Karl Keirstead: OK, thank you, everybody. I'm honored to lead this keynote with Alysa Taylor of Microsoft. I wanted to just thank the whole Microsoft team. You've been fantastic partners at the event this year. I don't think it would surprise anybody that Microsoft and Nvidia are the two most requested companies for one-on-one meetings. I think that says a lot about where we are in the tech curve. So thank you, Brett, Kendra, Mary and the team for making yourself available for so many one-on-ones, it's fabulous; and Alysa for making the trip to Scottsdale to chat AI and a number of other topics with us.

Alysa Taylor: It's a pleasure to be here. Thank you.

Karl Keirstead: Oh, great. Good. Alysa, do you want to take a minute and share with the group your role at Microsoft?

Alysa Taylor: Absolutely. So I'm responsible for our Azure business, so that comprises all of our AI services, as well as our data infrastructure, digital app and innovation and infrastructure. So that's the entirety of the Azure portfolio. And then I also have responsibility for our global industry, so everything from our regulated to non-regulated industries, including sustainability. And that's on the go-to-market side, so bringing all of the things that we build, buy and partner, how do we then construct that in a way to bring them to markets.

Karl Keirstead: If this was a two-hour keynote, we could fill it up with all of your expertise. But we'll try to condense it to 28 minutes.

So let's hit on a couple of hot topics to start. So I think this might be Microsoft's first occasion to at least sit down with the investor community more formally post- the drama of a couple weeks ago with OpenAI. So understanding that you're probably somewhat limited in what you can say, I'd love if you could share some perspectives. I can share with you that some of the questions we received had to do with IP ownership. It had to do with the extent to which Microsoft might be better off, assuming they are, given how things settled down. But I'd love your perspective, so if you don't mind?

Alysa Taylor: So there was drama with OpenAI? But yes. No, it's been an interesting week. You know, what I would say is we are committed to our partnership with OpenAI. We are excited to see that Sam has been reinstated as the CEO. And we have a great roadmap of innovation ahead, and we have perpetual use rights for the IP with OpenAI, and we have everything we need to fulfill on our roadmap of innovation. So business as usual.

Karl Keirstead: OK, nice to hear. Let's talk about another big event recently, and that is I think you played a central role with Scott Guthrie and the team at Ignite. So that was a big event. We had one of our team members there to talk with customers and partners and learn. But what were the couple of key takes for you coming out of Ignite?

Alysa Taylor: It was a really exciting event for us because we introduced a bunch of innovation, and I would say across the entire stack of the Microsoft Cloud. And so some of the highlights that I would bring forward, our big announcement around silicon I think was a pretty prevalent part of the event, so it's our first entry into first-party silicon. We introduced Azure Maya, which is our AI accelerator. We also introduced Azure Cobalt, which is our cloud-native CPU. But that really is part of our larger what we call silicon-to-systems, so we also had -- they announced the generally availability of Azure Boost, which is the technology that allows you to offload networking and security onto purpose-built hardware, so that improves performance; also our hollow-fiber core, which improves networking. So lots of infrastructure.

And then we also announced the general availability of Microsoft Fabric, which is our new analytical data platform. This is a very important key asset in our AI services because it allows organizations to be able to take a bunch of -- a lot of disparate data from either on-prem, different data sources, different cloud providers, Snowflake, GCP, be able to bring that data into an integrated data environment called OneLink. And then we have native AI services that are built into Fabric. So really exciting announcement there.

And then, when we think about the tooling aspect, so we also brought the Azure AI studio into Preview, and that includes all of our responsible AI tooling, like content safety, our model-as-a-service so that you can actually govern and reason over third-party models. And then Azure Arc, which is our central control plane that allows an organization to do the management and governance of all of their cloud and on-prem assets. So kind of this next generation of hybrid. We introduced what we called the adaptive cloud, which allows organizations to manage in a central integrated control plane all of their on-prem, multi-cloud, cloud assets and to be able to project those assets to their edge environment. So it's bringing together all of the decade innovation that we've had around edge, IoT, as well as application management into one single source.

Karl Keirstead: OK. Let's make -- can we unpack the first one a little bit?

Alysa Taylor: Of course.

Karl Keirstead: You led with it, so let's go there, the new Maya chip. So I think one question on everybody's mind is Microsoft has an enormous and successful partnership with Nvidia. So I think everyone's question was, how does Maya fit into that relationship? Is it targeting a different workload type? Maybe you can elaborate a little bit on that.

Alysa Taylor: Yes. It's interesting. The AI workloads are -- like HVC and AI workloads are different than enterprise workloads because you actually -- when you have these workloads come in, they are -- they're asynchronous, so you can have them both at peak times, at burst capacity or not at all. And so when we talk about the silicon-to-systems approach, the systems approach is really allowing us across a network of hardware to be able to load that one so that we can optimize for peak performance as well as cost.

And so, for us, our first-party silicon was just yet another addition to what we have built with AMD and Intel and Nvidia. And so that's the systems aspect of it, so it allows us to be able to control the performance of those workloads across the variety of the systems. And it's where things like boost come into play, as well as the innovation that we've seen in the fiber layer, as well.

Karl Keirstead: So there's a cost benefit to Microsoft as well internally.

Alysa Taylor: Yes. So if you think about it, GPU resources are very resource capital-intense, and so you don't want them to go unused. And so this allows us as you -- it is opaque to a customer but allows us to be able to balance the workloads on the back end. So we have no plans to replace our Nvidia or our AMD investments. It's just adding to the ability to do the load balancing.

Karl Keirstead: I'm assuming you spent Ignite talking about AI endlessly with the customers. So I think everybody could also benefit from your perspective on how those conversations are going, where we are on the adoption curve, where customers that are in the use case discovery process. What are those use cases that are coming up in that initial wave? In other words, where are they getting the greatest ROI near-term? That might be helpful for everybody.

Alysa Taylor: The use cases that we see emerging, really we look at it in three dimensions. So we look at it by very specific industry use cases, what are the line of business use cases, and then what are the individual function use cases. And so I'll give just a couple examples of some of the more prevalent that we see.

In industry, healthcare is one where one of the greatest challenges that healthcare organizations face is physician burnout. So you actually have seen in the U.S. physician burnout has increased from 43% to 52% in the last three years. So that's post-pandemic we're still seeing physician burnout on the rise. And so with technology like DAX Copilot, it is a combination of ambient and generative AI that can record an interaction between a physician and a patient, whether that be in a telehealth or an exam room visit. It then automatically analyzes, summarizes and generates a clinical note that can be uploaded into the electronic medical health record on behalf of the physician. So that reduces the administrative burden that a physician has.

And you see organizations like Atrium Health that have deployed DAX Copilot. They have actually recorded that their physicians are, on average, getting 40 minutes a day back, and they're improving the overall care of their patients. And so that's a great use case of using AI technology to combat a very critical industry problem.

In the horizontal space, customer service is one that we are using -- we're seeing that as one of the most prevalent use cases. Internally at Microsoft, we have deployed the Microsoft Copilot for Service. And in a matter of months, we've been able to increase the first-call responde) by 31% and actually improve satisfaction by 12%. So you see these use cases where the technology is being deployed and there's real outcome associated with it.

And then in the function, a great example of this is cyber professionals. And if you think about the sheer amount of data that a cyber professional has to reason over to be able to detect threats, Microsoft Copilot for Security actually aids in that process of being able to comb through all of the potential threats, pinpoint the most high-targeted threat, and then allow that cyber professional to actually work against that threat.

And so that's where you see sort of industry, line of business and function. And in addition to the use case, I often get asked, well, what is the return on this? What is the actual tangible monetary return? And we worked with IDC. We surveyed 2,000 global companies, and there was really interesting stats that came out of that. The first is that over 70% of the organizations had already deployed AI technology, so that sort of shows the interest level of AI technologies. 91% of those organizations had indicated that they had an AI project up and running in less than a year. So if you think about just -- we talked for years about digital transformation, that speed. So it is the pervasiveness of the adoption, the speed of adoption. And then, probably most interesting to this community is that, for every dollar -- it was reported, for every dollar that the organization had spent on average, they were seeing a return of 3.5 return on that dollar. So that's a really -- so it's taking those use cases and now starting to see clear patterns of ROI on them.

Karl Keirstead: Interesting. Let's talk a little bit about how Microsoft is scaling its AI infrastructure to meet that incredible demand. So I think there's a perception that there's a significant supply shortage of Nvidia chips out there. Can you comment broadly on the severity of that supply constraint and how Microsoft - you, Satya, the leadership - are scaling up that infrastructure to meet the demand?

Alysa Taylor: Absolutely. When we think about capacity planning, it's very multi-faceted. So we start with demand planning, so that is both short-term demand and then projections for long-term demand. And then we couple that with resource allocation. And so resource allocation is both the availability of the infrastructure, but then we also take into account things like geo-availability, land, power. So you can see it's a very multi-faceted approach to how we think about capacity planning.

And really, the systems approach that I keep talking about is so core to this, because that is how we are able to be able to do capacity planning across our GPU allotments, whether that be through our partnerships across Nvidia and others, as well as now our own first-party investments in it, and then being able to do demand shaping by geo. And we do this all wrapped in our commitments to sustainability. So we will actually come -- we will meet. We are on path for our 2025 carbon-negative footprint within our data centers. And then we also have a goal by 2025 to protect more land than we use in our data centers.

Karl Keirstead: OK. And Alysa, as Microsoft works to stand up that AI infrastructure as fast as you can, I recall a blog post in August. I think Satya and the team were talking about getting live on a number of H100 clusters, and that serves hundreds of thousands of H100s that are coming live over the next year. So that's a big revenue unlock, it feels to me, for Azure. So one thing I'd love to understand a little bit better are what the constraints are to get that GPU supply ready. You mentioned a few of them, not only getting your hands on the chips, but the power requirements are different, literally the land, the data center, the networking architecture. Are any one of those constraints any greater than the other in terms of standing up that infrastructure? Or is it a collection of all of those?

Alysa Taylor: It's a collection of all of them. I mean, I think that's why we take that integrated approach to capacity planning. You have to, because it is both the infrastructure, the geo-availability, the land use, the power structure, it's all of it. And you mirror that against demands today, and then the forecast against demands. And then you demand-shape, like I said. You then sort of help -- we have different data centers where we will -- certain workloads are more attuned to a certain geo. There are more sustainable data centers than others.

And so we work directly with customers to make sure that they are in the best data center possible, and then we, on a daily basis, monitor capacity of those both data centers, and we have a weekly socket that we review across our leadership team on things like GPU availability and capacity. So it's a very hands-on integrated approach.

Karl Keirstead: Without giving up any of your secret sauce, AWS and Google are also trying to stand up their AI infrastructures as fast as possible. Are there any differences in the way that Azure's constructing its new AI infrastructure that might give it a competitive edge a couple years out?

Alysa Taylor: I think the thing that we point to, we'll continue to point to, is the entirety of the system. It is not just about the chips. I think people like to talk about the chips. We have, for many years, had deep partnerships across Intel, AMD and Nvidia. We continue those partnerships. You probably saw yesterday AWS just entered into a partnership with Nvidia. That is -- we have a long history with Nvidia, and it has been part of how we've built out our infrastructure.

But it's every aspect of it. It is the networking. It's the security. And we're building all of this to be able to be the most performant and the most cost-efficient for our customers. And we actually -- there is a benchmark of super-computing in the industry of 500 cloud providers, and we are number one in the super-compute category. And so we feel like this integrated approach to both how we build out the infrastructure, as well as how we do capacity and demand planning, is a real benefit and it's paying off, not only in industry benchmarking but also for our customers to have the most performant and cost-effective.

Karl Keirstead: OK. Let's talk about the pricing a little bit on those Azure compute resources. A lot of us watched Sam Altman and the team at the recent OpenAI Developer Day. There were a number of cool announcements from that day, but one thing that struck me as really interesting is that Sam talked about essentially reducing the per-token price of their premium models, like GPT-4, by an extraordinary amount, 3X. So my first reaction was, boy, if they could reduce the price per token by 3X knowing that the consumption of Azure resources are a big part of that cost structure, does that mean that LLMs are becoming incrementally less compute-heavy, compute-centric? Which probably wouldn't be a great thing for Azure. How would you respond to that observation?

Alysa Taylor: Well, I'll start with saying we are obviously -- we continue to be very bullish on our AI growth. But the misnomer in the industry is that AI is an enterprise workload, and the reality is, is that digital-native small businesses benefit from the power of AI services. And so by being -- by making the models more efficient, and then bringing down the price, we are actually opening up our addressable market.

Karl Keirstead: You're almost like democratizing it in the P x Q phenomena.

Alysa Taylor: It's a P x Q, correct. And so it allows us to be able -- that efficiency allows us to provide these services to more and more customers where the barrier to entry before was price. So efficiency in AI services is actually a good thing, because exactly to your point, it allows us to democratize the services to more and more users.

Karl Keirstead: Maybe a similar question, but your answer might be the same. But another question that investors have is that, as the bulk of compute needs changes from training of the models to inference of the models, is that less compute-heavy on the inference side such that, as we go through that training to inference-shift, does that change the resource needs for Azure?

Alysa Taylor: Well, there's this new emergent category, right, which is LLM ops, it's the ops of it. And so how do you do things like prompt engineering, RAG, fine-tuning, all of that? We actually -- one of the announcements that we made at Ignite was model-as-a-service. So to really bring -- so that isn't a very labor-intensive. We're actually wanting to bring more and more tooling to the LMM ops space, which is a good thing, because then that allows people to train more models, use more models, have kind of the different models for different use cases and be able to reason and govern over those models.

So there's the compute aspect of the models and the inferencing of the models, but then there's the data aspect of it. And so more and more, your AI is only as good as your data estate. And so you see it's not just about the compute of the model, but it's also the ancillary services that come with when you are building or modernizing an app, data being a key component of that, storage being another key component. And so there's a -- we look at it from the AI platform side, compute, data services, networking and storage.

Karl Keirstead: I'm going to ask you about the data services in a quick sec.

Alysa Taylor: Oh, good. This is one of my favorite topics.

Karl Keirstead: But first, let's summarize this up to the Azure level that Brett and his team conveyed to the Street. So Amy on the last call gave guidance for stable Azure growth after the December quarter, and there's a couple dynamics in there. It feels as if it's reasonable to conclude that the AI contribution to Azure's growth should improve, and so stable means that the core, or ex-Azure part of -- ex-AI part of Azure might decelerate slightly. Maybe that's scale economies. But can you comment a little bit on that guidance? Why would ex-AI Azure growth moderate, really? Is it just -- is it scale?

Alysa Taylor: Well, if you think about the sheer number of workloads that you have to bring online to be able to deliver that stable growth, which is our forecast for H2, and then the other thing that I would say is you think about Azure from the entirety of the Azure platform, right? You have the infrastructure layer that we've talked about. You have the management and governance layer. You have the data layer. You have all of the AI services, the tooling, everything that we provide at the app layer. So there is in -- the workloads that make up Azure are fairly comprehensive and significant.

And so when we look at Azure growth, it is both around growing new workloads, and then you'll see customers do, which is normal course of business, which is continuing to optimize those workloads for performance and cost. And so our commitment is to grow workloads across every aspect of the Azure cloud while making sure that we are working hand-in-hand with our customers to maximize their price-to-performance.

Karl Keirstead: OK. Can we hit a little bit, before we get to data, on the optimization side? I think there's acute interest in that. So I'm sure at Ignite and through the course of your day you're talking a lot with customers about that, amongst other things. So where would you characterize Microsoft's Azure customer base in terms of how far along they are in terms of that journey to optimize their spend?

Alysa Taylor: Well, optimization is an ongoing occurrence because, if you think about when you bring a new workload online, whether that be migrating a on-prem workload to the cloud, whether it be building a new application, you're constantly optimizing the performance of that workload. And so optimization is just -- it's an ongoing part of cloud infrastructure and cloud compute. And so, for us, there's two things that we want to do. One is work very closely. We provide a number of resources, whether that be Azure landing zones. We have a service called Azure Advisor that actually helps customers build well-architected workloads and then how they construct those workloads in the most performant way. And that's part of optimization.

And so that I think is a normal course of business. And then, as they optimize, they bring on new workloads. And so our goal is to help customers build and construct the best workloads at the best performance, and then make sure that we are working with them on that next workload. And so that's where, when we talk about workload growth, it is in conjunction with making sure that our customers are building well-architected services at the beginning.

Karl Keirstead: Got it. Let's talk about data now, a number of interesting topics here. So one that intrigues me a little bit is this notion that, in advance of, or perhaps concurrent with enterprises moving forward on AI projects, they need to, quote, "Get their data estates in order." Are you hearing that? And to what extent is it starting to pull through, even if it's just in conversations today, the rest of Microsoft's data suite?

Alysa Taylor: Well, as I mentioned, your AI experiences are only as good as the data that the AI reasons over. And so it is very much step one is what is the data set that you want to be able to apply AI services against. And that data set has to be in the cloud. So there is kind of step one, which is being able to access or migrate your on-prem data, and then organize it and construct it in a way that the AI services can pull from it.

And so there's a number of things that we are doing to aid customers in being able to get their data estate in order. One is I talked about Fabric. Fabric is, for us, a very meaningful investment in our AI pursuits. So Fabric is our analytical data service, as I mentioned. One of the unique capabilities is the OneLake aspect of Fabric, which allows you to bring in various different, disparate data sources through shortcuts, so you can shortcut into data. You can mirror competitive data in our competitive, so in different -- whether you have data in GCP or AWS, you can bring that into OneLake, and then you have an aggregate data.

And then we have applied our AI services directly within Fabric. So if you think about how you then -- you take that data and you organize the data to be able to call against very specific data sets, things like vector search become very important, so we've brought vector search into Fabric through -- it's called AI Search. We've also brought it into our distributed data scale, Cosmos DB. You can have AI Search natively integrated into that. So it is both the aggregation of the data and then actually how you construct and access the data. And those are two really critical, important things.

And so Fabric and AI Search are very major investments that we've made as organizations are bringing their data into the cloud, organizing their data to apply the AI services on top.

Karl Keirstead: And also, I'd be curious. Are you seeing any evidence that AI is starting to accelerate the pace at which organizations are embracing the cloud? Are they -- is it boosting the on-prem to cloud data migration effort? Is it causing any acceleration in the pace at which you're running your databases in the cloud versus on-prem? Is that beginning to happen? I'm sure it's been a constant process. I'm curious if AI is acting in any way as an accelerant.

Alysa Taylor: Absolutely. If you think about there -- when you think about AI in the terms of applications, you're either modernizing an application or you are building a net new application. So if you even take Microsoft, what we've done with M365, Dynamics, GitHub, that's modernizing an existing application. And so, when you modernize that existing application, you're bringing in the Azure OpenAI service, which are the API calls into your data set. That's why your data is so important. But then you also see an increase in things like storage, because you actually have to store those processes. So you see when it is not just about the API calls, but it's about the aggregation of that data, the data services increase, as well as ancillary services like storage.

When you get into build, that is about kind of the entirety of the stack that you build upon, and that is everything from how you -- the developer services that you use, so whether that be GitHub and the GitHub repos, GitHub Copilot to actually do the coding; then bringing data into a operational data service, so Cosmos DB, as I mentioned, is our distributed at-scale database; then being able to apply the app services on top of it, as well as the developer ops management. And so you have this -- to create a net new application, you have an entirety of services that you construct to build a net new modern AI application.

And so both in the modernize and build, we're seeing really unique pull-through across the Azure platform. And so a great example is UI Path is a customer. They modernized their business process software with Azure OpenAI. Their storage and networking increased. They then started to build out net new applications, so building their data services as they brought new data services in to build out new applications increased.

And then another great example is the customer Real Madrid, so they are continually trying to figure out how to engage with their fan base. They were able to originally only culled from seven data sources. Now they are able to actually -- or sorry, five data sources. They're now able to cull 70 data sources by being able to do that data aggregation that we talked about in OneLake.

And then a really unique thing is they've actually increased their fan profiles by 400 percent in two years. So then it's a nice marriage of our ACR increases and, most importantly, our customer business outcomes increase, as well.

Karl Keirstead: Yes. This might be a good time to ask you a little bit about the Oracle relationship on the database on Azure. So by the way, Alysa, your profile among investors went massively up when you were seated in between Satya and Larry for that session. I think everybody in the room probably watched you then. How fun must have that been, right?

Alysa Taylor: It was very -- it was an incredible experience, nerve-wracking.

Karl Keirstead: Right, rivals can be friends.

Alysa Taylor: Nerve-wracking to have the two of them watch you as you deliver their intro, but it was an incredible experience. It was Larry's first time in Redmond, which was really kind of a momentous occasion.

Karl Keirstead: Yes, and it's a big deal not just for investors but for large organizations, UBS. It's public information. We're an extraordinarily large Azure customer, but we also have a massive Oracle database footprint.

Alysa Taylor: As most customers do, yes.

Karl Keirstead: Yes, exactly. So it mattered a lot to us. So the question to you is what does that mean for Microsoft? How are you going to benefit from this partnership on the database side?

Alysa Taylor: Well, as we've been talking about, how critical data is to building out unique AI experiences. The partnership that we have with Oracle was actually a direct request from our customers of being able to bring their Oracle databases into Azure. And so we have a very unique partnership where you can run your OCI database directly in Azure, so as you're modernizing, that governance management and modernization can happen in a very integrated way.

And for us, it gives Oracle's 400,000, over 400,000 customers the ability to bring OCI directly into Azure and run natively in an Azure environment. And so that's a very -- we've worked to be able to architect that in a way that is most beneficial to customers, so they can just literally take OCI, run it in Azure, and then have all of the ancillary Azure services that we talked about to be able to do the management, governance and modernization.

Karl Keirstead: Yes, huge partnership.

Alysa Taylor: Huge partnership. It's a very -- and the incredible thing is, is we're seeing a lot of interest from organizations come forward for this reason, to say it has been prohibitive for them to be able to bring their Oracle databases to the cloud, and now this allows that way for them to very seamlessly bring it into an Azure environment.

Karl Keirstead: Let's talk about a couple of the applications that bring a lot of your AI capabilities to the customer, so the Copilots. So I'll ask you about two. So on GitHub Copilot, color, Alysa, on how the traction is progressing, what kind of developer productivity gains you're seeing in these early days?

Alysa Taylor: So we talked about kind of the use cases by industry, by line of business, by function. Developer is one of the key functions where we see enormous potential. There is a -- and this has been happening for years, but it's what's called the app gap in the world, which is there are more applications being -- wanting to be built than there are developers. And so this is why you see things like the low-code tooling come online. And so it is both about making the developers more productive and giving them the tools to be able to do that.

And so GitHub Copilot has actually been a phenomenal productivity gain for developers. So developers that are using GitHub Copilot have increase of 55% productivity. That's an enormous number if you think about it, making developers 55% more productive. And we have -- and that base of users is growing. So we have over a million GitHub Copilot users, paid users. So it is a tool that has become very instrumental to developers.

And there is the making them more productive, but are they happy about it, and you actually see this really nice marriage of they are more productive and they have reported an increase in satisfaction in their day-to-day work, as well. And so that for us is a sign of true success, productivity plus gains in satisfaction.

Karl Keirstead: Got it. Let's now talk a little bit about M365 Copilot, only been GA for a couple of weeks, so understanding it's super-early. But what can you share with everybody listening in, and here in the audience, about the traction that Microsoft has seen so far? Can you share anecdotes or any -- probably not metrics, but anecdotes about productivity?

Alysa Taylor: There's a couple things that I -- yes, these guys will not let me share too many metrics. But as you mentioned, recently GA'd. Copilot, intentional from a naming standpoint, is that it is a co-pilot for individuals, right? So when we talk about GitHub, when we talk about M365, it is about enhancing the productivity, and both enhancing the productivity and what we call unleashing the creativity, so allowing humans to be more creative and more satisfied in their work.

We at Ignite released our work trends index, which reported of the Microsoft 365 Copilot users, 70% of them, they indicated that they were more -- that they would not give up Copilot, that it had become instrumental into their day-to-day environment, and that it had made them more satisfied in their work. And so you see this. For something that is so early to market, to see that large of a base say that it is critical to their day-to-day work is pretty impressive.

Karl Keirstead: The UBS CTO has promised that I'll have one eventually. I'm waiting. So as soon as it can arrive, I'll welcome it.

Alysa Taylor: I will tell you, PowerPoint summarization is key. I highly recommend that. If any of you get PowerPoints like I do, which is dense and large volumes of PowerPoint slides, there is a summarization button that can take those very dense, large number of slides and summarize them into a pithy summary for you.

Karl Keirstead: I look forward to that.

Alysa Taylor: Quite key.

Karl Keirstead: Let's give you a chance to make some closing remarks. I know we're out of time. So maybe, Alysa, when you think about the next three-plus years managing a business as broad as Azure, what are the couple of things that get you most excited in terms of the growth trajectory?

Alysa Taylor: Well, I would say the moment in time that we're in, depending on the age of some of us. We've lived through some pretty big shifts in the technology industry. If you think about Internet/mobile/cloud, we're now at this next inflection point, which is this AI transformation. And you've seen adoption happen at a rate that I have never seen before in terms of just the groundswell of people realizing and evaluating the potential of what AI can do, both for their company, for their individuals, and I just think we're in a really unique point in time in the market. And so it's exciting to be here.

And so it's not only just as we've talked about the individual AI services, but it's the ability across different industries, different functions, different business units to really re-define how they work and how organizations interact with their customers, how they reinvent business processes. So for me, the most exciting thing is just this inflection point that we are at, and we are early, early days, very early days of what's possible.

Karl Keirstead: You're going to have an exciting 2024, I predict.

Alysa Taylor: I think so. I think so.

Karl Keirstead: Microsoft, Alysa, thank you so much for attending the event. I enjoyed that conversation.

Alysa Taylor: Thank you very much.