Scott Guthrie1:54:32
Thank you, Judson. Well, thanks, everyone. It's awesome to be here. Now, everything you've seen so far in this keynote runs on Microsoft Azure. It's the infrastructure that is powering all of the Microsoft Cloud and all of our AI services. And Azure provides both the infrastructure and the cloud platform that you can use to supercharge your organizations and make them Frontier. Microsoft has over 70 Azure regions worldwide today and offers data residency in 33 countries, more than any other cloud provider. And we're continuing to dramatically expand our data center capacity so that you can put your apps and your services closer to your customers and employees everywhere around the world. And we're providing clear sovereignty commitments, giving you control over where your data and your AI lives and how it's managed. And we're optimizing Azure to be the world's AI supercomputer, and we're delivering AI innovation across the entire system. Last week, we unveiled our latest Azure AI Data Center in Atlanta. This Azure Data Center is the size of 20 football fields and contains hundreds of thousands of the latest NVIDIA Grace Blackwell GPUs. It has the highest density of GPUs of any data center in the world. Let's watch a video of it.
Now, this one Azure data center delivers ten times the performance of the world's fastest supercomputer. And this step change is what's accelerating the next wave of AI model improvement and AI capability breakthroughs. And AI infrastructure like this is also enabling exponential reduction of AI cost. If you look at where we were even two years ago compared to today, the price of a model like GPT-4 has plummeted by 93%. And continued infrastructure and model innovation is going to continue to drive cost reductions and, in turn, enable you to leverage AI for even more use cases. Now, we recognize with all this power and scale also comes responsibility. Microsoft is now one of the largest buyers of renewable energy in the world. And we're on track to meet our goal to have our data centers powered by 100% renewable energy by the end of the year. And this includes the data center you just saw in that video, as well as all the others that we're building around the world. And we're not just using renewable energy. We're also using innovative sustainability approaches to ensure zero water waste. The data center in that video is a liquid-cooled facility, meaning we cool the servers using water as opposed to traditional air coolers. And in this picture here, you can see the liquid chilling system that feeds cool water into the data center. And we use what's called a closed-loop system to ensure that the water is continuously reused with no evaporation or water loss. The initial fill of the water into the system equals only about 20 homes' worth of water consumption. And then once the initial fill is done, we'll reuse that water for six years without ever having to add any more.
You know, with Azure, we're setting a new benchmark for sustainable operations. Now, GB300s are the latest NVIDIA GPUs, and they're the most advanced GPUs in the world. GB300s deliver 11 times the performance of NVIDIA's previous-generation H100s, and delivers 12 times the performance of AWS's Trainium2. Microsoft Azure is the first cloud provider to deploy GB300s in the world, and Azure has more GB300s in production than any other cloud provider today. Customers are already running production workloads on GB300s in Azure. Here, you can see a cluster with tens of thousands of GPUs in it. Being first to run the latest generation of NVIDIA hardware means that you can train and inference AI models faster, more efficiently, and at a dramatically lower cost. And we're investing in differentiated silicon across both GPU and CPU workloads. Microsoft Azure's Cobalt silicon delivers industry-leading ARM64 price performance compute in the cloud. Cobalt-based VMs provide up to two times the performance improvement with .NET apps. And developers from Snowflake, Databricks, Elastic, Adobe, and our own Microsoft Teams are already taking advantage of Azure Cobalt in production, seeing up to 45% better performance, which translates into 35% fewer compute cores and VMs needed, which ultimately saves a lot of money. And today we're excited to announce our new Azure Cobalt 200 offering. Azure Cobalt 200 is a major leap forward, with more compute cores, more cache, faster memory, and higher performance, all built on the latest TSMC process technology. Cobalt 200 will deliver up to 50% better performance, and will be Azure's best price performance VM in the market. Now, all of this Azure infrastructure innovation, AI model improvement, and cost reduction is powering a new era of apps and agents. And it's going to enable all of you in the audience to integrate AI into every workflow and build transformative solutions that weren't possible before. AI leaders and more than 95% of the Fortune 500 are already building differentiated solutions on Azure. This slide here just includes a few of them. Now, one of the companies is OpenAI, who builds ChatGPT. ChatGPT is built on Azure and uses the exact same Azure services that you can use as well. Services like Azure GPU VMs, Cosmos DB, Azure Kubernetes Service, Azure PostgreSQL, and Azure Storage. ChatGPT needs to be able to scale their application tier across tens of millions of compute cores around the world. And they do it with just about a dozen engineers, which is pretty incredible. Now, how do they do that? Well, they do it with AKS, which provides a highly scalable Kubernetes service for Cloud-native applications. AKS is a fully managed Kubernetes service available in every Azure region, and it streamlines operations at any scale. AKS offers automated deployments, auto-healing, automatic patching, and built-in security safeguards. And these enable applications like ChatGPT to scale without significant operational resources. And with our new AKS Automatic capability, we're making it even easier for teams to get started with Kubernetes. AKS Automatic has built-in best practices for security, reliability, and governance. And it allows you to go from code to fully deployed Kubernetes clusters in minutes. And I'm excited to announce that AKS Automatic is now generally available as of today.
Now, data is the fuel that powers AI. Great AI solutions are built on highly capable data platforms. If you want to build a solution like Microsoft Copilot or ChatGPT, you need a database capable of storing petabytes of data, handling trillions of transactions, and supporting limitless growth. And Azure Cosmos DB provides that. Azure Cosmos DB is a globally distributed, multi-model database service with guaranteed millisecond latency and uptime. With Cosmos DB, we've built a database service that can automatically replicate your data to any Azure region around the world to give your users lightning-fast performance, regardless of wherever they're accessing your applications. And with both Copilot and ChatGPT's examples, as users interact with the apps, conversations, prompts, and metadata are stored using Cosmos DB. And this enables these apps to maintain context across sessions for hundreds of millions of users, delivering a natural user experience with low latency and high reliability at truly global scale. Cosmos DB also allows you to elastically scale your storage and performance throughput with zero application downtime. You can start small and scale to exabytes of data and trillions of transactions per day. And you can start with processing, say, just 100 operations per second, and then scale to millions of operations per second if you need to. And best of all, with Cosmos DB, you only pay for the storage and the performance throughput that you actually use. And Azure Cosmos DB delivers incredibly fast response times and 'five nines' of availability, being the most demanding performance and uptime needs for any application. That's why innovative leaders across industries and around the world are using Cosmos DB to power their most critical apps. This slide includes just some of the logos of companies using Cosmos DB today. Now, we also know that customers want to be able to scale without limits with the relational databases as well. And today, I'm really excited to announce that we're introducing a new Cloud-native database solution to our Azure PostgresSQL offering family. Azure HorizonDB enables you to power mission-critical apps with performance and speed at any scale. HorizonDB is fully compatible with Postgres, and it supports scaling out a Postgres database to over 3,000 cores and 128 terabytes of storage. It supports sub-millisecond, multi-zone commit latency. And HorizonDB also allows you to build smarter apps using in-bit database AI, vector indexing, and Semantic Search, all integrated with Azure AI for seamless innovation. Now, data is everywhere. It's stored in NoSQL databases, relational databases, and unstructured sources. It's in every app, every system, and every user interaction. And while that data holds incredible value, today it's too often fragmented, making it hard to manage and even harder to use effectively. One of the biggest things that we hear from all of you is the need to bring all of that data together into a single, cohesive foundation. If you can unify your data, you can unlock better analytics and provide the fuel that powers your AI solutions. And that's where Microsoft Fabric comes into play. Microsoft Fabric is an end-to-end data platform that takes you from raw data to AI and BI value, all in one unified experience. It brings together purpose-built workloads for engineers, analysts, and BI professionals. Earlier in this keynote, Asha announced Fabric IQ, the semantic layer that gives your data meaning. With Fabric IQ, Power BI and Copilot agents can now share the same underlying understanding of your business. And Microsoft Fabric is built on OneLake, our unified AI-powered data lake that brings all your enterprise data together in a single, trusted foundation. OneLake provides a unified AI-powered data lake for all enterprise data, structured, unstructured, and everything in between. What I'd like to do now is turn it over to Patrick to show a demo about Fabric and OneLake work in action.