Melanie Nakagawa25:13
Sure, this is a great question. I love the chance to be able to explain and talk a little bit more about the details of what's going into our data center builds and really the breadth and depth of our employees across Microsoft and our partners around the world in the supply chain. Because it is like a baton pass. When Katie starts to reduce her emissions, her embodied carbon in those chillers, she baton passes to me a lower embodied carbon that I'm reporting, and then back and forth. And then my cloud is a lower embodied carbon to her services as well. So it really is that partnership piece. It just keeps ping-ponging more and more megatons and gigatons of carbon saved the more we keep on racing in this way, and the planet benefits. So really to this question about AI and energy use within data centers, it starts with physically how we think about the growth of data centers. We think about first what is that load growth going to look like. We really are at this inflection point in time in history. I mean, I think we will be talking about the time before and after generative AI. So much of our daily lives will be greatly enhanced from everything from the mundane to the more complex. And so when we think about that, it's an incredible responsibility for a company like Microsoft that is building out these new technologies and building out the capability to do so. And so the way we think about it is how do we build, operate, and apply AI in a sustainable way. And that starts at the infrastructure level. So on the build side, it starts with the models themselves. How are we thinking about optimizing the models of AI and generative AI to be as energy efficient as possible, as model efficient? You'll see that we keep on coming out with both large language models and small language models. The Phi models, the most recent one, it's massively more compute efficient and energy efficient at the same time, and it's a faster trained model as well. And so what's great about this is that we think about with our IT department, our engineers, our designers, our coders, our developers, how are they bringing in principles of green software and how they think about coding and optimizing these models. So that's first on the build side: making sure that the compute process and the modeling is done with sustainability principles attached to it. The other piece is also the infrastructure inside. So I mentioned you've got low carbon steel, low carbon cement, but the physical build we're doing more. We're also piloting a 50 percent lower embodied carbon steel in the hot air cooled aisle section of our data centers. So think when you charge a phone or computer, it can get quite warm. Now picture that at a massive scale with a data center and servers. So we use steel to cordon off that section of a data center so that area can be more efficiently cooled with Katie's work and your work. But that steel now is going to be built with a piloting 50 percent lower carbon steel inside that building. So that's on the build side and the intentionality there to make sure the building is done more energy efficient. The second is operations. So we're growing, we're building out more, and these models do require more energy. And so there we want to make sure we're scaling our procurement of carbon-free energy just as quickly as we're scaling the growth of our business. We now have 34 gigawatts of carbon-free power procured by Microsoft to date, and that's part of a broader remit across many tech companies that are helping to really scale the clean energy marketplace through power purchase agreements. We're putting it on the map, and we're actually getting net new power. Most recently last Friday, you might have seen some news that Constellation Energy announced the bringing back of Three Mile Island Unit 2, the reactor that had been operating safely for about 17 years. And Microsoft is entering a power purchase agreement there to help bring back that facility, which is going to bring 800 megawatts of carbon-free power onto the grid into the wholesale market of PJM, which is a already constrained grid. So where you see AI load growth grow, you're seeing big bets and big investments from companies like Microsoft into adding net new power, and in this case 3,400 jobs, to ensure that we continue to grow and operate with more power in the Pennsylvania market and the PJM market. So that's on the operations side. And then we think about application, which sometimes we don't often talk about, but it's so exciting. It's exciting because those data centers are being built, they're delivering models that are optimizing energy systems, that are coming with new material breakthroughs, that are doing things not just futuristic but today. So a company based in the UK called Line Vision, they're using AI today with UK National Grid to optimize the existing transmission line in the UK. Through their program, they found that it can bring 600 megawatts of more offshore wind onto that transmission line. That's roughly 60 percent more renewables moving on existing transmission line today, and that's using these models. So this is an area where it's not just about what models can do or will do, but we've got, you know, we started the AI for Earth program in 2017, so we've got several years under our belt with partners around the world applying AI already today with use cases that are optimizing energy systems, predicting weather events, and really helping to bring new energy systems onto the grid. So I say all that because yes, we know the growth is coming, and I share this to say that we are preparing for this growth, we're readying for this growth, and we're making sure that we build, operate, and fly so that we can do this sustainably.