Microsoft Earnings Analysis: Hide Me The Money ⭐

With its latest quarterly report, Microsoft gave up all pretense of explaining where the money is going. This should scare everyone.

A quick recap: Yesterday, Microsoft reported that it earned a net income of $35.8 billion on revenues of $90 billion in the quarter ending June 30, 2026. Those figures are double-digit gains on the same quarter a year ago, but this quarter also marked the end of Microsoft’s fiscal year 2026. And it set all kinds of records, both good and bad.

In the good column, Microsoft earned revenues of $133.7 billion, up 31 percent from the $101.8 billion it earned the previous fiscal year.

But Microsoft also spent an astonishing $145.3 billion on capital expenditures (CAPEX) related to its AI infrastructure buildout in that fiscal year, also a record, of course. But the real miracle here is that Microsoft found a way to lower the perceived cost of this spending going forward. And you are never going to believe what happens next.

The seven stages of reinventing accounting

Microsoft has spent the past two decades or more steadily reducing the transparency of its financial reports, obscuring the data that potential and existing shareholders need, and ignoring its disclosure requirements. The strategy has worked: Microsoft talks in ever vaguer numbers, Wall Street analysts who should have the backs of us, all investors, and the economy play along but do not ask the difficult questions that are their literal responsibility, and it just gets worse every quarter. Microsoft isn’t alone in doing this, I know. But I cover Microsoft more closely than others, and I see this clearly and have for a long time.

In the past several quarters, I dug into each question these financial analysts asked of Microsoft and relayed the astonishing suck-up nature of that relationship. But this time, I saw a curious thing in the questions, which came off as a sort of greatest hits of the questions from the past few years.

There was the question about Azure growth, the question about the volatility of hardware costs and Microsoft’s CAPEX spending. The question about Microsoft 365 Copilot growth, which in the suckiest suck-upedness I’ve seen in, oh, a quarter, was “obviously, very strong.” The question about cybersecurity. And yes, the question that should have been at the center of your Microsoft financial results Bingo card, about ROI. It was all there.

And it was mostly uneventful. But there were a few interesting remarks in there that led us inevitably to yet another accounting change, another shenanigan, that should have had Wall Street reeling–to be clear, in a bad way–but was instead met with a collegial and congratulatory “well done, Microsoft, well done.”

But first, a bit of insanity that leads directly to the financial shenanigans Microsoft slowly unveiled. When asked about all the hardware costs and Microsoft’s ever-expanding CAPEX spending, CFO Amy Hood answered first, and we’ll get to that. But then CEO Satya Nadella chimed in with, “All of us are reading this 1873 as the book to be read. And so, in my mind, I think you’ve got to get the product shape right.”

Yes, I had to look it up. 1873 is a book about the Rothschilds, the Great Depression, and the making of the modern world. So it’s like Nadella, and apparently the Microsoft senior leadership team, are basically doomscrolling historical precedent for the way this company and the rest of Big Tech and Big AI are going to tank the economy. It’s OK, everyone, we came out of the Great Depression just fine. All it took were two world wars, and then everything has gone great since then.

Amy. Please. Show me the money.

“You’ve seen our CapEx really pivot toward what I would call and do call short-lived assets, which really, that’s CPUs and GPUs that have relatively shorter lead times,” he started when asked about hardware and AI spending. “The investment into land and datacenter builds is actually quite flexible. It’s [meaning land and datacenters] a smaller percentage of the overall cost structure, and timing can be changed on much of that, especially on the builds, or you can stagger the timing of the buildout of, as I was saying, some of the GPUs and CPUs that you plan to put in. And so, when you think about being able to manage through that, hyperscalers have been doing that for quite a long time in terms of having the flexibility and the understanding of manage those changes in demand.”

Allow me to translate that into plain English.

What Hood is saying is that when it comes to the cost of building AI infrastructure, which is mostly datacenters and everything that goes into that, there are three types of costs: Short-lived assets in the form of chips, longer-lived assets in the form of other components, and then real estate, which is a whole new ballgame for Microsoft, at least at this scale. The math of these things has to do with depreciation and how Microsoft pays for–or spreads out–the cost of components over time, and then the more traditional financials (though not for Microsoft) of land and property ownership.

This is important to understand because she later expanded on how Microsoft is shifting its accounting to spread out the costs of this massive build-out ever further into the future in at least two big ways.

“Quite frankly, my math has changed in terms of how I [track return on investment, or ROI] over the past year,” she said later in the call. “I would say the way to think about it for me is more the confidence in the [total addressable market] expansion, the margin levers that we have in terms of both product improvements [and] the infrastructure improvements.”

What she didn’t say explicitly here is that Microsoft has for the second time in the past years doubled the time period over which it accounts for the cost of hardware components and other assets like land and property. That is, in the pre-AI era, Microsoft and other companies followed standard depreciation and accounting for the cost of physical things that needed to be replaced (like components).

But now it doesn’t do that. So while a short-term asset like a CPU might realistically be expected to last 4 years, it would suddenly be viable for 8 years. That’s outrageous enough. But that represents two-thirds of Microsoft’s CAPEX expenses each quarter. And Hood, an accountant and not a hardware engineer, literally made the claim previously that Microsoft found that components actually got more efficient the longer they used them. I equated this is using bald tires on a car because, after all, there will be less friction with the road and they will be more efficient. You know, until one explodes.

If 8 years is outrageous, what might one call a doubling of that number? Impossible is my choice. But that’s what Microsoft just did. Not because these CPUs and other components are actually going to last four times their expected lifetime. But because this accounting change means that Microsoft can spread the cost of acquiring these components out over a longer period of time and lower the perceived costs of its AI buildout each quarter going forward.

In the previous four calendar quarters, including this one, Microsoft spent $41 billion, $31.9 billion, $37.5 billion, and $34.9 billion. Given the above accounting change, one might think that these numbers will start going down in the next few quarters since Microsoft is accounting for the costs differently, by spreading out the costs over much longer time periods. An accounting trick, in other words.

But there’s more: Microsoft is also changing how it accounts for the land and property expenses. And it is doing so in a way that will take those costs off the CAPEX books, too.

“Effective at the start of FY27, we are extending the estimated useful lives of our datacenters and office buildings, from 15 to 25 years, reflecting our operating history and expected use of these assets,” Hood said. “The greater impact is on capital expenditures as more of our future datacenter leases will shift from finance leases to operating leases as a result of this update. Finance leases are included in capital expenditures while operating leases are not.”

So the cost of owning all that land and property will still be accounted for, somewhere. Just not in CAPEX, the number that Wall Street is rightfully following and is very concerned about.

Here’s the gut punch to these two changes: Microsoft will spend $50 billion on AI infrastructure CAPEX expenses in the current quarter. That’s roughly 25 percent more than in the previous quarter and over double the figure from the year-ago quarter. But it’s also much less than it’s really spending because now it’s not accounting for costs the same way and has removed some CAPEX costs entirely. I don’t know what the “real” number might have been, but I bet it’s somewhere in the $60 to $70 billion range. Basically, there have been a lot of questions about AI spending and Microsoft is resolving this issue by mortgaging the future in two new ways. It’s incredible.

Wall Street should be outraged by this magic accounting. But here’s how Reuters, not the most bombastic of publications described it: “Microsoft gave a capital expenditure forecast below Wall Street estimates after an accounting change for data center leases.” Microsoft’s stock price is up almost 16 percent as I write this because, to Wall Street, this math somehow makes sense. But Microsoft still expects to spend $175 billion on CAPEX this year, even after those changes.

Moving on.

Three quick Google-related bits emerged here, too.

When Google announced its most recent quarterly earnings last week, it was notable for two points: The company spent $45 billion on AI infrastructure/CAPEX in the quarter and raised its expected yearly spending to $205 billion, and it was cash flow negative for the first time in its history because of those costs. As incredible, $99 billion of Google’s revenues came from gains in its investments in other speculative AI companies like Anthropic and SpaceX. It’s making money by betting on its competitors.

Microsoft addressed the first of those two issues with its own business using the accounting shenanigans noted above. And Amy Hood thought to say out loud that Microsoft expects “to remain free cash flow positive in FY27.” I think that’s the first time anyone from the company has even considered the alternative, so this must have been driven by the Google news.

As for the investment bit, Microsoft lost $400 million on OpenAI–and will lose an expected $100 million in the current quarter–but its investment in Anthropic help offset that and deliver $2.8 billion in investment gains. In other words, Microsoft’s investment in Anthropic delivered $3.2 billion. Not quite the $99 billion that Google saw, but give them a few quarters. Amy Hood will figure it out.

Copilot, Microsoft 365 Copilot, and the Super App!

Laurent wrote about Nadella again mentioning the coming “Super App,” a consolidated Copilot client that mimics what we see happening at OpenAI and Anthropic. But for this quarter, Microsoft had to make do with Copilot, which no one seems to want or use on the consumer side, and Microsoft 365 Copilot, which is seeing strong growth in the commercial market because it’s starting out small and is still small.

From a shenanigans perspective, Copilot presents an interesting opportunity for Microsoft. Here’s how Satya Nadella explained that.

“If I think about Office historically compared to what Copilot is, [Office] is much more narrow,” he said. “This [Copilot] is the first time where you really have an enterprise-wide tool which has a both per-seat and usage-based pricing.”

Here’s the plain English version: Office was the first Microsoft business to deliver what we now call a subscription service pricing model where (commercial) customers have long paid Microsoft a monthly licensing fee for the product. This was true before Microsoft 365, and it’s true today. It tried to replicate that success by making Copilot an added per-month, per-seat cost and has seen just limited success. But by moving Copilot–Microsoft 365 Copilot, in this case–to usage-based billing, Microsoft has a potentially explosive new revenue stream and from the exact same customers.

I want to be outraged by this. But the thing is, this actually makes sense: Microsoft will only see explosive revenues from this usage-based billing system if customers actually use the service a lot. And for that to make sense, the service has to deliver a solid value. So this is an island of sanity in the sea of insanity that is these financial reports. We’ll see if it bears fruit.

In the meantime, not so much. Last quarter, Microsoft claimed over 20 million paid Microsoft 365 Copilot seats, which I pointed out was just 4.4 percent of the overall customer base of paid Microsoft 365 seats. But this quarter, Microsoft claims 30 million paid seats, and that’s a growth of 50 percent growth! Awesome, yes. But Microsoft didn’t provide a new number for overall paid Microsoft 365 seats, and if that’s unchanged then only 6.7 percent of that customer base is paying for Microsoft 365 Copilot. And that is less than awesome. ChatGPT has over 50 million paid subscribers and Claude has an estimated 18 to 30 million. And from an overall user base size, ChatGPT and Gemini bot have over 900 million users.

And while this isn’t directly related, Microsoft said that GitHub Copilot now has 50 million users. What it didn’t provide was a paid user number: Last quarter, that figure was 4.7 million, but Microsoft has since introduced usage-based billing, so I’m guessing growth there has slowed. And GitHub has over 225 million users, not paid, which includes over 90 percent of the Fortune 500.

XBOX

It’s not news that XBOX is struggling, and with a year or more before Microsoft even previews the next console, it’s going to be a difficult era of ocassional blockbusters–Halo: Campaign Evolved, whatever Call of Duty game, and so on–interspersed with hand-waving and minor but important updates like the expansion of Backward Compatibility on PC that don’t move the needle much financially.

The data on the previous quarter isn’t great.

XBOX content and services revenues declined 10 percent YOY, with Microsoft citing “strong first-party content performance” from the year-ago quarter, a curious assertion since I couldn’t find any interesting game releases from that quarter last year.

And XBOX hardware revenue, which one assumes has to hit zero at some point, continued the double-digit YOY losses with a revenue decline of 13 percent. It could be worse, the year-ago quarter saw a 32 percent decline in hardware revenues and the three quarters between the two saw losses of 33 percent, 32 percent, and 29 percent.

Oddly, the recent layoffs and studio spin-offs didn’t come up except in passing, with Hood noting “severance expense and impairment charges in XBOX” with no context and that “total company headcount declined 2 percent year-over-year” in the quarter. Nadella likewise barely referenced Asha Sharma’s plan to turn the business around. Instead, he just said that “we are making the necessary decisions required across our content portfolio, platform, and operations to reset the business for long-term growth” and that he expects XBOX “to return to growth in fiscal 2027.” Which means between now and the end of next June, by the way.

Hood said nothing about XBOX beyond noting that Microsoft expects XBOX content and services revenue to decline in “the mid-single digits” in the current quarter while hardware revenues will just “decline” year-over-year. So I guess that growth thing isn’t happening anytime soon.

Windows

As Microsoft predicted a quarter ago, and as is obvious given the state of the industry these days, Windows revenues from PC makers declined 7 percent YOY, with Microsoft citing lower PC market demand and the red herring that is the Windows 10 end-of-support winddown that it just extended a year without telling anyone.

The component crisis might seem like a net negative for Windows, since PC sales inevitably decline when prices rise so sharply, and when you look at that 7 percent decline, you might think that it makes sense. But no. This story is bit more nuanced.

Microsoft earns most of its Windows revenues from businesses on a very steady basis that’s not impacted by the component crisis, though that’s reported as part of Microsoft 365 over in Productivity and Business Processes. And it makes the rest of its Windows revenues from PC makers, which, yes, are impacted by the crisis but with a big caveat.

“[Windows revenues from PC makers] were ahead of expectations as OEM and channel partners continued to build inventory given increasing component prices,” Amy Hood said during the post-earnings call. That is, revenues from PC makers should have been worse, but because prices are only going up on the components PC makers need, they are buying more than the demand requires because they want to lock in today’s prices. A silver lining for Microsoft, I guess, but also a nice reminder that this decades-old business model is still paying off.

Satya Nadella addressed the “2026 pain points” work Microsoft is doing with Windows, and in doing so, he confirmed my theory that this was less about customer complaints and more about the need to bring AI agents and PC-based AI orchestration to the world and that it needs a stable and reliable platform for that to work.

“In Windows, we are investing to ensure that it has the best quality and fundamentals, while also ensuring it is the best place to run secure edge AI,” he said. “We see significant opportunity for Windows to become the offload for unmetered intelligence, combining powerful on device compute with enterprise-grade security.”

Right.

Looking to the current quarter, Windows revenues from PC makers “will be impacted by lower PC market demand as higher component costs increase device pricing, a prior-year comparable that benefited from Windows 10 end-of-support, and elevated inventory levels,” Hood said. Which is hilarious when you think about it: PC makers overbought Windows licenses last quarter, if you will, but now that they’re stockpiled, they will underbuy this quarter. Maybe my business model comments are a bit off.

Anyway, revenues will decline in “the low twenties,” which is really bad. Not just worse than 7 percent, but triple that. “The range of potential outcomes remains wider than normal,” Hood added. Ominously.

There is one more thing

Last quarter, I discussed Bing and Microsoft Edge in a meaningful way in a quarterly analysis for the first time ever because Microsoft discussed these products in a meaningful way in a quarterly report for the first time ever. This came up briefly in this quarter’s post-earnings call, and a Nadella comment last night explains something that confused me three months ago. I think.

“In search and advertising, Bing and Edge have both taken share for five straight years,” Nadella said, echoing the same vague comment he made last month. But then it dawned on me. When he says Bing and Edge there, he means Bing and Edge in the context of search and advertising. That is, revenues from search and advertising are delivered in large part through Bing and Edge because those services incessantly put users in front of Microsoft advertising.

Put another way, “Edge growth” isn’t growth in the Edge userbase. It’s revenue growth from ads. That said, Nadella did state that Edge “has taken share for 20 consecutive quarters” three months ago, a claim that was not repeated this quarter. But even 0.001 percent growth is growth. I think the value of Edge to Microsoft is really just about ads, and that’s where all the enshittification in that product comes from, obviously.