in its earnings announcement because this unloved product is the source of advertising revenues. This quarter, Bing is part of that Microsoft effort to “win back fans and strengthen engagement” with consumers. And as with Windows and Edge, we’re confronted by the contradictory point that Bing is somehow experiencing unprecedented success, again begging the question of why it needs fixing.

“Bing monthly active users reached one billion for the first time,” Mr. Nadella claimed. But this time, I think the claim is semi-accurate. It all depends on your definition of a “user.”

Yes, Bing’s share of the search market is even more dire than Edge’s share of the web browser market: Where Google Search has steadily maintained a roughly 90 percent share of web search during the same five year period noted above, Bing has grown from about 2.5 percent share to 5.13 percent. That’s 100 percent growth! But it’s also 5 percent of the market, which means two things: It’s in second place (by a wide margin), and it is an also-ran. Which explains all the antitrust interest in Google Search.

But what about those additional one billion users? Is that even possible?

According to an AI overview of a Google search—you’re welcome—Google processes over 13.7 billion search queries every day. There are an average of 3 to 4 queries per user, so the number of users is roughly 4 billion.

Given its 5 percent of the market, Microsoft cannot have one billion Bing users. But Nadella said monthly active users, MAUs, and that could mean all kinds of things. And if you want his Bing claim to make any sense at all, I think you need to include other Bing-related search activities related to ChatGPT, Copilot, and other AIs to that list. That is, much of this growth is really driven by the growth of AI chatbot usage.

Finding the truth of this matter is difficult. But consider this: In October 2025, Brave said that Brave Search had 100 million users across desktop and mobile and that the service was handling 1.6 billion queries each month and almost 20 billion per year. Brave Search is so small that it doesn’t even make the Statcounter search charts. The smallest player on that chart is Baidu, with a 0.53 percent share. So Bing usage could be over one billion somethings. I just don’t believe it’s human beings. I mean, come on. Who uses Bing, really?

Apologies. That’s the wrong question. And if you want to accuse me of pulling the magician hand-waving trick I mentioned up top, fair point. Though I wasn’t trying to deceive, I was simply speaking too generally.

The right question is this: Who uses Bing on purpose? And the answer to that question, statistically, is no one. What people do is use Bing without realizing it. They use Bing when they search from the Microsoft Edge address bar unless they explicitly find and change that setting in the browser. They use it when they use Copilot, on the web, on mobile, on Windows, or wherever else. Bing is used by some other search engines and by other chatbots. In Office, Microsoft 365, and Microsoft 365 Copilot. People also use Bing when they search in Windows 11, access the widgets in Windows 11, and elsewhere. There’s a lot of unintended Bing usage happening out there. One might argue that most Bing usage is unintended, but whatever. Any time Bing is used, Microsoft benefits from the virtuous cycle between it, MSN, and its advertising business.

And that Microsoft does discuss in its earnings, albeit briefly.

In Q3, Microsoft said that “Search advertising revenue ex-TAC [meaning, minus the cost of paying affiliates to direct traffic to Bing] increased 12 percent, with growth driven by higher volume and revenue per search across Edge and Bing. This is one of those profits vs. revenues things, but remember that this is revenue and that it excludes the cost of doing business. So it’s another cherry-picked data point that highlights only the positive, excluding the truth.

Here’s where the bad news that Microsoft can’t avoid starts: Ex-TAC Search advertising growth in the previous four consecutive quarters was 21 percent, 21 percent, 16 percent, and 10 percent. And it will fall to the “high-single digits” in the current quarter for whatever reason, Microsoft noted. So it’s on a downward trend. And maybe that explains why the sudden effort to “win back fans and strengthen engagement” with consumers is happening with Bing.

Then again. Who cares?

Xbox

There’s one more consumer product that Microsoft wants to fix, and it’s a big one: Xbox has been on a downward trajectory for what feels like forever, and though its 2023 acquisition of Activision Blizzard made it one of the very biggest game publishers on earth, it’s done nothing to help turn around this business. This year, we’ve seen what seemed like an alarming leadership transition turn into a wonderful and hopefully not temporary new era of transparency and engagement. But this new leadership hasn’t had time yet to right this ship. And that shows in all the numbers.

Xbox content and services revenues declined 5 percent YOY in Q3, with Microsoft blaming “a prior year comparable that benefitted from strong first-party content performance.” And Xbox hardware revenue fell an incredible 33 percent YOY, which feels impossibly bad.

“The team is recommitting to our core fans and players, and shaping the future of play,” Mr. Nadella told investors. “Last week’s Game Pass changes are one example of how we are staying responsive to customer feedback.” I’m not sure about the “staying” part there; I’d argue this is more of a “becoming more” responsive thing. But OK, fair enough, it’s a positive change for most subscribers and for Microsoft, which will no longer undermine Call of Duty sales each year with Game Pass Day One inclusion.

But as with Windows, Nadella did the hand-waving thing by making a positive claim about this business that is, by all accounts, circling the drain.

“We set new records for monthly Xbox active users in the quarter, as well as game streaming hours,” he said.

That’s neat. But here’s the thing.

Xbox hardware revenues declined by 32 percent, 29 percent, 32 percent, and 6 percent in the previous four quarters (in reverse order). This quarter was the second in a row in which Xbox content and services declined, with 5 percent in the previous quarter following gains of 13 percent and 8 percent in the previous two quarters (again, in reverse order). And where Microsoft said that overall gaming revenue increased by 5 percent and 10 percent, respectively, in the first two quarters of this fiscal year, it neglected to specify a growth/decline for this and the previous quarter. In other words, overall gaming revenue is on the decline. It’s not just unprofitable; the revenues it brings in are going down. Hence the comment about a drain being circled. This is not sustainable.

As bad, the bloodletting will continue. In the current quarter, Microsoft expects Xbox content and services revenues to decline “in the low teens” (11 to 15 percent), and here it blames not just “a prior year comparable that benefited from strong first-party content,” but also its own decision to the lower the price of Xbox Game Pass. More ominously, Xbox hardware revenues will “decline” YOY. They’re not even guessing by how much. For the love of God.

AI

At the end of every post-earnings conference call, Microsoft allows analysts to ask it questions. Not a single question (or answer) included the words Windows, Edge, Bing, or Xbox, nor did anyone really ask about Microsoft 365, Microsoft’s biggest business, at least the core offerings. No, Wall Street doesn’t care about proven money makers or well-established—dare I say legacy—businesses like them. It cares only about the future and about the potential for growth. And so all the questions, all 7 of them, were about AI.

Which, to be fair, is somewhat understandable. Microsoft said this quarter that it spent $31.9 billion on AI infrastructure-related capital expenditures, a mammoth 49 percent increase over the $21.4 billion it spent in the year-ago quarter. That’s lower than the $37.5 billion it spent in the previous sequential quarter, but Microsoft had already explained this would happen at that time, and said the dip didn’t indicate a drop in demand—I’d call it a return to sanity—but just a coincidence of timing related to real estate leases and hardware buildouts.

At that time, I noted that if Microsoft’s AI spending remained flat, it would spend $145 billion on AI infrastructure in calendar year 2026, about double the spend of the previous year. But that wouldn’t happen, I wrote, and I expected Microsoft to spend roughly $180 billion to $185 billion this year on this frivolous “but everyone else is doing it” expense.

I was close, but it will be a bit more than that.

“We expect CapEx spend to increase to over $40 billion [in the current quarter] as we continue to bring more capacity online,” Microsoft CFO Amy Hood said during the call. “For calendar year 2026, we expect to invest roughly $190 billion in capital expenditures.”

Hood said that $25 billion of that is tied to higher component pricing. But no worries: She “remains confident in the return on these investments given higher demand signals and increasing product usage as well as the efficiencies we’re already driving across the platform.”

OK. Let’s get to the Q&A. This is Wall Street’s opportunity to show that its collective eye is on the ball and they’re going to challenge all Microsoft’s pretend numbers—like those of Microsoft Cloud, a business that does not exist—and force them to explain the insane AI spending. (That analysts later used the term commercial cloud, which is outdated, is interesting. As is Microsoft executives repeating back that term.)

This time, I will ignore the subservient flailing that each analyst provided at the start of each question–”Congratulations on another really solid quarter,” etc.—to focus instead on the meat of their valid concerns. Which in this case was:

“Those Microsoft 365 Copilot numbers are super impressive.”

Goddamn it.

Right, those “super impressive” Microsoft 365 Copilot numbers. Microsoft mentioned the term “Copilot” over 40 times during its prepared remarks, most of which was nonsense. But it said earlier in the call that “it was another record quarter for Microsoft 365 seat adds,” with an increase of 250 percent YOY, its fastest growth yet. “We now have over 20 million Microsoft 365 Copilot paid seats.”

20 million is a big number if you ignore context. But compared to ChatGPT, which has over 50 million paid users, 9 million of whom are Microsoft’s business customers, it’s lagging behind. (And not-paying customers see ads, which also generate revenues.) Google said in its earnings report, issued yesterday like Microsoft’s, that it now has over 350 million “paid subscriptions,” though most of that is likely YouTube- and not AI-related. (It did say that Gemini Enterprise “has great momentum with 40 percent quarter-on-quarter growth in paid monthly active users” and that it was “the strongest quarter ever for [its] consumer AI plans, driven by the Gemini App.”) And there are no reliable numbers for paid Anthropic Claude users.

So yeah, I guess 20 million is pretty good, if you ignore that this is still a tiny percentage of the total Microsoft 365 user base, most of which are businesses. How tiny? Even Microsoft Edge and Bing have better usage numbers in their respective markets; that’s how tiny. You may recall that in the previous quarter, Microsoft said that it had 15 million paid Microsoft 365 Copilot users and 450 million Microsoft 365 seats, so less than 3.5 percent were paying extra for Microsoft 365 Copilot. This quarter, that figure is 4.4 percent. So, yes. Super impressive.

Anyway. To the actual first question, which basically amounts to what the real demand is here and, more specifically, “How does [Microsoft’s AI spending] get paid for?” In other words, the same question everyone asked last month. What’s funny, however, is that this analyst asked Nadella to answer. So he responded with this:

“You want to start, Amy?”

Classic.

While Nadella scrambled to check his notes, Hood said that Microsoft 365 has historically been a per-seat revenue model, but that with the rise of AI, it was evolving into “a worker plus an [AI] agent,” or “a license business plus a consumption business.” Over time, the revenues this business generates will still include seats (workers), but it will also “have a meter, just like you see in Azure. And it may not all flow through bookings in the same way. You’ll just bill for usage.”

This is the pricing model that GitHub Copilot is moving to, and it will result in much higher monthly costs for the most active users. But it’s not clear how this will help Microsoft 365 Copilot: At the current price, an additional $20 per month per user, only 4.4 percent of the user base has elected to join. What will this look like when the prices are higher?

Nadella confirmed that Microsoft will move to usage pricing for its AI offerings across productivity and security, as it is with coding (GitHub). And the idea, stated plainly, is to charge the most active customers a lot more than they’re being charged now, with the hope that they see the value and, not stated, that Microsoft’s customers will do the same, lessening the impact of switchers. This is particularly important because those AI agents that are going to eventually change the world by working on our behalf in the background are going to dramatically increase AI usage. The other side of the sword is the automation of a product that’s quite expensive to operate. (Hence the $190 billion spend this year, which doesn’t include the cost of running that business, by the way.)

“This is more about getting intense users and intense usage, and that’s what we’re focused on,” he said.

The second question was, you guessed it, also about capex spending and how Microsoft can be so confident that this crazy plan can work. So, yes, just the same question again and again.

Amy, predictably, feels “quite good” about Microsoft’s strategy and how it will map to reality. In the second half of this calendar year and beyond, she believes that Microsoft will “put pressure on efficiencies, speed up the deliveries [of components] into [Microsoft’s] datacenters and make that what [she] would call revenue ready as quickly as we can.” There will be continued pressure on “first-party usage” (Copilot) and Azure demand (third-party usage). But she feels good about things.

“Terrific,” the analyst responded to this non-answer.

The third question was a nice surprise because, just kidding, the guy asked about the cost of AI again. He noted that Microsoft and Google both showed “higher margins,” meaning the cost of running this business is eating into profits, and he was curious what he and other investors were missing.

Hood claimed that Microsoft’s margins on AI were actually better than they were with cloud computing at a similar point in that ramp-up 15-ish years ago. And she said that the “value” Microsoft could capture here was tied to “consumption and usage-based pricing models,” which I will point out is exactly the Azure cloud computing business model. What she didn’t say, explicitly, is that Microsoft is finally going to start charging customers the true cost of the AI services they consume, instead of subsidizing it as they have been. What no one knows, including Microsoft, is whether that will be sustainable. Meaning, it will find customers willing to pay a lot of money for the value received.

To put this business over the top, Hood says that Microsoft will “leverage the IP” it has, meaning its rights to use OpenAI models through 2032, make the hardware it uses in datacenters more efficient by “taking the margins out of the infra[structure] stack,” and get as much capacity into those datacenters as it can. “when you move to usage-based models, you have to make sure you’re delivering incredibly high value to customers,” she added, which more than kind of ignores the fact that most find Copilot to be terrible.

Put more simply, you focus on the customer, create value, and reap the rewards. “The ROI will be very good.”

Hilarious.

Question number four: This guy really turned the conversation on its head by focusing on capex spending. Yes. Again. “Obviously,” he said, commercial cloud is growing really fast, Azure is growing really fast, and AI is growing even faster, but there’s a “disconnect” between the rise of capex spending and the revenues. This, he added, was making investors “a bit nervous.”

Last quarter, Amy was aggressively dismissive of this kind of mamby-pamby thinking, but this time she took a more conciliatory tack. She compared the AI ramp-up to the cloud computing ramp-up and pointed at the “acceleration” that Microsoft started seeing last quarter in commercial cloud, which, again, does not exist.

“I think the thing that investors have been asking, and [what] you’re asking about, is when we’ll start to see that show up in revenue growth,” she said as if that wasn’t obvious. “And I think that’s the first place you point to. We can also point to it, and I think you’ll start to see it in GitHub, where you see revenue growth rates and usage consumption models result in acceleration in the top line.”

GitHub is the tiny freaking business in the Microsoft empire, and its revenues wouldn’t impact Microsoft’s financial results in a meaningful way if they were simply expunged. To hold that business up as a model during a time in which it is transitioning to the new usage-based billing system that may make or break this entire product is likewise astonishing. Put another way, it’s not clear that the new GitHub Copilot bill system will work—it may literally drive customers away—and this business is so small, it’s unclear how its presumed success can work elsewhere, especially given that Microsoft 365 customers have shown nothing but an unwillingness to pay extra for AI.

Hood does make an astonishing admission here. She says that Microsoft has over $600 billion in revenues that it still needs to deliver, presumably to make its AI infrastructure spending make any semblance of sense. “I do feel very good about it, frankly,” she added, just in case anyone was worried that can never happen.

So say we all.

Question number five, finally, didn’t directly reference capex spending but instead concerned Nadella’s “reflections” on Copilot and what he’s learned about customer adoption to date. But this is still implicitly about ROI, right? What’s working and what’s not working.

Nadella talked about AI “form factors,” which include Chat, Chat with reasoning, agents like Researcher and Analyst, plus custom agents, and now Edit Mode and Cowork. These, to me, are all one thing, an evolution of the AI chatbot. But … whatever. “The usage of this [not clear on what “this” is] is at the same level of Outlook,” he said. “This is not, even to the previous question, are people using it, finding it useful? I mean, this is a daily habit of intense usage.”

He also discussed intelligence, meaning multiple models coupled with context (meeetings, documents, email, and other data). And he, too, compared this work in Microsoft 365 to what Microsoft is doing with GitHub Copilot. “We are seeing that all play out,” he concluded as if he had said anything.

The sixth question asked about the OpenAI agreement, which, to any outside party, looked like a lopsided win for OpenAI, given its Amazon AWS announcement less than a day later. But Mr. Nadella said that he feels good about the Microsoft partnership with OpenAI. And that he loves win-wins in partnerships. But he also positioned this partnership change as a win for Microsoft.

“We have a frontier model [meaning OpenAI’s], royalty free, with all the IP rights that we will have access to all the way to [20]32, and we fully plan to exploit it,” he said. “The second part, of course, is them as a customer of ours. They’re a large customer of ours, not just on the AI accelerator side, but also on all the other compute side. We want to serve them well. And then, of course, we have our equity … I feel very good about where we are.”

And then we finally arrive at the last question. Which brings us full cricle back to the capex spending and the consumption-based pricing that Microsoft has not even formally announced.

This time, Nadella didn’t immediately defer to Amy Hood, which is refreshing. Instead, he expanded on the never formally announced consumptin pricing by noting that Microsoft 365 will continue to have seats, those seats will have some “base usage rights” for AI built-in, and that customers who needed/used more AI would pay for “consumption packs.” And those customers who obtain long-term consumption commitments will of course get discounts on pricing.

“But how will you make money?” someone asks in the classic SNL skit Change Bank. “Volume.”