 All right, so our final speaker for this morning is George Castro. George Castro comes from us from the Cloud Native Computing Foundation. He manages DevRel there. And he's going to talk to us about what's happening at the CNCF and what the Linux Foundation is doing to help shape the future of GenAI. So, George, please come on up. Hey, how y'all doing today? All right, I'm George Castro, a little bit about myself. I grew up in the Linux era, worked on Ubuntu for about a decade, moved over to Kubernetes when that started to come out post 1.0. And I've worked as a community manager for the Kubeflow communities and Cloud Custodian. These are all Cloud Native projects. And I am on month six of working at the CNCF on the projects team. There are 174 projects in the CNCF. So when you are perusing the buffet of technology to consume, we are charged with keeping those projects healthy and ensuring that contributors are finding their place and helping to drive the innovation that is consuming so much good stuff. I cannot wait to talk to you. This is my first AI anything. I've been learning a lot, playing with it locally on my computer and just kind of crash coursing myself. I find it having grown up through the Linux era and the Cloud Native era, it feels like everything just keeps accelerating. And just when I started to get a handle on the software, the hardware is also moving at this absurd rate. So I'm going to tell you today about some projects and organizations that are using Cloud Native technologies and production. Talking about 20,000 nodes, high throughput computing, about 100,000 jobs scheduled per minute, this kind of scale. This is Kubernetes now. There are CNCF members who are doing this today. I'm going to highlight some of the projects that people are using and organizations such as Bloomberg, CERN, OpenAI, NVIDIA, and Adobe, and of course, Hugging Face. I'd like to talk to you today about the next trillion core hours. We just had KubeCon and Cloud NativeCon in Chicago last month. And one of the key notes was talking about, first of all, it was a reminder that Kubernetes turns 10 next year. And we're starting to come to the point where projects such as this start to need to look at what the next decade looks like. We kind of cut our teeth figuring out the first decade. I was there. It was a lot of, how do you get the flywheel going? How do you become more efficient? How do you on-ramp folks? Documentation, all of the things that make open-source work. When Jim yesterday talked about how users are kind of demanding that AI be open and collaborative, these are the kinds of things that we're talking about. And I'm gonna talk a little bit about those in a little bit more detail at the risk of preaching to the choir. Because I think it's important for us, especially those out in the audience that might be watching this on video and things like that, that we need to start thinking about the long-term sustainability of these contributors and how we can continue to keep things moving. So what does open look like? So in a Kubernetes mindset, great primitives, API-driven, strives to be a universal interface for resource consumption. And users love the extensibility. A lot of the people that are using Kubernetes in production today are using it because of the ability to extend that API and get what they need done. But open also means exercising the processes. That means when end users take the extensibility of those APIs and they're able to move and get what they need in production, that they are able to communicate back to those open source projects to ensure that the common primitives are making it back to core for a few reasons. Maintainability and it's cheaper. So as it turns out, having end users be able to extend these APIs and then communicate that back into core, making that loop nice and tight and as fast as possible is something that we prioritize. I'm gonna talk about a few things today. The first is hardware enablement. This is something that's very front of mind for Kubernetes folks and associated projects, GPUs, TPUs, FPGAs, all of that kind of things. I'm gonna highlight two Kubernetes enhancement proposals. Those are what CEPPS means. The first is gonna be dynamic resource allocation. This is an alpha. Is there anybody here involved in these discussions? These are some of the more lively discussions that are happening in cloud native today. This is a very nuanced topic that takes a lot of thought because it does kind of change a little bit on how you might consider how Kubernetes is allocating resources and those conversations need to be concluded. And of course, CEP3545, which is the improved multinomial alignment in the topology manager, another CEP. There's multiple CEPs in these areas and then I'm gonna show you the website at the end where you can look these up and checking them out. But the TLDR is you all wanna slice and dice your GPUs and all that hardware and all sorts of different combinations. Next I wanna talk about workflow enablement. This is about reducing complexity. Index jobs actually landed in Kubernetes 1.24 and job tracking for massively parallel batch workloads and this is where you're getting up to 100,000 concurrent pods that actually landed in Kubernetes 1.26. There's a few batch systems that I like to put on your radar. The first is Volcano, a CNCF project and Q with a K. Make sure you take a picture of that one. And there's also two batch working groups I found out, one at the Kubernetes level for the cluster folks and then one at the CNCF level that kinda covers everything else. So do be cognizant that when you say batch, sometimes you need to be a little bit more specific about where you need to be. Lastly, I wanna talk about application development. So I'm a former Kubeflow community manager. Kubeflow just made it to release 1.8 in November and it continues to be used in production, notably at CERN. Always have great conversations with those folks on how they're consuming that and that project is now in incubation in the CNCF. I'd also like to point out Carmada which went into incubation yesterday which is for multicluster application management which I'll get to in a minute. And lastly, I like to talk about Kube edge of Sedna which is about pushing that AI workload to the edges. There's many nascent projects in this area and those are three that I thought you might wanna check out. And of course, as this starts to be consumed more, we're starting to talk about things as they consume those primitives, users are gonna wanna be able to grow those integrations with all the other tools that they're using in their clusters like Averno and OPA, so all your policy management and all that stuff is lined up. And as I'm talking to people who are doing this, sometimes they kinda wonder what's the relationship between OCI artifacts and alimonyms? Are we just shoving these into registries? I don't know, but I'm really looking forward to seeing the discussions that people come out with. Because for us, it's gonna be about unlocking these new primitives and bring them down to the cluster level. Clayton said, inference is the new web app in the first kind of decade of Kubernetes' growth. It was all about getting those web app. The first demo I ever ran of Kubernetes was like a web app. So if inference is gonna be the new web app, what I really wanted to say today is that Kubernetes is already the web server for a lot of production AI workloads. And as these features and the feedback loop from end users feedback into the system, it will only get better from here and then y'all keep adding zeros to everything. It just makes it really hard. And we're now moving what I call scrambling up that stack where it's not just about the individual cluster or the individual pods or the nodes. You're now thinking at a multi-cluster level and how we're gonna manage that across regions. I think you've sat in enough talks to see the level of scale that some of you are consuming this technology with, and it's really exciting. We'll also take multiple organizations. I'm just highlighting three here. The level of complexity that we're dealing with is just something that just one organization can handle. It takes multiple organizations and multiple verticals working together. I know there's many people in this room. You are multiple hats, multiple foundations. And I just wanted to point that out that this is something that we're all doing together because that's what we do. So I invite you to build the scaffolding with us. If you go to kubernetes.dev, you will find all the kept enhancement proposals there that you need to be able to go as with anything with open source. Everything is open, meetings are open. Meeting videos are always on YouTube and we're always looking for people to participate if you're an end user. Thank you and I will be here all day if you have any questions on how we can help you and hope to see you all in Paris and next year.