Hey, good morning. Absolute pleasure to welcome everyone today to the Goldman Sachs Communacopia + Technology Conference. I'm Gabriela Borges. I cover software here at Goldman Sachs. Huge pleasure for me to welcome Amy Hood to the stage, CFO of Microsoft. Thank you for taking the time today.
Of course. Great to be here.
Amy, one of the more interesting conversations that we've had with investors over the past several months has been a little bit of a back to first principles thinking on software moats.
Yes.
If we think about the history of software companies, Microsoft has had a 50-year history of enduring software moats in the enterprise. Tell us a little bit about, in your own experience, how you would describe what has made Microsoft an enduring software company, such that there are moats that are relevant for 2026.
I think that has certainly been one of the questions, and I think thought processes people have gone through. I think what's made Microsoft enduring, but maybe more importantly, what makes it relevant for the next, we'll call it decade at least, is that we have, and have built, a very complete platform.
Whether it starts at the infra layer, up to the management layer of the platform, up to apps, and that's maybe been something people are used to. Now what you're seeing is you're adding intelligence into the platform, you're adding intelligence in what we call Work IQ, the intelligence of the Microsoft 365 work, and then on top of that, you're adding an agent layer.
I think in many ways, what you look and see is a company that sort of had a full stack and now has, I would call an even fuller stack as more and more of those pieces have come. We're lucky to be able to have also, I would say, a pretty sophisticated distribution network of partners, and a sales engine that I would argue is highly effective.
But even more so, I would say we also have longstanding customers who understand that trust, security, sovereignty, the protection of IP, their IP is a fundamental part of a platform company and the promise. I think we've been a platform company since even before I was there, which must seem like billions of years ago, given how long I've done this.
But what we've always said is that being a platform company meant being able to support and be relied upon to support and supply products with quality and with that trust. So I think when you put together our distribution network, really what we stand for, I would say, with customers in terms of reliability of our positions, and then I would say finally, a pretty sophisticated technical stack. I do feel like that is the way that I think about what defines enduring moats in the software business.
Maybe I'll pick up on this thread of IP and the trust and reliability that Microsoft has built up, and I really like how Satya Nadella had described it in his frontier ecosystems post. Tell us a little bit about how just in the last several months it feels like the zeitgeist is changing at the enterprise level about the need to protect IP, the need to think about, for example, multiple model strategies. How is that changing for your enterprise sales guys, that conversation on the front lines?
I do think it has changed a bit in the past months. I think the importance of supporting and being relied upon to support multiple models has always been something we have talked about. I think it has gotten more attention more recently in terms of, I think I could be off a little, but I think there is more than 11,000 models supported on the platform.
The importance of supporting everything from frontier models and open source all the way down to highly specialized models, some of which are things that our internal MAI team does work on.
I think what Satya is trying to say at least, and at least when I hear the explanation of a frontier ecosystem, is how important it is and always has been, as a platform company, I think this comes more naturally, to have enterprises be able to own their own continuous learning loop and what that means as being able.
If that happens in every enterprise, and whether you call that enterprise a small enterprise means big to a lot of people, but it really just means an entity. Whether that entity is small or the entity is very, very big, or whether that entity is a country, it is that if everybody owns and benefits from AI, then the distribution of the economic benefits will accrue across and be broad.
That is what we mean by a frontier, is that the frontier should in fact exist in a breadth way. That in fact grows the TAM available for everyone, but also, I think, reminds people that at the end, what the goal is to have things be better. That includes the economic outcomes to be accrued where the IP is created.
One of the questions that we get from industry a fair amount and from investors a fair amount is, okay, put on your prognostication hat and predict what the mix between frontier tokens and open source tokens is going to look like. The way that I would frame that question to you, because I am not sure that is the right question.
The question I would ask you is, what do you think the different types of SKUs look like? Meaning, investors will say Microsoft has all of this exposure to OpenAI in the backlog. Is it a zero-sum game, a trade-off? Maybe just address why the diversification of models is important and how you can support multiple models for different use cases.
Yeah. It's interesting. I understand the nature of what people are asking you to prognosticate on. I think that also is a challenging prognostication. I do think ultimately what people have always wanted, and I think the way I think about it more, is less that you ought to deliver the best value for the outcome that's needed.
If the goal is an outcome, and to get that outcome, you'd like to spend the least amount you can do to be able to gain that outcome, then that's the right way to think about delivery. I think that conversation's already happening more and more with customers as they're focused on the outcome. If the outcome can be, and have a positive ROI to them, then it's worth continuing to do that work. If it doesn't, then it's probably not.
The ability to be able to use Frontier Models for frontier problems, but really also to be able to use a mix of models to deliver the best value has been really where we've focused. I think actually that's not going to be unusual.
I think many people and enterprises at least that I talk to, are focused on how to make sure they're getting a high ROI, a high return on the outcome that they're looking to achieve. I think that sophistication continues to grow.
I think that's why we talk a lot about, a product called Microsoft Agent 365, because basically, the intent is that people have to be able to both measure and manage that spend profile to make sure that they're getting what they want.
We do the same in all our 1P apps, is to just make sure that if we have the ability to be able to deliver the right outcome at a lower cost, that we do that. I think that is one of the core promises of really Microsoft 365.
Absolutely. The other topic where ROI is top of mind is, of course, in the Azure business. Very consistently, I think it's been a number of quarters now, you've said, "Look, this is a supply-driven guide." Then you can tell in the commentary when you report earnings and there's 100 basis points upside, 200 basis upside, there is a measure of pride in being able to deliver that result-
Yeah
because of how hard your efficiency team
Yeah
is working. Talk to us a little bit about the kind of levers that you are pulling inter-quarter with that team to be able to drive efficiency out of every unit of CapEx.
Yeah, I think it is almost happening at, I would say even every layer of the stack, but it is almost at the shovel layer. You are really trying to make sure every quarter that you can take however many days, hours, weeks, months out of the time from when we can bring something online to when, I would say that a piece of hardware is revenue ready.
Our focus of the teams has been, and they have done a really nice improvement journey, especially in H2, which I think people have seen, being able to reduce some of our, what we call dock-to-live times over the course of a year by more than 50%. It has been really good work.
I would also say it happens at the software layers, in terms of being able to grind out more throughput, whether that is from the Azure platform teams or from the app layers above it. I think it is important people understand, I think of it almost like there is one big unit, and the more efficient I can make that thing, the more capable we are of converting it to revenue faster.
That tends to land in Azure in quarter. I think we have got a lot of work to continue to do on that front, and the teams are hard at work. But I do think it is layers. It is almost everywhere. I tell people we went through this journey on the Comm Cloud side, feels like forever ago. But it's happening much faster, and the work is happening much faster than it did the last time, which is good because demand is good. And so it'd be great if supply is good too.
The interesting derivative of supply and demand for FY 2027 is that you've guided to positive free cash flow. So talk to us a little bit about the mix between short-dated and longer-dated CapEx, because the language that you're using suggests that there's more flexibility in how you can lean in and out of those demand signals today than perhaps there was in 2025.
Yeah, I do think, and we've talked about that a bit, it's important to have that distinction. When you're talking about the long-lived assets, it's a very different thing. You're building construction spend or leases that move around a little bit. But on the short side, it's just demand-driven.
And for us, you kind of know how much space you have that's coming online, and you know you've got demand. So the estimate of CapEx is basically based on a fill rate. How quickly can we make that happen? Now, things that adjust that, of course, and it's why it's important that it's short-term is you're grinding away on efficiencies.
So efficiencies bring, quote-unquote, "the supply needed to meet the demand" brings it down. But then you're looking at your demand signal, and it's pretty healthy. So you can probably get more out of every dollar you spend.
I think that's really the focus that we've got. And short-term does allow a lot of flexibility in that. But right now, I would say the focus is far more on how quickly we can get supply revenue ready than it is trying to figure out if I have $1 billion of CapEx here or there to move. It's much more how quickly can we get it into the system.
There is a school of thought that while the demand signals could potentially get even greater over the next 12 months as we think about agentic architectures, world models, et cetera. How do you respond to the inverse of the question, which is should you be being even more aggressive with a little bit of an arms race dynamic with you and your peers on investment?
Yeah. I think right now, it's a funny question. Over a 12-month period, very little can get-
Fair, yeah.
built and come online in the next 12 months. I think what people are focused on, and what we're all focused on, is making sure that you're making the right long-term investments into land and power, and making sure that you've got a lot of flexibility over the long term in terms of being able to build and be aggressive if the demand signals warrant that.
On the short term, again, it's less about whether you're aggressive or not. It's more like you've got what you got, and what you should work on is pulling it in as fast as you can. Because at that point, it's not going to get built in that timeframe. So, I think that's probably why I separate them a little bit when people ask that. It doesn't really correlate in terms of sort of matching that dynamic.
Maybe give us the longer-term perspective on finding and securing supply, particularly given we are going into the midterms. There is a lot of conversation around data centers and finding all of the pieces to put together. How do you think about the things that you can control in an environment like that?
Well, I think we get asked about that question in lots of ways. We have talked about this. We have done long-term agreements, LTAs, with a lot of the supply chain to make sure we have, I guess, I am not sure the right term is matching sets, full sets of things. I think that is the best way to say it.
A lot of work has been done over the past 12 months to make sure that there has been significant investment to make that happen. I think people probably did not fully appreciate that it runs not just one piece of the kit, it is the whole thing, and now I think that is better understood by the market and by others.
I do also think that in terms of making sure you have got and can keep doing building projects is making sure a lot of local focus on positive impact in communities is a real need for people to make sure that we are all focused on and the positive outcomes that communities can see.
I think that end to end, you have to sort of make investments to think about kind of ending where we started in some ways, which is you need to have a distribution of the outcomes. I think we have been quite focused on that when it comes to making sure we can continue to deliver the plans that the demand requires.
One other supply question that I think is relevant here is on your internal semiconductor strategy. We were so excited to see the benchmarks from Maia 200. Talk to us a little bit about how Maia 200, Maia 300 could scale over time, and how much of an economic leverage does that give you when you can control more of the semiconductor stack in-house?
Yeah. I think people sometimes forget when we went through the CPUs and the journey through last time in the cloud. We started really with one provider. We diversified to two. We built our own piece, the Cobalt unit. Now that will be in, I think, 25 data centers over the course of the year.
We have seen diversity in that. It has been part of what has helped over time with margin improvement over the course of a decade. That does not mean we only rely on 1P. It means we have relied on the market and the ecosystem to make sure that we have, I think, a competitive platform to make sure. I think we have been very consistent. You will see the same from us. As I think others have said, I don't know that we would have a different strategy.
I think the plan is that you want to make sure you have great price-performance. It is not just price, and it is not just performance. We need both of those things, and when you tend to have that, and whether that is a 1P part or a 3P part from multiple partners, I think that will and has been and will remain the plan. That being said, we are excited by the Maia benchmarks we released. Obviously, the price per fund, that would be healthy. We look forward to being able to scale it.
Fantastic. Okay. Let me ask you about the capital structure framework, specifically around how you think about potentially going to the debt markets, potentially going to the equity markets. Is there a scenario where Microsoft's optimal balance sheet actually is one that holds more debt?
We have accessed the debt markets multiple times over the period at least that I have been the CFO for various reasons. I feel like access to the debt market to us, I feel like we have a lot of access.
We have probably a pretty fair price. For us, it has always been about if we need it and the ROIC is very high, that makes good sense, then we would be happy to go to the debt market. It is not that I sit and say, "Gosh, don't go to the debt market."
It is more just if we need it and the ROIC is there, you should expect us to use the capital and find the capital we need to deliver outsized results. I have thought a lot about it being a big deal for us to access the debt market, given we have used it numerous times in the past when it made sense.
I want to come back to your earlier comments on the enterprise and the applications business. As we think about some of the really big numbers tied to the AI cycle, whether it is on the CapEx side or the Frontier Model side, I would love to talk a little bit about this idea that the software TAM gets bigger, and more specifically because instead of selling into the labor market, you are actually selling potentially a unit of labor or a unit of work.
It becomes much more expansive. Tell us a little bit about how that shapes your pricing strategy. We get asked this question a lot, where does the new budget come from to support all of this new software spending? How do you think about pricing?
Yeah. I think you have started to see. Let me first say, I think with the changes basically in the coding market, and I think we are one of the later ones to make that change actually, to more usage-based billing. Then with the addition in M365 of consumption-based pricing as well as user-based pricing.
I think what you are seeing is the shift to a little bit of what we talked about before. It is that you want outcome-based processes. If there is a unit of work or a unit of outcome, then the price to pay to get that outcome is probably best done with consumption pricing.
We will do and make. I do think the user license still is important because companies still do like predictability. In fact, all budgeters like myself would prefer predictability above almost all other things.
The user license has always provided a way to have a level of predictability while gaining access. That remains an incredibly important point. Our ability to continue to deliver value in the user license, while probably delivering more and more value over time, quite frankly, in that license is important. It is an important and part of the process.
But the transition to also have outcome-based billing, I think will be a key part of a model that says if there is expansive TAM, it is because you have delivered an outcome at a higher value than it was prior possible or not possible to deliver before. I think that is why we do view it as expansive as opposed to, I think some people when they ask the question, it seems like a replacement. I think it is expansive.
I think if new outcomes are possible, I think we often get asked about it as a cost line item. It also has to be, going back to the very first comments we were talking about, it has to be a growth concept. That new businesses are built, new outcomes are created, new revenue is generated because of the outcomes.
I think when we focus on it only as a cost lever, we are not probably thinking about that correctly. I do feel like if this is done the way I suspect we are seeing, it is that it will both grow revenue for industries at large and be able to help grow the TAM.
To your point, Microsoft recently made the pricing change with GitHub.
Yeah.
Tell us a little bit about what you learned from making that change, customer willingness to pay, and then how do you apply that to the rest of the portfolio, and we can bring Copilot into the conversation.
When you are late to making a change, I am not sure you learn anything other than you were late. I do think you have to step back and say, "Wow, if we needed a point that says if you add value and change the productivity of an individual, are they willing to pay for that?" They are. We are not the only company showing that in our results.
So if you have a high-quality product, if that product's user feels that the value they are getting exceeds or far exceeds, depending on their view, the amount they are paying, they will continue to use the product knowing that it costs more when they do. I do think there is a lesson in that. There is also a lesson that people do value and like predictability, right?
I think we get the totality of that feedback, and including, I think it is also what makes it incredibly clear that what we call Agent 365, what I would call SynOps, basically, and controlling and measurability, observability will be an incredibly important part of delivering solutions to companies of all sizes to make sure that they feel good about that.
It leads nicely into a little bit of the rock and a hard place question we sometimes get on Copilot, which is if Copilot's amazing, customers are going to use it more.
Yeah.
Up until recently, and maybe still now, you have a fixed price point for Copilot. That would hurt Microsoft on the gross margin side. Now, the alternative is way worse, which is the product isn't good enough, so it doesn't get usage. I think what people miss in that two choices is all of the efficiency improvements you can make on the back end with Copilot and the potential to go to usage-based pricing. Maybe talk about that side of the coin.
Yeah. I think what we've worked hard on, this has been true, I know people sometimes forget. This was actually true before we had AI as well. People stored more documents. People wrote more emails. People kept more things. It's always been true of Microsoft 365 that because it is the repository for really the world's work, cost structures went up, and the offset to that was efficiency and productivity improvements.
I do feel like that same logic is true here. We do and have said there'll be limited things in the USL, in the user license, but increasing usage, which we have seen, is a good thing. The increasing productivity that we've seen to offset that has also been a good thing. We will continue to see that.
It's also why we've said making sure we're using the right model for the right job has been incredibly important in Microsoft 365 too and will continue to be. Then, of course, the more users of Copilot you have and the more engaged they are, the more likely they are to use tools like Copilot Cowork, which is a consumption-based model, and other tools that will be more usage-based.
I do think that it is a bit of a short-sighted question in that we do have, and have always had, a focus on making sure that we could deliver good value to customers that's predictable, otherwise known as the USL, the user license, and the opportunity for usage that needs beyond that to be charged for in a consumption and more usage model. The beginning point for building consumption growth will be having Copilot deployed and used and the quality to remain high.
Absolutely. Talk to us about the recent decision to re-segment.
I wanted to make sure you had something to do in August. No. I think this has been a change. We've been talking differently about the company, I think, really over the past year. I think really as we went through the second half of the year, how we were managing the company and how we were running the company really has changed significantly.
When that happens, it's important to make sure that how we talk and show our results internally should reflect that change. We've been working on this for a little bit of time. I do think when we finally talked about it feels like forever ago, but I think it was last week.
Like a couple of days, yeah.
I do think it really makes far more sense, especially as coding and security and knowledge work come together. We've been talking about that. Even our products are starting to look more and more like that. Having that be under the single banner really of Microsoft 365 makes sense.
The more and more integrated the stack has become in terms of delivery of those units means that it's very hard for us to think about agents and apps and infra as being separate. We don't run them that way. We don't run the company that way. Our teams no longer operate that way. We've got teams called one thing that work on a different thing.
It just got to the point where, although I really don't like having to change things that don't have immediate ROI in terms of work product, because that's a lot of work to do it and system stuff, but it was the right thing to do, and I think it far better matches the way both Satya talks about the company, but probably also the reality of where the markets are too.
I want to end here on something that you started with, which is the sophistication of Microsoft's technical stack. When those of us who spend time in the developer ecosystem or even on Twitter for a couple of hours a day, you sort of see the potential of what really sophisticated agents can do.
Then for the majority of us, we turn around, we come back to our work desks, and our ability to use the sophistication of the agents is somewhat throttled because of all of the reasons that we can talk about in regulated large companies. Microsoft, to us, seems somewhat uniquely positioned to help bridge that gap.
Yeah.
Talk to us about what that looks like in practice. Maybe Agentic Harness is part of the conversation. Maybe Copilot Cowork is part of the conversation. But we'd just love to hear from you because you do have a unique position.
Yeah
to be able to bridge the gap.
No, I think there is a couple of things. I think a lot goes into when we watch, then we come back to our desk and say, "Well, why can't I work exactly that way?" I do think there is a couple of things which we are working hard, and I do think we are in somewhat of a unique position to deliver on.
Number one, we have, as I said, most of the Work IQ that exists in our platform, making that accessible, reachable. You are right, part of it is an Agentic Harness. Part of it is making sure it is multi-modeled so you always have access to the best. Part of it is making sure you have access in a secure and thoughtful way to your own tenant, meaning company IP.
Part of it is the investment we have made in our new Frontier Model and make sure we have support for companies looking to do that and make that transition and teach their people. Part of it is making sure it is easy for developers to be able to build apps that extend it. I think we feel really good about the progress that we have made on all those fronts.
Then, of course, I think you have to make CFOs comfortable that the value in the budget is worth it. I call that thing Microsoft Agent 365. Again, it makes me laugh. I think it is because I have to manage budgets.
I think when you look at all those end-to-end pieces, and you say what has always made Microsoft unique, it has been that we have been able to help companies deliver that productivity output to their people.
I feel good that we are able to do that again in an age of, and I think that is why we call it the segmentis, in the case where agents and apps and humans will work together. I feel like Microsoft 365 is the safest and most trusted place for that to happen.
Absolutely. Amy, thank you so much.
Thanks, everybody.
Join me in thanking Amy for her time.