Microsoft Earnings Analysis: FY25 Q2 (Premium)

I focus almost exclusively on the shrinking client side of Microsoft’s business, which makes these quarterly analyses a bit difficult. But this time, I’m going to do something a bit different. Because there isn’t much going on with that side of the house, I will instead focus almost exclusively on AI. Given the seismic changes this past week, one might have expected some fireworks during Microsoft’s quarterly earnings conference call. And there were, given how dull these people are.

First, here are the top-line numbers.

Microsoft earned a net income of $24.1 billion on revenues of $69.6 billion in the quarter ending December 31, the second quarter of its fiscal 2025. Those figures represent gains of 10 percent and 12 percent year-over-year (YOY), respectively. It has three top-level business units, two of which are doing great:

• Productivity and Business Processes: $29.4 billion in revenues, up 14 percent YOY. This was 42.2 percent of Microsoft’s earnings.
• Intelligent Cloud: $25.5 billion in revenues, up 19 percent YOY. This was 36.6 percent of Microsoft’s earnings.
• More Personal Computing: $14.7 billion, unchanged from the year-ago quarter. (Literally flat YOY.) This was 21.1 percent of Microsoft’s earnings.

Looked at another way, the software giant attributed $40.9 billion of its $69.6 billion in revenues to the so-called “Microsoft Cloud,” which is not a business unit of any kind, but is instead a rough metric compiled from across the company. What it’s trying to communicate there is that 59 percent of Microsoft’s revenues come from the cloud, directly or indirectly. Growth in Microsoft Cloud, which, again, does not exist, was 21 percent YOY. Point being, Microsoft is, was, and will continue to be a cloud superpower. One that today is heavily investing in, and marketing, AI.

And this is where I usually diverge from the traditional analysis of Microsoft’s business: I could not care less about the cloud, especially the commercial cloud that dominates at Microsoft. That said, Microsoft’s AI efforts are interesting, given that the software giant kicked off the current AI era and is still very much in a leadership position. So I will look at this part of the business, at least. In fact, I will do so in great detail.

Let’s dive in.

? AI

AI has finally and belatedly reached the “show me the money” era, thanks to two factors: The toll of Microsoft’s ever-growing investments in infrastructure and the explosive and unexpected arrival of China-based DeepSeek this past week.

Microsoft’s cloud investments are well-known but I’ve long wondered how long its shareholders would sit by idly while the firm throws roughly $20 billion at AI infrastructure each quarter while seeing little if not nothing concrete by way of results. Threading this needle, CEO Satya Nadella and company have presented a rosy picture of a future that’s been upended by the (supposedly) low-cost alternative vision of this future delivered by DeepSeek.

I know DeepSeek is controversial, but I find it difficult to ignore the parallels here with Linux, which arrived over 30 years ago as an alternative to Unix. Linux never seriously threatened Windows on the desktop, despite the obvious fears, but it defeated Windows Server and is today the top workload on Microsoft’s Azure service. Is DeepSeek to Microsoft AI/OpenAI what Linux was/is to Windows Server? It’s worth considering. I assume Microsoft is working through that issue as we speak. But it’s important to know that with AI happening at such a rapid clip, we won’t need to wait 30 years to see if an open outsider can upset this market. It’s going to happen quickly.

In the meantime, Microsoft’s capital expenditures–virtually 100 percent of which are tied to AI infrastructure build-out–continue to rise. The company previous said that it would spend about $80 billion on this during its current fiscal year, so that works out to $20 billion per quarter on average. But it spent $20 billion in FY25 Q1 and now, this past quarter, that number jumped to $22.3 billion. (It was $19 billion in FY24 Q4.) At the current rate of growth, Microsoft will spend about $95 billion on AI infrastructure this fiscal year, which is a lot closer to $100 billion than it is to $80 billion.

With that and the DeepSeek announcement as context, the messaging is changing, as it must.

“We have been seeing significant efficiency gains in both training and inference for years now,” Mr. Nadella claimed early on in last night’s earnings conference call. “On inference, we have typically seen more than 2X price-performance gain for every hardware generation, and more than 10X for every model generation due to software optimizations. And, as AI becomes more efficient and accessible, we will see exponentially more demand. Therefore, much as we have done with the commercial cloud, we are focused on continuously scaling our fleet globally and maintaining the right balance across training and inference, as well as geo distribution. From now on, it is a more continuous cycle governed by both revenue growth and capability growth, thanks to the compounding effects of software-driven AI scaling laws and Moore’s law.”

This is a smart way to shut up the doubters: Thanks to the cloud, Microsoft grew from a has-been into a global superpower, and AI conveniently arrived right when that growth engine had finally cooled. With Microsoft making almost 60 percent of its revenues from the cloud today, Nadella is arguing that AI is how it can continue this success.

That said, the “continuous cycle” he mentions has thus far been negative, financially. The good news is that Microsoft can afford it, albeit barely. Microsoft’s cash flow from operations this quarter was $22.3 billion (and up over 18 percent YOY), the exact amount it spent building out its AI infrastructure. Coincidence? Honestly, yes. But this is an important metric to consider each quarter, and it’s free cash flow fell 29 percent YOY to just $6.5 billion, reflecting the cost of that AI build-out. (The difference comes out of net income, which is profits.)

Here’s another fascinating data point tied to this build-out: Microsoft has more than doubled its datacenter capacity in the last three years. With the understanding that not all of that additional capacity is AI-related, most of it is. And so one might argue that AI, is to Microsoft, roughly as big–or as important, or whatever–as all its other cloud-related businesses combined. That is rather astonishing.

This investment has to slow at some point. Indeed, that’s the next major milestone to look for, some coming quarter where Microsoft’s capital expenditures fall, quarter over quarter or YOY, or when the capex figure is once again below its cash flow from operations. It will happen, but timing is everything. And I’m curious if DeepSeek speeds up that process, because now investors–and, as important, analysts–are finally asking questions.

For example. “We really want to start to see a clear road map to what that monetization model looks like for all of the capital that’s been invested,” Zacks portfolio manager Brian Mulberry told Reuters.

I was so curious about how much push-back Microsoft was getting that I jumped forward to the end of the conference call transcript. There were 8 questions. And all of them were about AI (directly or indirectly) and, more to the point, the cost of AI.

Financial analysts are a curiosity to me and always have been. The first question is an example of why. It was about Azure growth (which was 31 percent, below the 32 percent analysts expected) being “a little bit disappointing,” which is insane to me: Azure is a mature and sizable business now and we’re long past the heady days of 70 percent-ish growth. And Microsoft’s response to this was curious. There are two sides to Azure revenues, CFO Amy Hood said. There’s Azure AI, which “were better than [Microsoft] thought” they were going to be. And then there’s the non-AI stuff, “pure cloud,” for lack of a better term, that she admitted were impacted by all the focus on AI. Microsoft is still constrained from a capacity perspective, thanks to AI, so 31 percent growth is amazing. And Microsoft feels the AI investment is justified by the “encouraging signals” it keeps mentioning.

“The AI growth rate is actually better than what we expected,” Mr. Nadella added. “The [traditional, non-AI] enterprise workloads [are] also in good shape. [With regards to balancing the old with the new,] you would rather win the new than just protect the past. And that’s one of the things that we definitely will lean into, always.”

With regard to actual revenues from AI, Nadella said that its divided between Azure AI and Copilot [Microsoft 365 mostly, I assume], and that the latter was “better” [bigger, financially, or more lucrative] than the former. Microsoft saw “strength” in seats [licenses] both new and additive, and usage, which he admitted doesn’t impact revenues, though I’d argue it does, negatively. “Price per seat was actually quite good,” he said, which is super vague. In short, the argument is that this is going better than Microsoft expected.

Inevitably, one analyst asked about DeepSeek, which Nadella had mentioned upfront in the call, and how AI scale at lower costs might impact Microsoft. Nadella downplayed DeepSeek, explaining that “what’s happening with AI is no different than what was happening with the regular compute cycle,” that it’s just “Moore’s Law that’s working and hyperdrive,” and … yeah, sure. I’m reminded of the time I confronted the Windows Mobile team in the wake of the iPhone and asked what they were doing to counter that threat. They told me that the iPhone proved what they had believed all along, that a consumer smartphone would be successful. To which I asked them, if this so obvious to you, why didn’t you do it first?

“I think DeepSeek has some real innovations,” Nadella said. “And that is some of the things that even OpenAI found in o1 [its latest AI model] … Obviously, that all gets commoditized, and it’s going to get broadly used. And the big beneficiaries of any software cycle like that is the customers, because at the end of the day, if you think about it, what was the big lesson learned from client server to cloud? More people bought servers, except it was called cloud.”

In other words, the advances DeepSeek unveiled are a tide that will float all ships, and in the AI space, there is no ship bigger than Microsoft. This is a theme Leo brought up on Windows Weekly yesterday: The Jevons effect dictates that when technological resources become more efficient, overall demand paradoxically increases so much that total resources combined still continue to rise. Point being, yes, Microsoft is building out AI infrastructure at an astonishing clip, and, yes, DeepSeek just showed a possible way forward for doing AI more efficiently, but this is a net gain for Microsoft, not a net negative. In the end, it will benefit from this shift.

Nadella said exactly that.

“When token prices fall, inference computing prices fall, that means people can consume more,” he said. “And there will be more apps written … That type of optimizations means AI will be much more ubiquitous. And so, therefore, for a hyperscaler like us, a PC platform provider like us, this is all good news as far as I’m concerned.”

The next question was about OpenAI and Stargate, and the ongoing weirdness of the Microsoft/OpenAI partnership.

“We remain very happy with the partnership with OpenAI,” Nadella began. Yes, of course. “And as you saw, they’ve committed in a big way to Azure. And even in the bookings, what we recognized is just the first tranche of it. And so, you’ll see, we’ve given the role for we have more benefits of that even into the future. And obviously, their success is our success because even all the other commercial arrangements that we detailed out … with that announcement.”

Well. That requires some explanation.

Tranche. I stumbled on that term immediately, and though I though I understand what he was saying, I was incorrect. Probably. “Tranche” means portion or part, as expected, but I believe he meant this in a financial sense. And in that context, it means more specifically a portion of a payment owed. That is, Microsoft has recognized (put on the books) only a portion of the money that OpenAI owes it. It will recognize more of that in future quarters. Neither company has admitted to this, but the ongoing belief is that OpenAI gives Microsoft 20 percent of its revenue. So whatever OpenAI makes on Stargate–using non-Microsoft infrastructure–benefits Microsoft financially (and also otherwise, since it relieves the Azure capacity constraints). “Their success is our success,” indeed.

But he says something very interesting here. In reiterating that “the fleet”–the AI infrastructure it’s building–is fungible, with the right balance between AI training and inference, and geo-distributed, he notes that Microsoft is “working super hard on optimizations.” And then he goes in an unexpected direction.

“We are working super hard on all the software optimizations, I mean, just not the software optimizations that come because of what DeepSeek has done, but all the work we have done to, for example, reduce the prices of GPT models over the years in partnership with OpenAI.”

Woah.

Anyone who doubts that DeepSeek is real, or whatever, should read that quote over and over again. When Nadella discussed optimizing Microsoft’s AI infrastructure, he put the work DeepSeek has done not just in the same category as ongoing efforts with OpenAI, but ahead of it. He clearly believes that something fundamental has shifted, and that the optimizations DeepSeek unveiled will directly impact Microsoft and its costs. That is fascinating, that this company may have sped up the pace at which AI becomes more affordable to Microsoft as an infrastructure provider, and so suddenly. It’s a strange thing to admit, because it immediately makes one wonder what the plan was otherwise.

Nadella also discusses an important component to the AI build-out that I can’t say I’d ever considered: By being so ahead of the game here, so to speak, Microsoft has to be careful not to buy too much hardware that will be out-of-date and inefficient in the near future. It’s a delicate balance.

“Remember, you don’t want to buy too much of anything at one time because of the Moore’s Law every year is going to give you 2x [efficiency/performance improvemetnts],” he continued. “Your optimization is going to give you 10x. You want to continuously upgrade the fleet, modernize the fleet, age the fleet, and at the end of the day, have the right ratio of modernization and demand-driven monetization to what you think of as the training expense.”

Here, Ms. Hood jumped in with some financial clarity, Microsoft, she said, has “close to $300 billion of RPO that is committed customer contracts that need to be delivered on.” RPO, or remaining performance obligation, is a set of contracts with customers who will pay it that amount over time, assuming Microsoft can meet the demand. Not all of that is guaranteed, but the point is that Microsoft is spending $20 billion per quarter in part because it has a potential gain of $300 billion on the books. Once again, the point is the shut up the doubters.

The next question was about Copilot, but again touching on the concept of cost vs. profits. Here, Nadella talked about the evolution of Copilot to include agents, with the net result being more seats (Microsoft 365 Copilot) licenses. And he notes that the addition of OpenAI o1 globally brings down the costs of delivering these services. Microsoft is recognizing these gains in GitHub Copilot and Security Copilot too, he said, thought both are likely tiny customer bases right now.

The next question was about proprietary AI models (like those offered by Microsoft and OpenAI) vs. open AI models like those offered by DeepSeek and Meta. His answer was predictable, that there will be a mix of both, just as there is in software generally, and that Microsoft will choose accordingly.

“You want to always build your application with high ambition using the best model that is available, and then optimize from there on,” he said. “You build models, you finetune models, you distill models. Some of them are models that you distill into an open source model. There’s going to be a combination.”

The seventh question went back to Copilot and where the big growth markets are.

“The initial set of seats were for places where there’s more belief in immediate productivity, a sales team, in finance or in supply chain, where there is a lot of, for example, SharePoint grounded data that you want to be able to use or in conjunction with web data and have it produce results that are beneficial,” he said, meaning traditional Microsoft 365 commercial use cases. “But what’s happening [now] … is that people collaborate across functions, across roles … I call it ‘think with AI and work with people’. And that pattern then requires you to make it more of a standard issue across the enterprise. So, that’s what we’re seeing. It starts maybe at a departmental level. Quickly, the collaboration network effects will effectively demand that you spread it across. You can do it by cohort and what have you. And so, what we’ve made it easier even is to start with Copilot Chat plus this. And so, that gives the enterprise customers even more flexibility to have something that’s more ubiquitous.”

The final question was about about commercial RPO, expanding on the comments Microsoft had made earlier about bookings. Here, the ongoing OpenAI commitments are “a big component,” Hood said, one that will “absolutely” grow, benefitting Microsoft. And then there are “core motions” across its customer base to upsells like “Copilot or GitHub Copilot or other processes” that are “important” too.

This bit is almost humorous.

“We also had a good E5 quarter,” she added, alluding to one of the more lucrative tiers of Microsoft 365 commercial, “which when we talk a lot about Microsoft 365 Copilot, I sometimes forget to also talk about suite momentum. And we saw that as well this quarter, which we felt very good about.”

I guess I’m not surprised that Microsoft collectively ignores the established businesses that continue to drive revenues as it plows forward on new initiatives, in this case, AI. All companies, including Microsoft, do this all the time. Wall Street cares about growth and opportunity, so that’s the messaging, but traditional businesses–and, yes, Microsoft 365 is suddenly a traditional business–earn all the revenues.”

?‍♂️ The rest

Like Microsoft, I have little to say about the company’s other client businesses, as there was little movement anywhere. But I do have some notes.

? Windows

Windows only came up 14 times in the post-earnings conference call, compared to 16 times for OpenAI, over 60 times for AI, and almost 50 times for Copilot. Windows is just sort of there, I guess, having plateaued at whatever level despite a push for AI PCs and Copilot+ PCs. Microsoft says it sees “momentum build” as the Windows 10 end of life date approaches, but there’s precious little in the way of positive hard numbers.

Windows OEM and devices revenues (read: revenues from PC makers, including Surface) were up 4 percent, above the PC sales trends we see because of PC maker pre-builds ahead of the Windows 10 EOL upgrade shift. But Microsoft expects declining revenues from PC makers in the current quarter because “inventory levels will normalize.” So there’s your Windows EOL bump: About 2.7 percent when compared to the overall market. Eh.

Microsoft appears to be hiding Surface losses inside one of the more profitable parts of the company now, which is interesting.

Microsoft Edge commands less than 5 percent usage share worldwide, but Microsoft claims that its browser is now used by 30 percent of users in the United States. It has experienced share growth for 15 consecutive quarters (US or worldwide, it’s not clear).

 ?Copilot+ PCs

To date, there has been exactly one report about how well Copilot+ PCs are doing in the market, and it’s one I never trusted or believed. But now we have a more authoritative source with admittedly vaguer data. Satya Nadella said that 15 percent of premium-priced PCs in the US over the holidays were Copilot+ PCs, and he expects “the majority of PCs sold in the next several years” to be Copilot+ PCs. This is a low bar. As I’ve said, almost all PCs will be Copilot+ PCs in the near future. But it is perhaps ironic that Nadella also pointed out that “the most powerful AI workstation for local development is a Windows PC running WSL 2 powered by NVIDIA RTX GPUs.” That PC is not a Copilot+ PC. Perhaps that can change this coming year given the superfluous nature of the requirements.

?‍? Surface

Microsoft did not mention the term Surface at all during the post-earnings conference call. As noted, Surface is now part of Microsoft’s revenues from PC makers, so it’s a net drain. The only mention of Surface comes implicitly in the company’s earnings slid deck, which notes that the 4 percent revenue gain from PC makers was “partially offset by a decline in Devices,” meaning Surface.

? Xbox

Xbox was an unmitigated disaster in the quarter. Overall Gaming revenue down 7 percent. Xbox hardware revenue was down 29 percent in a quarter in which Microsoft offered the most mild of console refreshes imaginable and little in the way of sales or other incentives. And Xbox content and services revenues were up just 2 percent, “driven by growth in Xbox Game Pass.”

“All-up, Game Pass set a new quarterly record for revenue and grew its PC subscriber base by over 30 percent, as we focus on driving fully-paid subscribers across endpoints,”Nadella said. That bit is interesting, and no doubt reflects Call of Duty finally appearing on Game Pass. Indeed, Nadella also mentioned that “Black Ops 6 was the top-selling game on Xbox and PlayStation this quarter – and saw more players in its launch quarter than any other paid release in franchise history. ”

Xbox Cloud Gaming experienced a “record 140 million hours streamed,” and over 4 million people played Indiana Jones and the Great Circle.

And … yeah. That’s it. But here’s the strategy for Xbox.

“We are focused on improving the profitability of the business, in order to position it for long-term growth, driven by higher-margin content and platform services,” Nadella said. “We are delivering on this plan.”

Maybe so. But this is what I think of as a holding pattern. It’s not a good look today, and not what fans want to see. Worse, the near-term future is no better. Gaming revenue should rise by single digits this quarter, while Xbox content and services revenue will rise by low to mid single digits, and Xbox revenue “will decline” YOY. As usual. But how much longer can it decline? When do we hit rock bottom?

? Microsoft 365

Microsoft 365–which Microsoft executives routinely called “M365” throughout the call–continues its steady upward trajectory, with a reliable 16 percent increase in commercial revenues and a strong 13 percent gain in consumer revenues, “significantly ahead of expectations”. Paid Microsoft 365 seats (licenses) were up 7 percent, driven by small business, medium business, and frontline worker offerings, and revenue per user was “up,” but we don’t know by how much.

This should be embarrassing to Microsoft, but revenues from the perpetual Office 2024 offering were higher than expected, and this had a positive impact on commercial revenues.

Microsoft 365 cloud revenues were up 8 percent, slightly ahead of expectations.

Microsoft 365 consumer revenues were up 8 percent, too, and consumer subscriptions were up 10 percent to 86.3 million, thanks in part to the Microsoft 365 Basic offering.

There’s just not much to say here. Looking ahead, the current quarter looks strong, meaning the same as ever. Microsoft 365 is a well-oiled, money-making machine.

That said, I will end on Microsoft 365 Copilot, which, yes, is AI, but also part of Microsoft 365.

“Microsoft 365 Copilot is the UI for AI,” Nadella said during the call. “It helps supercharge employee productivity, and provides access to a swarm of intelligent agents to streamline employee workflow. We are seeing accelerated customer adoption across all deal sizes, as we win new Microsoft 365 Copilot customers and see the majority of existing enterprise customers come back to purchase more seats.”

More specifically, customers that adopted Copilot early have expanded their seats collectively by 10X over the past 18 months. The number of people using Copilot every day has doubled quarter-over-quarter. Copilot usage among employees (?) is up 60 percent quarter-over-quarter. Over 160,000 organizations are using Copilot Studio, and they’ve created over 400,000 custom agents in just three months, up 2X quarter-over-quarter.

“We see Copilot, plus agents, disrupting business applications,” Nadella added. “And we are leaning into this.”

They sure are. Buckle up, folks. The Copilot train isn’t slowing down anytime soon.