Welcome, everyone. I'm Amy Hood, the Chief Financial Officer here at Microsoft, and I'm here with Kevin Scott, Our Executive Vice President of AI and Our Chief Technology Officer. I just wanted to welcome Kevin to talk directly with investors, because given the significance of the next generation AI and our leadership position in this AI era, we wanted to spend some time talking more about this after the myriad of questions that we've gotten, both in meetings and from investors on conference calls. I think this can help frame the growth opportunity ahead, really speaking with what I consider to be the person in charge of running the show here. With that, let's jump right in.
The first question, and I get it gets asked this way to me, Kevin, so I'll ask it the same way to you, which is: Listen, we've seen multiple hype cycles in tech, crypto, metaverse, so how transformative will next generation AI really be?
There have been multiple hype cycles in AI. When I was in grad school, we were in what is called an AI winter, because there have been so many boom and bust cycles that, like, there's a Wikipedia page describing what the bust look like. I will say, like, I've been doing machine learning and AI work since I stopped being a computer science professor in 2003. We've been on this really interesting exponential trajectory over the 20 years, and I think, you know, we've kind of been in the flat part of the curve for most of the 20 years.
Like, we really started to see an inflection, like, starting first in 2012 with, you know, this realization that we finally had all of the compute and all of the data that we needed to really extract performance out of deep neural networks. Like, what we've seen just in a remarkable way, and it's one of the reasons why we've done our investment with OpenAI, and why we've built all of this supercomputing infrastructure, and why we have been able to build this new category of applications that we're calling Copilots, is the models over the past handful of years have become to behave really like a platform.
They are things where you're not training them and conditioning them for a single use, but, like, they're big pieces of infrastructure that you build, and you can use across, like, many, many, many, many things. You know, if I look at how technology cycles have gone just in my lifetime, there's been personal computing, there's been the internet, and there's been the mobile revolution. I think this is sort of the next thing on that progression of the evolution of technology, where it really changes the way that we build software. The way that I've been talking about it to folks is, you know, just to try to describe how profound it actually is, we've been writing computer programs and using computing technology the same way for 180 years.
Ada Lovelace wrote the first program 180 years ago. Like, we've been doing things more or less the same way since then, with more powerful computing devices and more powerful software abstractions. Basically, if a human being wants to get a computing device to do something for them, you either have to be a skilled programmer or you have to get a skilled programmer to anticipate your needs so that you can run the program that they've written.
Yeah.
Now we have AI tools that anyone can use to express a thing that you want to accomplish, and in a very flexible way, you can get the computer to do the thing for you. Like, and we're not done on this progression yet, just in terms of, you know, how much more capable the technology is going to get.
This is an interesting one, Kevin, too, because I would say from my perspective in talking to customers, it's showing up in every conversation, and they aren't asking about the hype. It's very different. They're asking, "Hey, listen, how quickly..." Based on really what you just said, is that it makes it so much more approachable for problem-solving, real problems that people in every seat in an organization feel and see opportunity in. You know, I think they're more asking when they can deploy it, far more than, you know, "Gosh, is this a real thing or not?
Yeah.
You know, it certainly resonates. Given this broad sort of excitement, what should investors know about our AI strategy, and what differentiates that, really? I think the amount of news, sheer news, that's come out, in the past, it feels like a very long time, but it's really only been three or four months.
Yeah.
Maybe you can help distill that difference of what our plan is and what makes that different.
Yeah, I mean, I think we've had a strong point of view about where we thought this AI platform was trending for the past five solid years now. Like, we started a transformation inside of Microsoft about a year before we even did the partnership with OpenAI. Part of that is, like, we really believe that these models are becoming platforms. Like, we believe that we're gonna have a small handful of them, and that we're gonna use them as the foundation for all of the software that we're creating. We have rethought how we build all of our low-level infrastructure from the way that we design and deploy data centers to, like, what compute and networking looks like, you know, all the way up to, like, what the software development environment looks like.
We talked about a little bit of that, this year at Microsoft Build, like this notion of a Copilot Stack, which is, you know, just as distinctive, I think, as the, you know, the LAMP stack was at the beginning of the.
Yes.
Internet revolution. You know, it is yeah, the thing that I will say about us is, like, we just have this one through line that we've been investing on the past five years. The reason that you see all of this news is because, like, we've been focused in the way that we built the platforms on top of which the applications are getting built and delivered. There's just no way we would have been able to produce as many of these Copilot applications, so Microsoft 365 Copilot, Bing Chat, ChatGPT, you know. Like, I'm probably gonna say a whole bunch of Copilots even that haven't launched, like if I, you know, if I let myself get carried away.
Like, there's just no way that we would have been able to deliver all of that if there wasn't a real platform underlying what it was we're doing.
Right.
Because it would have been too expensive, too painful, too time-consuming to do each one of those as an independent thing.
I think what's interesting about how you talked about it, Kevin, is that because this platform is really, it's who we are, it's how we approach every problem as a company, people should think about our approach to AI being incredibly similar to that.
Yeah.
That I thought, and I would encourage investors to go back and look at the Build keynotes, because this key aspect is that there's every layer of the platform.
Yeah.
... will be built with this in mind. I think in particular, certain things really resonated. Microsoft Fabric is an example, right?
Yes.
It really resonated and broke through about what's really different. I think for me, it's also that, you know, it's hard sometimes to capture what these large technological transformations happen. You know, I think you and I would say this is sort of a generational moment, and in terms of watching the innovation that's showing up across the world, across companies of all sizes. I think that's probably what makes it most exciting, and in particular, for, I think, developers last week. I mean, it's certainly a changing perspective. I always tell people, you think, "Oh, about a platform?" Well, then this audience always asks me, "Well, where does revenue show up?" It really comes from two places, right? There's the tools and services that customers will use, right, to build AI apps and services.
They'll run the Azure OpenAI APIs, they'll run our Azure infrastructure, and then, to your point, it's gonna be built into every Microsoft cloud-
Yeah.
solution. Everyone.
Yeah.
Whether it's announced or not announced, whether it's named Copilot or not, it'll be that same aspect. There'll be something that shows up that's AI-enabled. I think that's the reason this will be our sort of the next generation AI business will be the fastest growing $10 billion business in our history. I think I have that confidence.
Yeah.
Because of the energy we're seeing.
Yeah, 100%. Like, the other thing, too, I will say, is that because it really is a very general platform, like, we have lots of different ways that, $10 billion of, you know, AI is gonna, like, first show up. There is, like, all of the people who want to come use our infrastructure.
Yes.
... they're training their own models, whether they are, like, running an open source model they've got-.
Yeah.
or whether they are, like, making API calls into, like, one of the big frontier models that we've built with OpenAI. You know, it's like all of these products, like, each of which has a slightly different potential for business model. Like, you know, the way that, you know, the way that we are-
Yeah.
going to be able to, you know, just sort of monetize all of this, I think is really interesting. And, like, I'm excited for the developers even.
Yes.
on top of this, because, like, we're probably gonna get some business models that, you know, haven't existed before in history. Like, we're gonna have business model innovation happening as well.
Yeah. That's really been a hallmark of almost any big technology shift, is you both have technical transformation, platform transformation, and business model innovation. I think it is always exciting. Having lived through maybe the first transition to the cloud that you and I were around for, the second one, I think in some ways, people, investors ask me, "Gosh, well, I'm so worried about all the money it takes to invest in these." I try to remind people, as a company, we're in a very different spot. We're leading this cycle. We understand how to run at-scale cloud businesses, and you said something really important earlier, which is we've built a single platform from the ground up effectively, that's used consistently under every service.
That's not where, frankly, we were a decade ago in terms of reusability and fungibility. I do think, of course, we'll see scaled CapEx, and there'll be COGS growth associated with that. We'll also be in a far better spot than we started from, and the ability to leverage that and the confidence we're feeling from customers on demand. I feel quite optimistic about that.
Yeah, I mean, the other thing, too, that I will say on CapEx is like, you know, it's CapEx that we're deploying, not because we feel like we've got to go spend it to catch up, but, like, it's CapEx we're deploying because we feel like there's such a gigantic opportunity ahead of us.
Yeah.
That there's gonna be such a big return on that invested capital.
I think that for you and I, puts our head in a different spot in terms of talking to customers and what we see from developers, too.
Yeah.
Another question I get a lot, from both investors and customers is our view on, I think, really a topic that we read a lot: ethics, regulation, the overall safety when it comes to what AI is really capable of. What can you share on these topics and the work we've already done with our responsible AI principles? Because I know this is one of the key components that you think about every day, in the suite.
Yeah. Yeah, it's, you know, one of those things where I think it's actually is another strategic advantage in how it is that we are building the platform. Like, I think you have to think about safety and responsibility-.
Yeah.
... and security of these systems as, like, one of the, you know, one of the necessary bits of the platform itself. Like, you just can't have a platform without it, for, like, just a whole number of reasons. Like, you know, maybe starting with customer trust. Like, you have to have trustworthy platform, trustworthy infrastructure...
Mm-hmm.
trustworthy applications that you're building.
Yeah.
we've been looking at this whole discipline of responsible AI for many years now. I think before I even joined Microsoft...
Mm-hmm.
Satya charged a bunch of people in Microsoft Research and then our legal team with building a real AI ethics and, you know, responsible AI practice. You know, we have been investing over the years on everything. Like, you know, what do we need to put in the legal team? Like, what do we need to do research on? Like, what infrastructure do we need to build to help our product teams make sure that they can test and assess their systems for, you know, safety and responsibility?
Yeah.
You know, one of the anecdotes I will share is like, you know, we had, you know, like, all of this press around, Bing Chat, you know, the infamous Kevin Roose article. You know, the interesting thing there is, like, just even by...
Yeah.
... we are learning so much about how not just to build things with safety and responsibility, because, like we-
Yeah.
... we believe that everything that we've launched has adhered very strictly to our publicly available Responsible AI Standard that's in its second version right now. Like, we're even sort of learning, like, how to use those safety systems to tune things to meet the preferences and expectations of users. You know, the, you know, the Kevin Roose thing was basically, I won't even call it fixed. Like, we adjusted the product within a couple of hours of.
Yeah.
you know, of that story breaking. The only reason that we were able to make the change as quickly as we did is, like, we had all of the safety infrastructure-
Yeah.
that we had built, that allowed us to make sure that the change that we were making, on that particular product, to, you know, conform it to the preferences of users and to make sure that, you know, people weren't, you know, using the product in ways that we hadn't intended
That's right.
... was because we had all of this stuff that we had built to, like, let us do that assessment, to make sure that the fix didn't cause more problems than it, than it was addressing.
I think important in what you said is that, listen, we're not waiting for regulation to take this proactive approach to responsibility, and I think, going through the process of that. I've heard you also mention that the regulation is needed, and we've said that.
Yeah.
... as a company. How do you talk about that in your conversations?
Well, I think if you look at any sort of technology that has ubiquitous impact on society, like, there's always regulation. Like, my favorite metaphor is electricity. You can sort of think of these AI systems, I think, and like I hope in time, like, they will have similar sorts of impact, where, you know, 10, 15 years from now, you will just have a difficult time imagining how the world could function without them. If we get to that point, like you must expect that there's gonna be regulation. Like, you'll have regulation on the generation bit. Like, the same way that we have regulation on the power generators and the turbines and the generation facilities, like, we will have that for AI.
Like, there will probably be regulation on distribution, as there is with electricity, and there will be regulation on, like, the sort of endpoint of the technology, like, where it actually gets delivered into, you know, like into a product that a user is touching. I don't know exactly what that regulation will look like. You know, Brad did a really brilliant job, I think, a couple of weeks ago, articulating, like, what our initial point of view is. We also believe that, you know, there's gonna be a lot of debate and public conversation about what it is the regulation really needs to be. Like, I...
It would be really suspicious to me, honestly, and it would be a sign that the technology maybe isn't as impactful as we think it is, if you believe that it doesn't need regulating.
I think that's an interesting way of thinking about it. The other thing I get asked a lot of from customers is a focus on privacy.
Yeah.
I think this is not a surprise, right? As people learn about the technology, and obviously, just so people are clear, they're not... Customer data belongs to the customer, just like it does today, that applies in the world of AI as well.
Yeah.
I can use the Azure OpenAI example. You know, customer data remains within the tenant boundaries. The LLM is fine-tuned on their data, remains with their data.
Yeah.
is not applied to the master, LLM. I think that's just important, and we keep reiterating that with customers. Are you hearing the same sort of questions as you-
Yeah.
talk to people?
Yeah, absolutely, and I think what you just said is, like, the thing that we have to make sure everyone really understands.
Yeah.
Like, we're not, you know, we're not using people's data to improve the foundation models themselves. Yeah, all of the covenants that we have with everyone around their data, you know, with respect to their tenant, are absolutely respected, you know, with this technology. Like, whether it's an OpenAI model or something that we built, and like it's, like we're super diligent. Again, it's, you know, one of those things where we're lucky to have spent the past, you know, 15 plus years-
Yes.
... building all of that infrastructure because, like, we've been able to rely on it as we've been deploying this new AI technology.
Absolutely. Now I wanna move on to, I would say, the second most popular question I get in investor calls, at least, is regarding the OpenAI partnership. Why did OpenAI choose us? How did that partnership get started? People are worried that we don't own the IP that's really required in this area to be considered a leader. Is that a valid concern? Those are the two big questions for you, Kevin, in this one.
Yeah. Let me start with the first one. I think, you know, the first and foremost, like, we've been able to do a bunch of really amazing work with OpenAI, and, like, the work that we've done is a testament to how well aligned we've been from the start. The initial intention when we did the first deal in 2019 was we wanted a third party, like someone that was outside of Microsoft, in addition to our own research and engineering teams, to be able to come in with an incredibly high level of ambition for what they were trying to do with AI.
They would put pressure on all of our infrastructure and systems software so that we could make sure that we were building the best possible platform for training and serving these new, large-scale frontier models, as we're calling them now. Like, I think if anything, like, that has worked incredibly well. Like, we've been able to design this infrastructure along with OpenAI. Like, we very quickly after the partnership, built our first AI supercomputer. That was the thing that GPT-3 was trained on. In addition to them using our infrastructure, like, we've also been able to partner on using their models in incredibly creative ways to power our products. The first of which was GitHub Copilot.
like, we had deployed GPT-3 in, like, a whole bunch of different ways, like, relatively silently in a bunch of products. GitHub Copilot was the first generative AI first product that I think anyone had really built, you know, where the generative AI was just front and center and allowed us to do a thing that people thought five seconds before we launched it, was actually impossible.
Yes.
Which is a little bit of the flavor of, like, a bunch of these applications. Like, you deliver them, and, like, you go from, you know, in people's mind, like, that was impossible to, like, "Wow, I can't believe it." Yeah, so that has been the nature of our partnership.
Yeah.
You know, we have the way to understand their IP rights is, we have a very broad license to all of the OpenAI IP for the term of the deal, and we work incredibly closely together on all parts of the stack. Like, they have pushed us in incredibly helpful ways to make sure that our infrastructure and systems software are world-class. I think you see that right now in just the number of people.
Yeah.
choose our cloud to do their own high ambition things. It has definitely helped us out in incredible ways at just sort of the scale and pace of delivery that we have. You know, we obviously benefit tremendously as well from OpenAI, who, like, I think is one of the smartest, you know, smartest group of AI researchers on the planet. We also complement the things that they're doing as well. Like, we, you know, we have, like, a really, really fluid partnership, like, where, like, we're very aligned on this platform vision, and we're very aligned on safety and responsibility.
Yeah.
Like, this real, you know, mission and, you know, incredible obligation that we have to doing it right.
Yeah.
... for society. Like we just are pretty flexible about how we help one another.
I think for a lot of people, this has been a little bit what's hard to capture, is that you can be in this relationship with a company and a partnership, also have your own internal research capabilities, and we can partner with others. It's just a unique partnership, fundamentally, I think, built on trust and mission alignment, which is always an important thing. Just to reiterate, I think the two key points you made is, you know, when Kevin talked about having a broad, perpetual license to all the OpenAI IP developed through the term of this partnership, that means even if the partnership were to end, we would still have those license rights to all the IP up until AGI.
The other thing that's critical is that, you know, I think, and Kevin leads a lot of these conversations, as does our legal team and our OpenAI ethics team, I'm sorry, our AI ethics team here. You know, we share the same values for responsibility and the safe deployment of this IP. I can't tell you how foundational that is to being able to make progress.
Yeah.
Both of us can independently make this technology available to our respective users and developers. You're seeing that happen. I think that's good for everyone.
Yeah.
Really keeps the innovative spirit. You know, this AI capability is built into the Azure foundation. I think, it's sort of a long way of Kevin and I are saying that we feel very good about the position we're in, both with the partnership and with access to the IP over the long term.
Yeah.
Let's go back to, I think, what a lot of people always ask me is about, it's like: "Hey, listen, are you really hearing from customers that they're gonna deploy this? What are the use cases we're seeing today? If I was to open my mind up, what are the use cases we'd see tomorrow?" I would say, maybe I'll say some of the things I'm hearing, Kevin. I think you probably hear even more, especially longer term, what's possible. I think we hosted CEOs on campus, not so long ago, Kevin, you hosted a conversation, I met one-on-one, but it was fascinating.
I mean, it was every industry, whether it's healthcare, where we talk about Epic with medical records, whether it was manufacturing and talking about how to detect and describe problems between the plants and headquarters. Whether it's education and having personalized tutors or retail or developers...
Yeah.
To your point, I think, it's coming almost... I would describe it's a global conversation, and it's a by industry conversation. What are you thinking in terms of use cases that are possible? I think you're closer to some of those industry cases than I am, for sure.
Yeah, I think very broadly speaking, this particular wave of technology transformation is going to be all about really driving productivity in places where people are doing cognitive work. This is, you know, this is just sort of.
It is.
overly wordy way of saying, you know, white-collar work. Like, you know, your job, my job, like, you know, anyone who is, you know, really-.
Yeah.
you know, using their mind to, like, do something. Like, what I think is gonna be enormously obvious in the coming months is, like, there's a huge amount of stuff that we all do in our cognitive work that is, you know, sort of routine, repetitive, like, we're probably not super excited about it ourselves. you know, it's, you know, the annoying, stressful parts of our job.
Yes.
Like, I think almost every one of those things, like, gets impacted in the short term by these tools. You know, the way that we're thinking about them is, like, this choice of the term "Copilot" to describe this new type of software is, like, a really deliberate choice because we believe that there are going to be lots of these Copilots. That they are going to be, you know, some of them very general, like, Bing Chat or ChatGPT, but, like, some of them, like GitHub Copilot, like, relatively focused on a particular type of cognitive work. Like, I think whether it's software development or it's marketing, it's customer support-
Yeah.
It's, you know, like billing, you know, doing prior auth for insurance.
Yeah.
Like, you know, just sort of pick your thing.
Yes.
you know, like, I know that I encounter these things every day, multiple times, where you're like: "I can't believe we're still doing things this way." Like, now is the moment where, like, you know, you sort of finally, I think, get past the, like, I can't believe we're still doing things this way.
Yes.
You get to go do them a different way.
Yeah.
I think in the longer term, like, you know, the real prize here are, like, these more generally capable AI systems that can really go solve big problems. you know, like, the thing that I talk about with folks, and, like, I don't know when this will happen, but it's sort of the aspirational thing. Like, if you had an AI that had scientific expertise and the ability to go think about a problem for a really long time, and you said: "Hey, I wanna go spend $100 million worth of compute on, you know, with this-
Mm-hmm.
With this AI expert to go cure pancreatic cancer in all of its forms, like, that would be the easiest $100 million that human beings had ever gone to spend. Like, the amount of pain and suffering and, like, economic losses and just general misery that, you know, some of these diseases cause. Like, that's so the aspiration for the future. Like, can you have systems that can help us go solve problems that we can't solve right now, even using the best tools that we have?
Yeah.
like, what is that better tool that helps us go solve those really nasty problems?
I think that's why, in so many ways, this time period where you can both be and have real concerns about the power of the technology, but far more than that, so much optimism about
Yeah.
the possibility. I think that those two things can coexist, and I think, frankly, always have in large transitions in history. Just a few things. First, Kevin, thank you so much for joining us today. In some ways, I think there's no better person to have to tell investors what our vision is, why it's different, than the person responsible for making sure our vision and strategy gets executed to the very best of our capability. Thank you both for taking on that large mantle for us and for joining us today. Thanks to everybody for watching and getting a bit more educated in terms of what's different about our approach.
I will say, we've waited a long time for me to get more productive here at the company, Kevin, so maybe this is the final secret that gets me there and increases this CFO's productivity at least. Thanks, everyone, and I look forward to delivering, really, Kevin-
Yeah.
... both for our customers, for developers, frankly, and for a society that I think is ripe for the adoption of this technology and our leadership position. Thank you, and we certainly remain committed to leading this wave.
All right. Thank you, Amy.