Hello my friends, and welcome back to BUILD. I am excited. I'm Seth Juarez. I'm here with my friend Scott Guthrie. How are you doing, my friend?
Fantastic. So are you enjoying Build? First of all, I am fantastic. So let's start first. I want to start because I want to get into this. What's top of mind for you, for developers as they're here, and tell us what you do so people can know. They all know, but I want you to tell them.
Sure. I mean, so you know, what my job is, is I run Azure and kind of the overall Microsoft cloud infrastructure. So everything that we showed today runs on top of that. So my team and I are kind of sort of involved in everything at the base level and then also all the data platform and several of the higher level services. And so, you know, things that are top of mind for me are, you know, how do we keep building out the infrastructure to support the AI revolution that's happening? So I spend an awful lot of time working on adding data centers and network and compute and storage to make all that possible. How do we do it more cost effectively? And so a lot of the innovations we talked about even today in the keynote around Azure Maia and Azure Cobalt in terms of our first-party silicon, how do we take, you know, a large percentage of the cost out of AI solutions? And then how do we enable AI that's rock solid, reliable, super fast, super performant and enables all the higher level tooling, whether it's GitHub Copilot or other tools, Microsoft 365 and our overall ecosystem solutions, you know, to be a platform that you can then build amazing AI solutions with.
So let's talk a little bit about the fundamentals, 'cause this is the cool part about Azure is that there's a ton of like operating system fundamentals. Can you give people a sense of the scale of this stuff? Like we have a ton of data centers, a ton of... Tell us about that.
Yeah. I mean, I think in our earnings report, you know, we kind of quoted that we added over a GW of data centers in 90 days and, you know, gigawatt's a lot, it's like the size of Seattle. And so it's, you know, we're growing fast and we're doing it, as Satya talked about in the keynote, responsibly. Our data centers are zero water waste. So, you know, they use less water than a restaurant does in an annual year. And, you know, we invest very heavily in making sure the communities that we operate in, we're good stewards in terms of electricity prices, taxes, community engagement, because we want to be long-term members of the community. And to do that, we have to act responsibly. And then, you know, there's an awful lot of work involved to do this all reliably at scale on a consistent basis in 80-plus locations and countries around the world.
That's the fun. You say that with like, there's a lot of... Tell us about the work involved because like these things, we just have them turned on and we're super excited about it, but there's a lot of work that goes on to do that.
Yeah, I mean, you need to get land, you need to get power, you need to get permitting. You know, in terms of the, you know, some of these sites have 6,000 workers on them in terms of doing construction. It's very precision tradecraft. You know, this is not just anyone can walk up. You're talking about people with 5, 10 years of experience in terms of electrical, cooling and other construction things. So this is really tradecraft and, you know, and you got to do it safely. And so we take safety incredibly importantly because at the end of the day, you know, the most, you know, we have signs that we put up on our job sites. The most important part of the day is you go home alive and safe. And, you know, and because this is high electrical equipment, this is heavy equipment, you kind of can't take that for granted. And, you know, thankfully we have one of the best safety records and again, one of the best environmental records in the industry. And we're kind of growing at scale.
And then there's the software that goes with it. Yeah, of course. Because when you're growing at the rate we're growing, we're constantly having to change the network architecture, change the storage architectures. You know, there's huge supply chain constraints in the world on memory, on SSDs, on hard drives. And so you're constantly also having to architect new ways of doing things to basically handle what is effectively a doubling of capacity every two years or so. And, you know, generally in software, almost no software when it handles 10x the scale works. And so you kind of, if you're doubling every two years, you kind of need to be starting on the next generation almost immediately upon shipping the current generation because you'll be at that next generation very quickly.
And that's surprising because I, it's obviously something I haven't considered because of your doubling scale. Like everyone's built software on your website works great when like your mom looks at it or your dad, right? But then all of a sudden when thousands and then hundreds of thousands, and you have to change the way you think about how to architect the software on top of the core. So what are some innovations that would be surprising to folks here and watching that maybe they hadn't said it needed to be done?
Well, I think one of the things, and, you know, we benefited with our partnership with OpenAI as an example with some of our data services and with Azure in general. But, you know, as you think about, just give one example, would be memory and chat history and authentication. You know, in a world where you've got an agent that's running all over the world, if all those agents have to go back to a single database at some point, your database will fall over. You know, it just will not be able to scale to the load of a ChatGPT or a Claude or a Copilot. And so, you know, how do you architect your database layer so that you can scale, you know, linearly? And so a lot of the work we did in Cosmos DB as an example was to enable that, where we can basically do a separate instance of Cosmos in every region, but then do eventual consistency of the chat history so that, you know, if any region has an issue, you just go seamlessly to the next region and away you go. But you don't have any synchronous locks anywhere. And that way you can replicate scale units, you can keep growing in a very natural way. And, you know, that'd be like one example.
And then, you know, the trick would be like things like attachments. When you upload something into ChatGPT, like if you store it in the database and everyone's uploading documents now and images, you know, that database gets really big. And so, you know, how do we do external attachments and spill over to blob storage where we effectively are doing like archival-level costs, but at the same time having the ability to write SQL against it and then having a globally replicated state store? You know, that type of thing is not easy. And what makes it particularly not easy is when these apps are evolving constantly. You can't like take it down for a day to upgrade the schema. You kind of need to keep the plane running in the air while you swap out the engines, and that's part of the fun of running a live service.
