Key Takeaways
- Microsoft MSFT isn’t a typical software company.
- What Microsoft’s key competitive advantages are and why they’re durable.
- How much Microsoft might spend on artificial intelligence and how quickly revenue could grow in relation to the company’s AI buildout.
- Forecasts for Microsoft’s revenue, operating margin, and earnings.
- Why Morningstar is more constructive on Microsoft stock today than the market is.
- The biggest risk to our USD 600 valuation on Microsoft stock.
In this bonus episode of The Morning Filter podcast, co-host Dave Sekera and Morningstar senior equity analyst Dan Romanoff take a deep dive into one of Sekera’s top stock picks: Microsoft. Microsoft isn’t just a software company; it’s a portfolio of businesses, and Sekera and Romanoff discuss the company’s individual segments. They cover why Microsoft remains a wide-moat company in the face of AI disruption and walk through its moat sources.
They unpack Morningstar’s forecasts for the company’s AI spending, revenue, and margins and explain how Romanoff arrives at a USD 600 fair value estimate for Microsoft stock. They wrap up with a discussion about the key risks Microsoft faces today.
Have an idea for a bonus episode of The Morning Filter? Send it to themorningfilter@morningstar.com.
Transcript
David Sekera: Hello. I’m Dave Sekera, co-host of the weekly podcast, The Morning Filter. Today I’m joined by Dan Romanoff, senior equity analyst on our technology team, who covers the software space. Today is Wednesday, May 20, and we’re conducting a bonus episode of our podcast to conduct a deep dive into our analysis and valuation of Microsoft.
Now, Microsoft stock peaked in October 2025, but it’s sold off pretty significantly since then, having been caught up in the broad decline in all of the different software stocks. As of today, Microsoft is a 5-star-rated stock, our highest risk-adjusted stock rating, and it trades at nearly a 30% discount to our intrinsic valuation. In fact, by market capitalization, this is the greatest differentiation between our valuation and the marketplace today.
Thank you very much, Dan, for joining us here today.
Dan Romanoff: Sure, Dave. Thanks. Happy to be here.
Sekera: Well, we all know Microsoft. We’re all familiar. We’ve all used Microsoft Word. We’ve all used Microsoft Excel. We’ve all certainly made slide decks and PowerPoint, but there’s a lot more to Microsoft than just Office. I was hoping maybe just start us off today with giving us a brief description of the company overall, some of the different business segments and maybe how those segments fit together.
Romanoff: Sure. Yeah, you’re right. There’s a lot going on here, and so Microsoft is in three segments the way they define it, which are Productivity and Business Processes, Intelligent Cloud, and More Personal Computing. Productivity is sort of like 40%, a little more than 40% of revenues. Intelligent Cloud is a little less than 40% of revenues, and More Personal Computing is maybe in the high teens, 19-ish % of revenues. And so that’s sort of the split.
Within Productivity, that’s where you have the apps, you were sort of naming, that as the main one in there, is Microsoft Office 365, as they call it. Also in there is Dynamic, so that’s sort of their ERP solution. That is a pretty big piece of software, which would be a substantial market cap if it was its own company. Also in there is Copilot. There’s certain parts of LinkedIn that are in there as well.
So like Talent Solutions is sort of like an HR software piece almost that they are licensing out. And so that gets rolled up in there. And so that’s what that business is. Intelligent Cloud is largely Azure, but there’s other stuff in there, too. They did a Nuance acquisition a few years ago for some healthcare-related stuff. That’s where that falls.
Those are the main things. Well, there’s the Windows servers and the SQL server business—that’s also in Intelligent Cloud as well. So those are the main things that are in there. GitHub, I guess gets a lot of airtime. That’s a pretty decent business for them in Intelligent Cloud also. And then in More Personal Computing, that is largely Windows. That’s where you find Windows and then gaming. So like the Activision acquisition, all the Xbox stuff, Game Pass, that all rolls up into gaming. And then lastly, you have the Bing, the news and search, the Google-competitive product, and Microsoft Edge, their browser. So that’s what—sort of, end devices for whatever is left in that business.
Sekera: Now you mentioned Azure, and of course everyone all wants to talk about artificial intelligence today, that being their cloud hosting platform. Could you maybe just describe a little bit more about what Azure does and maybe some of your thoughts on the growth prospects there?
Romanoff: Yeah, sure. Azure is like we call it public cloud is sort of like what everyone, I don’t know, grew up, I guess, calling it. And it’s sort of like, I don’t know, they’re one of the three hyperscalers. So it’s like there’s this oligopoly of Google GOOGL, Amazon AMZN, and Microsoft. And so the three of them own the public cloud market. So you kind of are renting out compute power and storage capacity and other software elements that all roll up into this large menu of cloud services. And so if you’re a business, it may be easier, cheaper, more flexible for you to build out some workloads in a cloud setting rather than buy your own servers and configure them and host the software applications. So that’s what Azure does. In terms of the importance of the industry, that is the main growth driver, I would say, within software overall.
It’s just traditional workloads moving to the cloud. That’s been a theme for years now. And then of course, more recently, I’m sure we’ll talk about this more, so I won’t front-run it too much, but AI falls into Azure as well. And so I’ll just leave that one there for now, but hopefully that kind of gets you what you’re looking for.
Sekera: All right. Well, continuing on the AI theme then: What is the relationship that they have with ChatGPT and OpenAI today?
Romanoff: That relationship has definitely evolved over time. And so Microsoft is a longtime investor in OpenAI. They were one of the early seed financiers for the company. And then I think it was about three or four years ago they did what was reportedly a USD 10 billion round of financing. And so they kind of ended up with a 49% interest in OpenAI. And so there was this exclusivity where Microsoft got to exclusively commercialize some of OpenAI’s models like GPT-3 at the time, I think it was. And so that commercial license was really valuable to Microsoft. And what OpenAI got in return was they got to use Azure basically as the all-purpose cloud computing engine that drives their business. And they got credits basically from Microsoft and their USD 10 billion. So that was how that arrangement worked. And then there were reports within a year or so that there was tension building between the two.
And so they changed the nature of their agreement, and Microsoft—and there were more financing rounds along the way. So Microsoft didn’t own as much. I think it eventually went down to 33%, and that exclusivity got thinner, and OpenAI was able to expand to other cloud providers as well. And so the current iteration is that Microsoft owns a 27% stake in OpenAI, and they’re both free to do whatever they want. So Microsoft can use Claude in some of its—Anthropic’s Claude—in some of its software ‚and they still have like a right of first refusal to commercialize some of OpenAI’s models. That’s kind of how it works. They own 27% of it now, but it seems professional, and they have a good relationship, and both are important to the other.
And so the relationship hasn’t soured, I wouldn’t say, and it’s still formal, and it is meaningful for both of them. So I think it’s good for both parties the way it is now.
Sekera: Now when I’m thinking about Microsoft’s portfolio of businesses that they have, how those businesses fit together, and think about that in the terms of our economic moat rating—we, of course, rate Microsoft with a wide economic moat—could you just walk us through what are those primary moat sources, and how do you think about those sources for some of those individual businesses?
Romanoff: Sure. So yeah, we rate Microsoft as a wide moat. And with software, typically the primary moat source we see is switching costs, and that is definitely the case for Microsoft. What really is going on there is if you’re a company and you install a system or a software application, that takes time and effort. Maybe you had to involve Accenture ACN or someone else to come in and do that work for you. And so you’ve spent a lot of time and money doing that, and then your business revolves around whatever the software that you’ve installed is. And so it becomes embedded in all of the processes that you’re undertaking. And so there’s a switching cost, and it shows up in metrics, too. I mean, there’s this retention metric that, in software, we talk about all the time. And for important systems, that retention is usually around 99%.
Companies do not change. They kind of bend over backwards to not change even if they hate the software product. And so that is the switching cost argument in a nutshell, and that’s the primary thing going on at Microsoft. There’s some other moat sources as well, but from a switching cost standpoint, it’s basically all of the software, whether that’s Windows or Office. I mean, there’s not really even alternatives for some of these things, but like the Dynamics ERP system, like definitely wide moat. It’s an ERP system. It sort of runs your entire business, and so, you really can’t change that. Those are some of the products. Azure, there’s a wide moat there. You do the same thing there. You’re building your business. There’s all these services that you’re going to consume from Azure, and so, sure, you can move data around, but there’s a lot more to it than that.
So once you build a business to function a certain way, it becomes really tricky to change it. So wide moats based on switching costs are pervasive throughout the portfolio, I would say.
Sekera: Now in conceptualizing artificial intelligence, I know over the past couple months, Morningstar’s Equity Research Group really went back to the drawing board, reevaluated our economic moat ratings on a number of different companies, specifically those that our analyst team thought could be most at risk in the future from artificial intelligence. There were a number of wide-moat companies that we downgraded to narrow moat and some narrow-moat companies where we actually stripped the moat away to a no-moat rating. So I guess in thinking about Microsoft and their businesses and their economic moat, what do you think really differentiates or what strengthens that wide moat to the degree that this was one that we did not consider moving down to a narrow moat?
Romanoff: That exercise was good. I think the main reason you saw for moat downgrades just in a nutshell was the uncertainty now is just a lot higher. And so there’s really no evidence that software is suffering from AI right now, but you can look out a few years and say, well, maybe things can start going south. And so there’s just less certainty about what the next 10 years looks like for sure. And so that sort of drove the moat downgrades. From Microsoft’s perspective, I mean, there’s a number of dimensions I would say you can explore to assess the threat from AI and it’s like, does your company or does your software generate or consume proprietary data? Are you reading or are you writing to a database? Do you sell database-related products? How important is the software to the business?
If it’s just some workflow-processing tool, maybe that’s not super important, but again, Dynamics ERP from Microsoft, like your business is built around that. It’s hard to replace that with AI. And so when you start looking at it through those different factors, and you look at the different products and maybe like, what’s the pricing model, you kind of start to see that Microsoft is pretty well insulated for the most part. It’s not perfectly. And then I always like to say when I’m talking to clients that, “Well, even if this goes totally south, Microsoft still has Azure.” And so if software does get eaten alive by AI, well, then all of your AI inference is going to run on Azure. And so it’s easy to see how, in our opinion anyway, that Microsoft can thrive in the world that exists today, and if software continues on and AI isn’t really the threat people think it is, Microsoft can be fine.
Or, in the other world where software is eaten alive, Microsoft probably does pretty well, too, because Azure is hosting all that inference, so.
Sekera: Their diversification in their portfolio of business has kind of naturally offset one another in different types of scenarios as far as what AI may or may not be able to do in the future.
Romanoff: Yeah, absolutely. Absolutely.
Sekera: When I think about artificial intelligence and I think about all the different aspects of artificial intelligence and all the different moving pieces, is there a way from that 30,000-foot view you could describe: What is a Microsoft’s strategy in how they’re approaching AI today, how they’re setting up their businesses to be able to make sure that they continue to be leaders in that business?
Romanoff: Well, from a strategy standpoint, I would say two big things. One is they’re trying to push AI into all software elements that they have. And so you can find AI in Microsoft Office. They’re building computing, like Windows has an AI-certified sticker that goes along with it. They’re building AI features into Dynamics. So you’re seeing it being pushed throughout the portfolio just as a software company, which is kind of normal in what all the other software companies are doing. Microsoft is a little different because they have Azure. And so they have this whole menu of AI services that they’re creating and offering out to clients. And so if you’re a big company, a small company, it almost doesn’t matter, they can serve you. They have Azure. And so as I mentioned, that’s where all the inference is happening, that’s where all the model training is happening.
And so Azure is absolutely critical in the AI world, is sort of capturing the lion’s share of the revenue just from the AI industry at large, if you will. Microsoft is a major player there, but even within those data centers, they have their own chips now as A) a point of differentiation, cheaper on a price-performance per watt basis. And so in a vacuum, maybe not as “good” as an Nvidia NVDA GPU, at that inference and model training, but from a specific price performance per watt standpoint, they perform better. It’s like any other ASIC chip—that’s basically they’re designed to do one specific thing, and they do it very well. Microsoft has gone down that road. They’ve created their own large language models now. So they’re not even aware of that one. When I tell them, well, Microsoft has Microsoft AI, MAI is their product now, MAI1.
And so they have their large language model. It’s not as well known as, say, Claude or GPT-4 or whatever, but it’s there. They have small language models. They have a whole family of them called Phi, the Phi, like the Greek character. And so there is an entire menu of services and solutions that they can provide you, and they want to be everything to everyone. They have managed services. You’ll find similar menus, incidentally, on Google, which I don’t cover, but on AWS, you have the same options as well or similar options. And so they’re all building out similar things, and they’re standing alone, the three of them. There’s just nowhere else you can really go to get this array of AI-related tools.
Sekera: Now thinking about AI, and of course, we’re in really still the early stages of this AI buildout boom, still running full steam ahead, in fact, in some ways it seems like it’s still continuing to accelerate—just how much money is Microsoft spending on building out artificial intelligence? So what are you forecasting for capex spending this year, and of that total capex spending, about how much of that do you think is going toward just the AI buildout boom?
Romanoff: Yeah. I mean, the numbers are staggering, and they’ve really gotten out of hand in a hurry. This year, we’re in fiscal ’26 for Microsoft, and we’re in their fourth quarter right now, they will spend approximately USD 120 billion in capex for the year that’s about to end. And last year it was, I think, USD 65 billion. So it kind of doubled in a year, which is—the dollar amounts are staggering. And in terms of AI, it’s not a number they give, but I think it’s safe to assume that most of that 120-ish billion dollars this year will be spent on AI. They will tell you that all of their data centers are mixed-use, and so there’s a lot of traditional workloads being allocated for, if you will, within the footprint of the data center, but it’s not obvious if it’s 50% of the floor spaces for AI and 50% is for traditional.
But the cost of an AI server versus a traditional server is mind blowing. And so maybe this will help the listener get their arms around this: For a single rack of servers for a traditional workload—so just like whatever, email, a bunch of email servers in a data center—that would cost you, you can get maybe 20, 15 or 20 servers in a rack, that would be loosely 100,000 bucks by the time you put the cables and the cooling and the switch on top of it and the power units to the rack itself. For an AI rack, it’s loosely USD 4 million. And so you’re talking about just an exponential increase in cost. And so that just leads me once again to believe that if you had to pin me down easily three-quarters of the spend—probably maybe not that much but, because some of it is still the building itself—but within the technology equipment, easily three-quarters of it, I would say, is for AI-related.
Sekera: If we’re getting close to the end of this fiscal year for the company, what are you projecting for their spend in capex next year?
Romanoff: Just USD 180 billion. It’s a walk in the park. A modest increase.
Sekera: Well, and again, one of the big things we’ve been hearing a lot about in the marketplace is with the amount of capex spending that these companies are spending, what does that do to their free cash flow? For a company like Microsoft, based on your estimates for next year, are they still going to be free cash flow positive, or are they going to be free cash flow negative, have to go to the debt market in order to be able to fund this buildout in the short term?
Romanoff: Yeah, it’s really staggering. As big as these dollar amounts are, believe it or not, Microsoft will still be free cash flow positive next year, or should be. And obviously there’s room for other stuff, but I think they should generate, I don’t know, at least USD 30 billion probably in free cash flow despite USD 180 billion of capex. I won’t say they have an unlimited capacity to fund a buildout like this because that implies that it gets even crazier, but they can continue to absorb this for a few years even at this—if we assume it’s an elevated pace, they can continue to fund it for several years without really breaking a sweat. The investors probably, the shareholders probably won’t agree with that, but just financially they could make that work.
Sekera: I guess the real big question, then, I always have is, how long is this spending going to be elevated like this? I mean, are we at the point where we’ll see a couple years of this elevated spending, they’ll get the capacity that they need, and then they’ll be able to dial that back, or to some degree, is this just a permanent shift change? When you’re thinking about modeling out capex, I don’t know if you look at it as modeling it as a percentage of sales or how you do it, but when you’re thinking about forecasting it three, four, five years out into the future, how are you conceptualizing that in your financial model?
Romanoff: Yeah, I would say my models have evolved on that as well, because I used to think that going back a couple years, it would be a couple years of a big building bonanza and then it would flatten out for a year and then maybe wind down even a little bit. And now we’re a few years into this building boom and the numbers keep going up sharply still. And so that USD 180 billion for next year turns into USD 200 billion for the next year after that. And that is pretty much still a consensus-y kind of number, but I’m still expecting more than 10% capex growth the year after that. And I no longer think the dollar amounts go down ever. It’s like this is just a new level.
Sekera: Yeah. So this is a permanent shift.
Romanoff: It is a permanent shift, but I would say, even if in five years, say, that number is USD 200 billion, becomes USD 175 billion or something like that, to me that still is the same thing. You still have a permanent step up in capex because now that you’ve built all these data centers and you’ve equipped them largely with this five-year depreciable asset, those servers are going to need to be replaced every five years basically. And so that’s the investment cycle you’ll be on. So you got to believe that Azure will still grow and so therefore they will be building new data centers along the way, but maybe not to the frantic pace that they’re doing now.
Sekera: And we’ll get into your forecast in a couple more questions, but when I’m thinking about the capex spending, that goes onto the balance sheet for now, but at some point in time, you’ve got to drive additional revenue in order to be able to make that economically viable. How do you think about that push/pull between the amount of money they’re spending today? How quickly do you think that they can bring on a new revenue in order to make it economically viable, and how does that flow through your thought process?
Romanoff: All of these questions are very challenging to triangulate, and so we do our best. And I would say, again, for the investor, Azure, for a book of business that’s USD 75 billion, it’s growing at 40%. I mean, it’s crazy how big this is and how fast it’s growing ‚and what allows for that is the pace of investment. As soon as they build a new data center, it is open and fully utilized immediately. So it’s not like the old days where you built a data center and you fill up the capacity over the course of a couple years. No, it’s fully utilized, but the problem is because the demand signals are so strong, Microsoft keeps plowing more money into it. And so you have this mismatch where revenues for Azure, specifically, are going up sharply, 40%, it’s an impressive number, but the capex is going up loosely 100% from USD 65 billion to USD 120-ish billion dollars this year.
And so until that normalizes, I guess it’s really challenging, but I definitely have done some work on this, and I would say if you look at the building and the equipment that’s going in it and the land, you get to this number that is around seven, seven and a half years of an average useful life of all that equipment. You can say that from USD 120 billion of new capex this year, your incremental depreciation should be around USD 16 billion from that, just based on that. They’re managing this fine so far in that margins are going up, operating margins are going up, gross margins are feeling a little bit of pressure. And so there’s this tug of war happening internally. It’s definitely challenging from a management standpoint. I would say they’re doing a good job so far, but what does that look like in a couple more years, once you see the USD 200 billion number coming out there?
We’ve attempted to do it in our model, and I’m sure our model falls short in some way, but we’re trying to capture it in there. And so I kind of see this, I don’t know, this tug of war where margins are loosely flat. They’re offsetting some depreciation pressure by internal AI usage, and they just did another headcount—a voluntary buyout—headcount reduction. And so they are finding ways to save money, but if this continues, I think that there will start to be a little bit of operating margin pressure. And so that’s one of the key debates probably about the stock is how this all develops.
Sekera: I know one of the big questions from the marketplace is as the depreciation rolls through over next couple quarters, next couple of years, and that’s a hit to earnings, is that going to be more than what the accretion is going to be to earnings? Could we actually see that reduce earnings growth for the next couple of years? And then, if so, would you then actually have faster earnings growth in the out years, or are they just really managing this such that as you were mentioning, they’re kind of being able to manage the amount of depreciation expense as that new revenue comes in so it’s not going to be something that would really end up impairing earnings growth in the short term?
Romanoff: Yeah. It’s funny, I spend a lot less time looking at earnings than I do at the rest of the business. And so if you started talking about it that way, I do have earnings growing slightly slower than revenue, say, over the next five years or whatever, but only slightly. And so a lot of that is from an initial depreciation wall that hits and so earnings growth slows in a one-time basis, if you will, and then grows and picks up from there. And so over the course of, say, five years, it’s not a great difference between the revenue growth and the earnings growth, but I do see that coming. Yeah.
Sekera: So again, let’s get into, then, forecasting the company overall, and, again, a company that’s just a behemoth, has at least three different segments that they report, each of those segments having a number of different business lines underneath, how do you go ahead and put together your top-line forecasts? I mean, how do you break out those segments? Do you break out individual products within the segments? How do you build that up to get to that top-line growth? For example, what would you be looking at for growth this year?
Romanoff: Because there’s a lot going on, it is challenging. What I found is the most effective way to do it, especially in the context of you’re updating your model in the span of a few hours and writing a note during earning season—so you have to have a way to do that somewhat efficiently—and so my approach has been to model the segments at a high level and just say, OK, Intelligent Cloud’s going to grow 25% or whatever this year and Productivity will grow 12%. Sort of do it that way and then get underneath it and do sanity checks and say, OK, well, if Azure is going to grow 40% or so this year, what does that mean for, whatever, the decline in Windows Server or stuff like that. And then on a quarterly basis, they do give you some flavor sometimes.
They’ll guide to the high-level segments and so it’s short-term. We obviously try to be more long-term than that, but it helps to just make sure that you’re being honest and staying in between the lines. But I am always checking what’s implied for the number of units of Windows or how many Office, commercial Office seats are being sold or are producing revenue in a given year. And so the bottom-up stuff is happening as like sanity checks, for my model anyway, more than it is for like specific product buildouts. It’s a little bit of a triangulation, but I mean, that’s how I do it. And then you asked about the forecast for the year.
Sekera: Yeah. So you said, I think we’re toward the end of what’s fiscal year ’26 right now and about to roll into fiscal year ’27?
Romanoff: Yeah. I think for this year, I mean, there’s one quarter left, but the number for this year is approximately 17% growth. I think that gets you to USD 329 billion. And so next year, growth will be a little bit slower, largely because you’re just dealing with larger numbers. And so Azure, for example, you can’t grow what will then be a hundred-ish billion dollar book of business. You can’t be accelerating that forever. And so growth should decelerate slightly. And so you come to this number overall that is probably more like 16%, something like that. I forget the exact number.
And then over the course of the next few years, the main driver of growth for sure is Azure and all the AI-related stuff. Every once in a while they’re spitting out data points that are sort of super helpful to say, “Ooh, I have this sort of off in my model, or pat yourself on the back. I can’t believe I got it exactly right.” That stuff happens, too.
And so we’re always thankful when they do provide a data point, but like Copilot is a number they’ve given recently, 20 million Copilot seats, and if you do the math there, they say they charge 20 bucks a month. I doubt that’s what it is. I’m sure there’s some enterprise discounting that’s happening there, but it’s pretty easy to see that that can add $3 billion of revenue. For example, that’s like a hundred basis points of growth, for example. And so there’s all these little levers that are being pulled all the time to help you get to your final number.
Sekera: Out of curiosity, as you mentioned earlier, we do try and think about these companies in the longer-term perspective. For the next five years in your model, what would be that compound annual growth rate for the next five years?
Romanoff: For Microsoft revenue, it’s 15% revenues compounded over five years. We were asking about earnings also a question or two ago, and that number is like 14% to 15% growth. And so there is some pressure that I’m seeing on margins, but like I said, the linearity of that is not perfect.
Sekera: That was actually going to be my next question as well: With the operating margins, how do you think through your forecasting there? How are you putting your numbers together? Again, are you trying to model out those individual business segments? Are you modeling out some of those individual business lines in order to come up with that margin and put together a mosaic? Or are you really coming down more with that top-down approach?
Romanoff: Microsoft has been remarkable over a long period of time now, over the last decade operating margin and, interestingly, they report GAAP. They’re one of the few software companies that’ll report a GAAP number. Their GAAP operating margin has gone from 25% to 46% over 10 years. That’s pretty astounding, I would say, for a mature company. The starting point of that measurement is important because you had a subscription transition happening, and then you had Azure being immature 10 years ago, and it’s kind of mature even though it’s growing as fast as it is. So you did have some helpers along the way to try and get you here. And so I would say that investors need to understand that a software company, every expense line is basically a leverage opportunity. So it’s not like a manufacturing company where you just have a fixed overhead somewhere that is overhanging the entire business.
You can get cost of goods sold, for example, to give you leverage. And maybe Microsoft isn’t there exactly yet because of the big data center boom that’s happening. But in a normal course of business, it’s happening that way. And so, yes, I do think of it from a business segment perspective. And so like the margin for intelligent cloud, for example, probably will go down a little bit to account for all the depreciation, for example. But they’re finding efficiencies elsewhere in the business. And so the overall margin probably is going to go up this—well, it will go up this year, and maybe next year it’ll be flatline. And so that’s the debate. Does it go down 10 basis points? I mean, to me, that is flat. If it goes up 20 basis points, it’s still kind of flat. But I do expect there to be a flattening for sure in the next couple of years just at an overall level.
And so once you get to a more of a steady state with capex and what Azure looks like and Intelligent Cloud looks like and once capex grows loosely as fast as revenues does, and so this may take five years still, but once that happens, then you should be back into a normal margin expansion of 25 basis points a year or whatever. Microsoft, for as big as it is, it’s sort of elite margins. They’re kind of at the top of the margin curve for software companies. And so that, I think, is a little bit of a limiter. And then right now the weight of Azure, which is a slightly lower margin business, pulling the corporate average down is like a counterweight. Those things are all counteracting each other. So it’s a challenge, but whatever. We’re doing what we can.
Sekera: Yeah. Well, it’s interesting. When you look at Microsoft stock price, it peaked essentially last October, and really it’s been trending downward ever since. It looks like maybe we bottomed out about a month and a half ago and we’re kind of coming off that bottom.
But generally it looks like it’s really been caught up with all the software stocks that have all been trading off. Some have been trading off for at least a year now at this point. I think the market is looking at the software space, in particular, trying to understand exactly how artificial intelligence are going to impact these companies. What are they going to displace going forward? What kind of disruption are we going to see?
I’m just curious, from your point of view, when you think about your investment thesis overall for Microsoft and how that gets you to your valuation, based on where it’s trading in the marketplace today, what do you think investors are pricing in?
Romanoff: The multiples for software companies, obviously, we use DCF models to do our valuation here, but when you look at the way the stocks have traded, it’s really shocking that the stock performance has been as bad as it has been in the face of, if you just start drilling down on any one company, software companies throughout 2025 were basically beat and raised on a quarterly basis. We cover more than a hundred software companies here at Morningstar. So this goes beyond my companies. The average was beat and raise every quarter, and that continued for the first quarter. So things are going well, and if you start digging into the average metrics, which we do every quarter in our quarterly pulse, there’s no change to anything. The retention remains 95% on average. On a dollar basis, it’s like 106%. I mean, that’s what it’s been for the last few years, so there’s no changes.
And so things look like they’re going reasonably well for software, but the multiples have compressed, the stocks are down like, I don’t know, say, 40% on average. And so all of that performance has been multiple compression. So to me, that is interesting. And so when I take our DCF, yeah, I like to run scenarios, and coming up with a fair value estimate and say, “Well, what if Azure grows faster? What if it grows slower?” And we do those kinds of things for sure. But I like to do another exercise, which is to just take the DCF model and say, “OK, the stock is trading, before we came in here, it was at USD 415 for Microsoft.” And so if I need to take my model and flex it to say the stock to show a valuation of USD 415, what do you have to do to make that be the case?
In the case, I mean, there’s two levers that I will normally pull on, and one is revenue growth and the other is margins. And so there’s obviously a million ways you can do this, but for revenues, if you just assume that this year and next year, they’re approximately correct and the consensus number itself will be good, and my estimate is pretty close to consensus—if you do that and then decelerate revenues to around 5% over—5% growth in year 10—and then for your terminal period, you assume like GDP kind of growth, you can get to USD 415.
And so in the face of the Azure boom that is going on now, like you have investors pricing in growth of GDP, that is crazy to me. And if you flip it around and you do the same thing with margins and you just say, OK, margins are great, and some incident will happen—maybe it’s all this AI stuff—will just kill margins. And so starting in 2028, you have to take margins and move them from 46% to 32%, and that’s your operating margin in perpetuity. And so that will get you a valuation of USD 415 per share. There are no precedents for that happening in software. I’ve been doing this for 25 years. It’s never happened one time. And so can it happen? Sure, it can. Do I think it’ll happen? No.
And so from a valuation perspective, I just think those things are unlikely. I’ve seen the death of software, I don’t know, at least a half a dozen times now. And so, is this different? Yeah, maybe a little, but it’s AI, and it can also help software, too. And so there’s a tug of war there. I don’t believe all of the fears. I think there are reasons to worry when you’re contemplating how much you should pay for a share of Microsoft, but at the same time, I think the future is not bleak and there will be no more software. I don’t think that’s happening.
Sekera: I mean, overall, do you think it’s just going to be a matter that they need to change the pricing of their business model? Again, something where it might be seat-based today, might be seat-based plus consumption, or is it just a matter of there’s always going to be economic value that’s being added. They’re just using that AI to add that value, but there’s going to be some way that they still extract that from their customers.
Romanoff: The software industry in general is struggling with that. And I think the answer has been obvious for a while, and there have been companies in the marketplace that have been using consumption or work-performed models for a long time already. And so the way this is shaking out is if you’re serving up AI of some kind, and so whether that is Microsoft or Salesforce CRM, who has some AI business, you’re going to charge a consumption basis for that, and then you’re going to charge a seat-based pricing model for the rest of it. And so it’ll be a hybrid model. That’s the way this is all going right now. The same is true for Microsoft, but so much of Microsoft, basically all of Azure, is already priced on a consumption basis anyway. This will be nothing new to anyone.
Morningstar is paying for our per employee per seat for Office, that’s normal. And then to the extent that we use, or whether it’s Azure or AWS or whoever for like cloud and AI stuff, we’re going to pay them for the consumption of that. So that’s the way all this is shaking out already, and I think it just continues to move in that direction. And even if more of it comes from a consumption basis than a seat-based license than we’re used to, investors shouldn’t be surprised by that. Software companies will extract their value somehow. However they can, they are going to do that. And to be fair to all these software companies, you cannot run your business without software. It’s obviously impossible, and so they’re going to get paid somehow.
Sekera: What’s it going to take for the marketplace to get comfortable with software companies? I mean, is there anything like management can do? Is there anything that they can show to the marketplace to prove to the marketplace: “We’re using AI. We’re adding economic value by doing so. We’re going to continue to be able to extract that economic value from our clients,” or is this just a matter of it’s just going to take years for this to play out before people really understand that software business model going forward in a new world of artificial intelligence?
Romanoff: If there’s a USD 64,000 question, this one is like the USD 128,000 question. There definitely are levers that all of these companies are—I’m just going to broaden it out for a second to talk about all of software—and so, what you see is management teams have been focused on margins and say, “Well, our growth is what it is right now. And so the one thing we can easily control is margins.”
So there’s been this push to drive margins up, and that is working for everyone, and investors don’t care about margins. And so then they’re saying, “Well, we’re going to drive stock-based compensation down.” And they’re doing that pretty well across the board, and investors don’t seem to care about that either. And so they’re buying back stock, so everyone is doing the same thing, and some of these companies are very aggressive, and some of them have new buyback programs for the first time ever, and no one seems to care about that.
Management is trying to do everything it can to send signals to investors. They’re having like special AI days to talk about the AI features in their portfolio, and none of it seems to be helping. And in the meantime, they’re continuing to put up good quarters, whether it’s Microsoft or someone else. The performance is generally good on a quarterly basis. And so I think the answer to this is you basically just need time for investors to see that, OK, it’s been, whatever, eight quarters now since this Claude tool was supposed to wipe out this particular kind of software, and that hasn’t happened. And basically there’s no inflection point in that business anyway. And so everything is still going pretty well. And maybe investors eventually just say, “OK, we’re done with the fear, and things can go back to normal now.”
Sekera: Getting away from AI a little bit, when you think about Microsoft, specifically and maybe just software in general, I mean, what keeps you up at night? What would be the biggest risks in your mind to your current base case?
Romanoff: I have spent some time saying I’m just not as worried about AI. And I don’t want that to be flippant because there is actually a risk that AI is bad for software. And so when I think of Microsoft, AI is kind of the biggest opportunity, and it’s the biggest risk. And thankfully for Microsoft, they’re kind of positioned to do well no matter which way the wind blows in this equation, but it’s definitely something to worry about, and it’s like the whole capex outpacing revenue growth by a significant margin—that keeps me up at night also. They’re doing stuff with finance leases, and so they’re trying to manage the depreciation of data center equipment, but there’s some financial engineering that’s happening. And so there’s a lot of unknowns, and you just can’t model it perfectly. And so I always tell people, “I guarantee you my model’s wrong, and I just don’t know where.” And that is definitely the case here.
But the one thing I feel pretty good about is—it may be wrong, but it’s not USD 415 a share wrong. It could be 10% wrong or something like that, but it’s just not as wrong.
Sekera: But again, that’s the nature of investing, whether it’s Microsoft or any other business that we cover. So again, that, I think, is why, when you’re looking at a stock like this, trading at that large of a margin of safety from the intrinsic valuation, that gives you plenty of cushion that the market sentiment becomes even more negative in the short term. Again, you’re still buying it at that big discount and, conversely, even if you’re wrong in a couple of different areas in your model, maybe the intrinsic valuation is a little bit lower than where it is modeling out to today. That’s still, especially for a company this large, at least in my mind, a significant discount where investors can get involved in that stock today.
Romanoff: Yeah. And that is definitely the way I see it. So it has a wide moat. All of its businesses are basically wide-moat. The management team itself is great. Satya has been great. They’re well positioned in pretty much everything that you’d care about. So from a cloud perspective, Azure, just long-term traditional workloads, it’s well positioned. AI, it’s well positioned with Azure just from housing all of the inference. They’ve done some of the things we talk about like their own model, but that’s—to backtrack for a second, that’s a risk too, right? Their own large language model generally doesn’t show up well in all of the discussions. Everyone talks about Claude and ChatGPT, and no one talks about MAI-1. Those aren’t words I’ve ever heard anyone say. And so to Microsoft’s point, if they can make their model more competitive or raise the awareness of it, that would probably help them as well.
Sekera: As much as we’ve talked about downside risks, there’s still plenty of upside risk here as well.
Romanoff: It’s a great company, and it is attractively valued. If you’re looking at multiples, it’s trading at loosely 20 times earnings for next year, which is, I think, just a modest premium to the market, which is pretty rare territory for a blue-chip, world-changing company to be in, that is positioned behind all of the right tailwinds. It just seems kind of unusual.
Sekera: All right. Last question here before we wrap things up. What’s next out there? We’ve been talking a lot about artificial intelligence. What are the other things that you’re thinking about and how they may be impacting software in general, technology in general, three years out, five years out? I mean, is it like quantum computing? Is that really going to be the next big wave that’s going to be coming, or is that still something that’s so far out into the future it’s not even on your radar yet.
Romanoff: Well, I tell you, what’s next for Microsoft? We are getting a little too far ahead, I would say, but quantum computing, I did write a report on it. That’s the next thing I get really excited talking about it. It’s just a fascinating topic, and Microsoft is right in the thick of all of that. It’s their single longest-running research product at about 20 years now. So just working on one thing that is not producing any revenue for you for that long is pretty amazing. And so I expect them to be driving that industry forward as well. They’ve made important advancements there, and I think there will be more to come. And when quantum computing breaks, it is going to rewrite all of technology. It’s not just going to be, “Oh, this is a cute little investment.” I mean, it is going to change everything we do now from hardware to chips to software.
Everything will have to be redone. It’s going to be remarkable. I can’t wait for it.
Sekera: Is that going to be a bigger catalyst than artificial intelligence has been?
Romanoff: Well, I think right now there probably would be more excitement around quantum if AI wasn’t a thing, but AI sucks all the oxygen out of the room, and so there doesn’t leave you room to get excited about anything else, but they’re still investing behind it because you don’t have to invest USD 180 billion, but AI also can be impacted by quantum, right? If you can get a large enough quantum chip that can do all of the inference processing, I mean the power of a quantum chip at scale would change AI as well. And so we’re not there yet, but it’s really exciting to think about.
Sekera: Well, that sounds like that’s going to be the topic for our next discussion then.
Romanoff: Yeah. I can’t wait for the next podcast to talk about quantum.
Sekera: All right. Well, that’s it for today’s bonus episode of The Morning Filter.
If you have ideas for future bonus episodes, including quantum, please send them to themorningfilter@morningstar.com. Viewers and listeners who’d like more information about Microsoft or any of the other stocks that we cover can visit morningstar.com or whichever Morningstar platform you use for more details. We hope you’ll join us every Monday for The Morning Filter podcast, which drops at 9 a.m. Eastern, 8 a.m. Central. Thank you very much, and hope you have a great day.
