AI Maturity Model & Cisco AI Tech Model

Hello everyone. Welcome to today's expert insights series session on AI maturity and Cisco AI tech stack. My name is Riley.00:10Course and that will be your moderator for today's session.00:20We're also joined by experts Dave Cole and Mark mcbrede who will bring this session to life by giving you a practical framework for AI adoption. During the session, we will also be answering live Q and A So if any questions pop up, feel free to type them into the Q and A panel of the Webex app, and we will get to those.00:23As soon as we can.00:43Today is session one of a learning series AI portfolio overview and customer maturity model. We are answering fundamental, the fundamental question, how do I adopt AI today? Future sessions in this series, we go deeper into networking for AI, compute for AI, real use cases.00:49And agentic AI.01:09But everything really starts today with understanding the journey and adoption. Towards the end of the session, we will provide more information on some of the future sessions including sign up information if you're interested in any of these as well. So before we jump into things, I'll start off with a quick agenda.01:11We have an introduction. We'll take a look at operationalizing AI and then AI maturity model. We will look at application to a use case, followed by tech stack for AI and then wrap up with some key takeaways.01:33So before we jump into anything, we have a quick Slido survey for you guys, so if you wouldn't mind scanning the QR code, if you were asked for a meeting code and password, it's going to be Cisco AI maturity. There's also another view where you guys can go ahead and scan the.01:52QR code. The 1st question here, what best describes your primary role? Looks like we got a couple of IT managers and directors in the call, some network architects, network engineers, followed by some systems engineers, solutions architects.02:12Awesome. Glad to see a nice spread of different roles here and see everybody's awake with us today. Awesome, we'll go ahead and stop the poll there. Looks like we got a couple. Oh, got some final answers coming in.02:32Awesome, thanks so much for participating in the Slido we can start off with introduction. Over to you Mark. All right, thanks Riley, thanks for getting us started. Good morning, good afternoon, good evening, everyone depending on where you're joining from and welcome to today's session where we're gonna go over some foundation.02:48No Topics, what does it look like to get started with a AI maturity model? We'll go over some of our portfolio and and help to tie it all together. So.03:07Before we get started, we'll take a look at how we see AI being used in the industry today. So most of us are probably very much aware of how AI began and how even even some of us were early adopters and we're seeing it all over the news and social media, but now we're starting.03:20Going to get into this transition where we're seeing AI in the enterprise space.03:40So here's a few examples where we can see AI being used in manufacturing, where it can be used for predictive maintenance, quality control, demand forecasting, helping to stabilize environments and minimize downtime.03:46In the public sector, we're seeing it being used to build smart cities to improve security and safety for public services. In the retail space we see how AI is being used for personalization, helping to enhance the online shopping experience.04:03Financial services is using it for a fraud detection, things around risk assessment, even trading. We're seeing the advancements that are being done in the health care industry as far as using AI for a diagnosis to be able to do.04:22Drive through optimization and and patient support and then even especially in the education space, AI being used for learning and teaching experiences and to be able to secure facilities. So with all these different examples, we can see how AI has.04:38Really move the needle and being able to allow these industries to progress more efficiently. And here's some examples of use cases that aren't really industry specific. These are use cases that can be applied everywhere. Some examples.04:57Include AI assistance that can be knowledge based co pilots. So think of AI assistants that can summarize your meeting notes, they can help organize your calendar, they can review documents.05:17There's also content and code generation using AI to create images, make videos, even enhance your coding or use AI to teach you how to code. There's also virtual agents that can be very specific.05:31Chatbots or they're speci specialized in a domain where they can help answer general questions that are coming in and things around visual computing where we can have digital twins, video analytics, and then even language translation. So now we can record videos and translate them and then share the.05:48In a real time or use real time translation enabled us to communicate with other people around the world.06:08And then of course with detection and prediction, that's what AI does best. It it predicts it's really good at forecasting, finding anomalies and then creating those deeper insights. So here's an example of what a, a full AI services.06:16Back and look like.06:36So we'll start here at the, at the bottom where we have our, our GPUs. So this is the foundation. This is what's making AI work and we can see where we could possibly have a, a multi tenant GPU. This could be a GPU as a service.06:40Leveraging technologies like Kubernetes, TensorFlow, pytorch or Jupyter.07:00Then in order to make AI work, we have to have the data, right? So this data can be on prem, it can be in the cloud, it can be data you're pulling from databases, it could be logs and metrics that you're pulling from applications. It can be a wide variety of documents that could be Word docs, PDF.07:06Box, maybe videos content. So this is where all your data lives and it's gonna fuel your, your AI.07:25Moving on up, we have our SAS based AI services. This is where you decide am I going to use a model that already exists or am I going to create and train my own model or maybe I use that existing model and do some fine tuning to make that model fit use case that.07:34I'm targeting. Then there's also things on data infrastructure where we can use ingestion and retrieval for documentation, LLM tracing to do agent tracing, and then things on model registries to make sure that we've got versioning control.07:53And moving over to the right, we have different types of platform services such as data platform tools that can be used for the AI and ML tools. You can also have AI coders that can help your developers to create new code, fix existing code, and even do testing.08:14Things on service management for business intelligence and then even conversational AI to pull those deeper insights from your, your AI models. And then at the top we have the the agent ecosystem. So this is kind of like the, the front end to your model. This is where you're.08:34Your chat about applications live. This could be where your AI.08:54Agents are being hosted at, it could even be multiple agents that could be part of a hub or even the agent platform. So this is really where your users are interacting and this is where the the tokens are coming in to be used by the model. So if we take a look back on just how AI.08:58Has it evolved and.09:18Here it says 1990s, but it's really started way before the nineties, but it was around 1990 or so when we really start to hear things around AI and machine learning and this is around the time where we're teaching machines how to identify and label, you know, pictures.09:20Of cats, we're feeding it information and when we're training these models. And then in 2022 we get chat GBT and things have really explode. So when chat GBT came out in 2022, within like the 1st couple of months, they already had a hundred million subscribers.09:40To put that in perspective, it took Facebook roughly around four and a half years to reach that number of subscribers. So this is where we could interact with a prompt, ask it questions, have it summarize articles, documents, and then get some answers back.10:012024 became the year of AI assistance. So here's where we use these AI tools to summarize our meetings. Tell us the highlights of a, a PDF file, create content, create a PowerPoint, things that makes life easier for our, our day to day work.10:20And then last year we start to see this movement towards a gentic AI. So instead of interacting with a prompt, I'm gonna tell the agent what I want it to do. And I'm gonna give it a specific set of skills and tasks and resources and let that agent.10:40Go off and do whatever I've asked, assigned it to, find its resources and then bring me back that data. And then now in 2026, we're really starting to see this evolve from the AI that's living in our screens and and in our phones, and now it's becoming a part of robotics and.10:58And and humanoids.11:18So in order to get where we're at, there's a lot of things and a lot of businesses that they have to think about what does it take to get here? What are some of the challenges that are along the way, you know, building an AI model and application that can be complex?11:21So today we're looking to cover, like what are some of the adoption barriers? What are some of the ways that you can help to make that, that challenge easier and, and break it down into different stages along the way. So some of the adoption barriers that we typically.11:39What we see is starting with data. So data is the underlying tool for AI.11:59And there's data everywhere. So it can be on prem, it can be in the cloud. Do you have access to that data? Is that data clean? Can it be used? There's also infrastructure, so the way that these AI models and app applications work are different than your typical.12:04Data center network, we're now introducing GPUs, we're now introducing lossless networks and different ways that the network behaves that can be completely different from your standard data center network.12:24Of course, we gotta have security in there. How do I know that my model is secure? What kind of testing have I done to make sure that there can't be any type of data model poisoning or or things in the model that I don't want to be?12:39Introduce and into the external environments. So that's where we have to wrap security around it. So starting with data, you can't have AI without data. You can have the most powerful AI cluster and the most recent GPUs, but you're not gonna get the value.12:55View out of it if you don't have the data to fuel it. There's many ways that we can get data in. It can be human generated, that can be different types of text files, PDF documents, recorded content, audio, video, all the things that you would want to use to, to train your model.13:15But it could also be machine generated, it could be metrics, logs, events, telemetry that you're collecting from your existing applications today and you want to leverage AI to, to summarize all of that data. And then there's always a challenge of.13:34Being able to build the right sized infrastructure for all of that data. So here, we're taking a look at example of of inferencing. So basically there's two types of AI workloads. There's building and training a model.13:54And then there's inferencing which we can see here. And basically with inferencing is when we have a chat prompt, we put something into it in order, something we're trying to achieve in this example, we're trying to create an agenda for a board meeting. We're saying who should be in the meeting, what are some of the topics that should be covered.14:09Word in that meeting. So when we put all of this data into this chat prompt here, each one of these gets broken down into a token. So a token is basically a, can be a word.14:29That can be broken down into individual tokens such as help me would each be a token, the word organized could even maybe be split up into two tokens. You see for the example here, each one of these words is, is a token that we're gonna feed into a model.14:47And then the model's gonna give us back a response. So if we're, we were just to take this prompt as it is and then send it through, this is what's known as aa1 shot response. So we're telling the prompt.15:05Here's what I need you to do. We send it in, it does its calculations, it sends us back tokens in the form of what we view as words or an answer, and that's a one shot token, and that consumes a number of tokens for it to come in as well as number of tokens that we're getting back.15:22But we can see there's different options that we can add to this prompt, and that affects the number of tokens that are coming in. If we add additional photos or files that we want to be part of this prompt, maybe it's meeting notes or it could be a recording from a previous meeting that needs to be included as part of this prompt.15:39That increases the number of tokens.15:59There's also options where we can do deep reasoning. So this is where we're telling the model to really think about what we're asking at like a human, go through all of the different scenarios and then give me the best possible outcome. So since we're asking it to do more during a reasoning, this is also going to increase it.16:03Token count by, you know, ten to 20 X because we're asking it to do more.16:23And then there's even options like agent mode. So this is where we can use the same prompt, but maybe in agent mode. It creates the meeting, it finds out who needs to be included in the meeting by checking the directory. It creates the invite, it sends out the invite, it creates the entire.16:28Agenda, maybe it's multiple invites that it needs to send out. So this can increase your token count by even shift either 200 times cause now we're letting the agent go out and do even more.16:48So you might be asking how does this relate to infrastructure? So all of these tokens is what's gonna be flowing through your infrastructure. Tokens that are coming in, that have to be processed by your GPUs, tokens that have to come back out. So your infrastructure has to be able to support.17:01The demand for what you're offering as as part of your AI services.17:21And of course, last but not least, but definitely the most important security. We need to make sure that our AI models can withstand the proper type of testing. We want to make sure that there can't be any prompt injections that our models don't have any harassment or hallous.17:30That they're trained on the proper data to be able to give the correct answers. We want to avoid prompt injection, we want to be able to test to make sure that users can't ask a model something and then either maliciously or maybe even indirectly get something back from the model that that they shouldn't.17:51So here's where we can use things like secure AI factory and an AI defense to make sure that our models are secure. So we're addressing all of these challenges head on and Cisco is providing the critical infrastructure for this AI area. And this includes hardware.18:09This includes software, this includes services, and everything in between to help you with your AI journey. So when you're getting ready to go through building out your 1st AI deployment, it's, it's not one person that can do all of this. It really.18:28Starts with the business leadership, who's setting the goals? What are the targets we need to hit? Who has an understanding of how AI can better our business?18:49From there we need people like data engineers that can help gather and prepare that data, make sure that it's labeled, it's clean, it's accessible. These are the answers that we want to be able to provide in our prompts. And then a combination of your data scientists.19:01Your machine learning engineers and your application developers that understand the AI services that are being created, that understand how the model's being developed, helping them to either add AI services to existing applications or even building new applications that have AI built into.19:20Them. And then of course there's model monitoring and management.19:40That has to continue once the model's been deployed. And then the IT operations teams, these are the ones that are building out the infrastructure. They know where the data is. They have the resources to help develop that model. They know how much GPU is available in the AI clusters. They know how to.19:46To monitor it and manage it and then how to scale it as the business grows and there's more need for additional AI services, the operations teams knows how to expand those services and that infrastructure.20:06So when it comes to operationalizing AI, it, it is complex. You start off with what's the problem we're trying to solve? Do we have the data to do it? Is, is this possible? Is the data usable?20:24Is it clean? Do we need to go back and relabel this data? Is there updates that we need to make? And then even deciding, am I gonna build and train a new model or do I just use an existing model and I can use techniques like fine tuning to then customize that model for what I needed to do?20:40You.20:59Then once you get to that point, you go through testing and you start to put it out into the real world where it goes through maybe your testers, developers, user acceptance testing until finally it's ready to go into a production deployment. At this point, you're wrapping security around it. You've got.21:00Bring in place to see how that model's performing.21:20Using tools like Splunk observability, Nexus dashboard, intersight. That way you have eyes into how your model is performing and what does that end user experience look like? So we've covered a lot in just the 1st few minutes, a lot of questions.21:24There's a lot of things that need to be answered. So where to start? So this is where we introduce the AI maturity model and it's, it's five different stages and it's broken out into two parts. The 1st one being how does this affect the business?21:432nd part, how does this affect our infrastructure? So in stage one, this is your awareness stage. This is where you're seeing how AI is being used in the world, how other companies are benefiting from it.22:01You're thinking of ideas, things that you could do internally coming up with questions like if we had AI, we could use it to accomplish XYZ On the infrastructure side, there's NO change. This is your, your day to day infrastructure that you're already managing. So there's NO change here.22:16This is the day to day business for your operations team.22:36Stage two becomes the active stage where maybe you get that 1st GPU server, you take that idea and you start to work with other teams on how can we get this idea into an actual production application.22:41And that becomes stage three, where you're now using use of validated designs and infrastructure that can support your idea for this new application. You're wrapping security around it as part of your 1st AI cluster.22:56And then this starts to evolve into systemic, where you now know how to use AI, you know how to scale that out, you know how to integrate it with existing applications or even how to use it to build new applications.23:15And then you reach stage five, just transformational where AI is just automatically part of the business and this is where you can see your ROI based on all of the AI investments that you've made. So breaking it down into these five stages will help you get started and all the way to where you can reach.23:31This is transformational stage.23:51So in order to put this into practice and to show how we can use the AI maturity model in action, where we can apply this to our portfolio, I'm gonna pass the ball over to Dave, he's gonna take us through on, on how we can merge all this together looking at our maturity model as well as.23:56Cisco's AI portfolio.24:16I'll pass it over to you Dave. Thank you very much Mark. Excellent. All right, so we're going to con you know go through the examples of what, what a flow through the the maturity model would look.24:19Like. So let's start with a a sample customer. So we've anonymized this, but this is a kind of sampling of several of the customers we've worked with to date. Imagine a large enterprise customer.24:39That wants ultimately to build a an AI shared services center of excellence, where basically all the business units come together and draw from a, a shared services model for.24:55AI workloads. The goal is to bring in all the the lines of business you know for an example HR, finance, legal, IT operations. So we'll start off with awareness.25:12So, you know, what does awareness look like? This is where we had, like basically where a lot of customers are today. There's interest in it. The executive have seen some demos of of AI possibilities out there.25:30Maybe some of the teams are, you know, working through their own, use cases, but there there's NO, you know, there, there's NO hint of an executive buy in yet. They're, there it's almost like stealth AI right now. If if a team learns some.25:48If the team earns some, experiences, they they there's not a lot of there's not defined lines of communication with other teams to share that share that knowledge.26:08So, you know, the it's and and at the same time, of course they're not probably not communicating with the infrastructure team to look at, you know, the lessons learned from, say, the data platform to get it ready for a full fledged AI deployment.26:23You know, from a infrastructure perspective, this is pretty much a traditional data center. We have some compute in there. Of course we're gonna be talking about, you know, UCS here. There's probably a mix of management platforms for the the compu.26:38Probably some of the older compute is still running say with Cisco UCS manager and maybe some of the new compute we've got into inter site already. Our networking is probably gonna be a mix of 1025, and a hundred gig, maybe a hint to 400 gig here and there on our spines.26:59We'll have some of the traditional storage vendors out there, you know, pure storage and apps, so block and file, pretty typical deployment for a lot of companies. And of course, you know, the, the th the dominant.27:18Virtualization layer here is probably gonna be VMware. Although this is kind of a a hot topic with a lot of customers these days, they're already coined with the idea of what to move on to. You know, if, if the licensing costs for instance, for VMware are.27:32Become a challenge, but this is pretty much where people are starting today. So this is, you know, we want to shift into the active phase here. So.27:50We want to start experimenting with a solution. This is you know the executive is curious, they want to see a good use case, they wanna see a success. They've made maybe a little bit of funds available, but they, you know, it not.28:05Many executives are going to open the wallet fully at this stage. Your knowledge may vary, but typically we're going to try and do this kind of on a shoestring budget. We are going to try and leverage as much of our existing infrastructure as possible, but one of the, you know, kind of one of the requirements is somewhere we've got.28:22Get access to a little bit of AI optimized hardware, be it we take, you know, we, we build or throw some GPs into a couple of existing servers or maybe we're gonna leverage our public cloud, but we're gonna go after one.28:41Good use case. We're looking for something that's low risk, we're looking for something that has a relatively sanitized, data source, if you will, and.28:57Something that should be able to show us fairly easily, you know, what, what AI can do for us. In this particular example, we're gonna be talking about a an an example workflow from HR for document processing. Think about documentation, it's all fairly standard.29:13File formats, fairly low risk and.29:33Yeah. So, but keep in mind, of course, we're, you know, this example is going to really focus on very limited funding. At the same time, this is an opportunity for our infrastructure team to start working on more operational aspects like standardization like a.29:38Scripting and and and making it easier to scale and deploy this new infrastructure. So with that idea, we're gonna start with as much existing hardware as possible. We are going to with.30:02Decide towards more scalable operations. We are going to standardize on Cisco intersight. We're gonna use as much existing hardware as possible. We're gonna try and leverage as much of the current network as we have. We're gonna just keep using the same traditional storage that we have. Again, we're just gonna focus on intersight.30:20What we're gonna do is we're going to throw in, actually my apologies, oh, I forgot the pointer.30:42Okay, so we are going to throw in a couple of GPUs into some compute and kind of the easy button here is to leverage the NVIDIA AI enterprise software suite, that will help.30:59Help develop and, and deploy, and adopt our AI journey. That is probably gonna be sitting on some kind of Kubernetes platform. We, Cisco is partnered with openshift, with Red Hat for OpenShift. That's one example of what we could use.31:17For the underlying virtualization and containerization layer.31:36And this is a great opportunity for our operations team to start, looking into orchestration tools like, like Ansible, Terraform, even just PowerShell to help deploy and standardize and script this deployment from the ground up.31:42Now, because this is an experiment, you know, we want, we need to pull, we need to be able to observe and pull telemetry from a lot of different layers to get an understanding of just, you know, what this journey looks like, what this this use case is going to look like to our existing and, you know, future infrastructure.32:03So this is where Cisco Splunk can come into play. So this can pull telemetry from all our compute and from our storage or networking. It can even pull in telemetry from the from OpenShift and from the AI enterprise suite to see how all these pieces are working together and help correlate this data into.32:22To dashboards and ultimately into reports to offer up to the executive to show the outcome of this active phase.32:42So we've we've worked through this, we've gotten to completion, we've been successful. The, you know, in this particular workflow, we saw 40 % time reduction in the time it takes to do our document processing, and at the same time we saw improved results from the AI being able to dig deeper and.32:53And pulling out more intelligence from this documentation.33:13Our infrastructure team had a fantastic opportunity to learn things. So they got to play with new shared tools and tool chains, automating this from the ground up, and they got to see, you know, faster deployment. We figured how to deploy one host. We now can deploy several of them by running a script. We've been able to standard.33:19Guys across this platform.33:37We with all the extra, with Splunk giving us all this extra telemetry and insight, we've shortened the mean time to repair. And through all of this, we got to look at our existing infrastructure and determine where the weak points are, where the bottlenecks are, and it helps.33:41Let's articulate that to management to justify taking, you know, new approaches to our storage and our networking.33:59So at the end of the day, the executive has seen the value, they've seen the improvement in this, this one low hanging fruit, use case, and they have chosen to fully fund the AI shared services center of excellence.34:07This is where we get to move into the operational stage. So, at this point we have a production platform deployed and it is following best practices, it is, it is designed to be scalable.34:26And we're bringing in more and more, use cases, more and more lines of business to benefit from this.34:45This is also an opportunity for us to, really dig into a major use case. So a major use case, for instance, would be say the finance team. They want a virtual assistant that is available 2047 and being a virtual assistant, if this is facing customers for an example.34:55That mean time to 1st token, that.35:14That that how interactive it is and how short the delays in response than that. That's gonna be a big selling point. That's gonna be a big threshold for us to hit. You know, when we look at the business constraints.35:18Governance and security gaps. This is gonna be, you know, this is gonna be a lot of work on our part. And there's gonna be a difficulty improving our AI return on investment. This is where we've got to work with the, the business teams, the lines of businesses to help define what that looks like going in and how that looks like after we've brought it in.35:35Into this center of excellence. And from infrastructure team, scaling beyond that 1st unit. Great, we've deployed it in product and we've deployed it in production, but.35:55You know, as we add more, we were going to have to expand beyond that. What does that look like? You know, how modular is our deployment? And as we add more, what is the, you know, how our operational environment is going to get more complex? How is that going to scale at the same time as our.36:07Our infrastructure scales. So this is where we get into the, the, the Cisco secure AI factory. So this is we've worked with Nvidia on this. So, you know, what, what is an AI factory?36:27The whole point of this platform is to get from our input, our data, and get through this infrastructure to get intelligence out of it. And ultimately it comes down to tokens. We need to generate tokens as quickly, safely, securely, and as cost effectively as possi.36:45So we've worked with NVIDIA to come up with this a a modular reference design that includes that that checks all the boxes here. This is intended as a full stack.37:11Solution including security and observability, and it's intended to be as a modular design, it is intended to be scaled as far as your facilities will will accept it. We start diving into the components of this.37:23So 1st we're looking at AI infrastructure. We're looking for an AI infrastructure that is again it's it's as full stack as we can make it. It'll include your AI software. This includes all your frameworks, libraries, model management, blueprints, all of that. And this also we're looking for workload and GPU orchestrator.37:39Option as part of the solution.37:59We're looking at a container platform, a Kubernetes platform that's gonna underpin all of this. We're looking at two network solutions. 1st, we're looking at a container networking solution that takes care of networking and load balancing, and we're also looking for a high performance ethernet's physical switching infrastructure.38:02From compute, we're looking at dense and modular GPU servers. And at the same time, we're looking for simplified manageability that can take it at scale. And of course we've got data storage where, you know, we're looking for high performance, scalability.38:22We need it to be secure, we need data protection as part of the solution. Alongside all of this, is this overarching, you know, security question. So I'll start in the middle here with workload and infrastructure security. Think of this as kind of your, your typical or your, you know, your firewall.38:38But we need to go much much further than that. Current, you know, typically firewalls can't scale to the levels that we need.38:59In this environment. So we're looking at distributed firewalls and next generation or next, next generation solutions. We're looking at security operations down here. So being able to, you know, provide actionable insights into.39:06You know, what our security is looks like. Oh, here we've got AI observability where again we are trying, we're pulling telemetry from all of this together so we can actually see it and, and, and provide intelligence from it.39:22And up here, back in AI security, we've got AI models and application security. This is probably one of the tougher ones to solve because we're not just looking at, you know, we're not just a typical firewall here where we're looking at what ports are open and make sure nobody's attacking the TCP handshake process. We actually need to.39:39See how the AI models are operating, how they're talking to each other, how the agents are operating, how the agents are pulling up data, and how the agents are providing responses to other agents and to our end users. This is a challenging one that I will touch on later to solve.39:59So how is Cisco helping you with this journey? So is, you know, as part of the what the work we've done with NVIDIA, this is how we're solving it. So with the AI infrastructure part, we have our, we have our Cisco AI pods.40:19This is intended as a full stack solution, that includes NVIDIA, the AI enterprise suite of software, as well as NVIDIA run AI, which provides workload and GPU orchestration, Kubernetes platforms.40:35You have an option, we have, we have options that you can deploy with. Of course, we've got red Hat OpenShift in there. We've got Nutanix and KP, and you can also run your own Kubernetes if you want on top of your own preverted operating system. Cisco AR networking.40:51So we have our physical solutions. There's kind of two approaches here depending on how you want to manage it. We have our on premises solution, which is you can manage your ne your your switches with Nexus dashboard. And we also have a cloud based switching platform or switch management platform called nexus.41:08This is hyper fabric.41:28When it comes to actual container networking, we offer the ISO Balant enterprise container networking solution. Cisco compute, of course, so Cisco has a Cisco UCS platform. We have models that are built on the NVIDIA HTX and MGX.41:29Reference designs, and we also support the RTX Pro GPUs.41:49As you, you know, depending on your workloads and your budgets and all of that we we can offer NVIDIA bluefield dpus. So these are, you know, super nicks or beyond that actually. So these actually include.41:57Embedded processors on them for distributed workloads, as well as for instance, in north south from an AI pod, you might be looking to leverage these to accelerate storage access, that sort of thing. And at the end of the day, all of this will be managed through Cisco Intersight.42:14And part of this, we've worked with partners to come up with lots of options, you know, from NetApp pure storage, fast data, which is one of the new AI optimized storage vendors out there, Hitachi and Nutanix and US.42:33From AI security, we have isovelant for your container security. We've got hypershields from our security team as a kind of a dis a distributed security platform. We've got Cisco hybrid mesh firewall.42:48We're gonna be using Splunk enterprise security to monitor all of this. And we have Cisco AI defense, which is a quite a large and barely unique solution, I believe that I'll touch on in a little bit. Just to recap, so the team you're, you're listening to today is.43:05Compute networking, we're, we're data center people. When it comes to the security solutions, that's another team. If you want to go deeper on those, I recommend you talk to your account teams and get some, you know, get some FaceTime with those teams. I'm going to touch these at a very high level, but there's only so deepaking.43:25So, and, for observe being able to get visibility into how all of these pieces are working together, again, we're going to be deploying Splunk.43:45Or we can offer Splunk. And deployment options, we have, you know, build your own. So this is, you know, if you want to build it, your own pieces and bring them together yourself, totally fine. If you want a more.43:56Validated approach. We have two options here. So we have our AI pods with on premises network management. So this would be our, you know, Nexus switches with, with Nexus dashboard. This is full stack, you buy it and you deploy it.44:15And this is gonna be backed by our Cisco validated designs, lots of documents online about how, you know, what we've done to validate these designs, bandwidth configurations, that sort of thing. And then we have an option of, AI pods with cloud based network management. So this is gonna be again our our Cisco.44:30Switches with with the hyper fabric. Likewise, full stack, buy it and deploy it.44:51So I have a couple of demos to show you. Let me. So I'm just gonna show you a couple of our products that fit into this framework. I'm going to be going at this at a very high level.45:00So at, you know, after the demos, I'll also remind you lots we actually have specific, you know, specific sessions that will dive deeper into some of these topics. So.45:17Start off with just intersight. So this is our compute, centralized management option, a management and monitoring option. So we have it includes things like dashboards to be able to see the health of our servers, you know, from here we can see things like our, you know, what's firmware is installed on.45:33Our servers, what those server models are, see the health of the servers themselves, all of that.45:52There's a the operational layer, so this is where we can, you know, we can, we can see the servers, we can check their inventory, we can see, you know, see their health and everything. This even allows us to see the, the, you know, potentially the hardware build.46:00From, you know, multiple aspects front, we can see the the rear of the server here with the GPUs installed. We can see the top of the CPU or top of the host, and this includes health overlay. So if there were say a dimm fails, it would show up as red here.46:17We see the health of the fans and all of that. Another thing we can do here if we dig into a server, and I think this is probably one of the part things that many companies.46:38Wrestle with the most. Like 1st of all, we can see inventory of all our hosts, we can see what, you know, network adapters and all our memory that's installed, all of that. But we also include a hardware compatibility list. So this is drawing on lots of.46:51Lots of external sources to help see, you know, are we running supported software or we running supported drivers? Is the firmware on our various hardware, is our firmware acceptable to the software that we're running? In this particular example, we can.47:08See here that the validation has actually failed on our, on our RAID and network drivers.47:27And this can even show us, you know, ok, there's our, our rate adapter. We're running 07:28 and 07:30 is actually the recommended version. So this tells us we need to look into that. Likewise, our ethernet drivers, we're running an old 1.1 oh 35.47:34And it's actually not qualified anything less than two oh 17. So we can see right away that, you know, we may not.47:54Have the most stable platform until we take care of this. And there's lots of different operating systems that are included in here. It covers a lot of different versions of code. So this is probably one of the the biggest value prop propositions for intersight.48:03We can even see into chassis, see how the chassis are physically deployed. We can see the inventory in them, you know, PCI knows so there's this PCI card installed in here. We can see right down into, you know, what the connectivity looks like.48:22So, you know, from our fabric interconnect into the integrated or the intelligent fabric module in the back of the chassis, we can see that there's a link here and we can even click on and see the telemetry we can get from that link. Likewise, we've got all these links across the backplane into the servers, we can see how those go.48:41How those are virtually wired in the backplane, all of that.49:01This also allows us to do like to explore so we can get lots of metrics across this and actually set thresholds and have it alarm if, say, network interfaces are are getting too busy, that sort of thing. And we can also do some automation.49:06So one thing also I want to touch on, there are storage vendors supported in this. So there's NetApp and Pure and Hitachi today. We hope to see more in the future, but this actually allows us to integrate our storage into intersite.49:23So we can do things like C inventory, we can actually, you know, look at the disks that are present in, in our storage. We can see what our various data protection policies are, what our snapshot policies are, and.49:40This then takes us into our workflow capabilities, so we can actually offer automation inside intercite. And this allows us across the stack to be able to do, like have workflows that cover multiple components, not just our compute side. So this example here, we've got this data.49:57Flow for deploying a new VMFS data store where it actually steps onto the storage, creates a new storage volume.50:17Creates a lun. It actually then goes forward and maps that lun into your hypervisors and flows, you know, creates the VMFS file system on there and takes you hopefully to success or failure.50:25And, you know, that's if you want to use this particular automation option. Intersight has a full API to it, so, and there are several SDKs available, so you can throw, you know, Python scripts to automate your deployments. You can, you can use Ansible and.50:41Terraform with we have lots of options online for how you can automate this entire platform.51:01Next I'm gonna just breeze over Splunk, what Splunk can do for you. So here's an example of Splunk, and the whole idea is that Splunk is going to give you visibility into your entire infrastructure from one place.51:09So, you know, as an example, we've got our A AI pod overview here. We can see, you know, GPU temperatures, how are our GPUs being utilized? How is our GPU memory being utilized? What's the cost of this platform?51:23You know over a rolling 31 day period, power consumption, all of that. It'll even give us insights into Red Hat openshift, so we can actually see, you know, how is, how is OpenShift operating? Are there any choke points? You know, how much memory is it consuming? All that.51:39But.51:59You see, you know, intersight, it can pull telemetry from Intersight and bring that together into this tool. Next is switching. We can even dive into your local, your large language models and how are we deploying those? What are we running? How are, how is our KV cache utilization looking? We bring all that together.52:02Here, even pulling information from our, our vector databases, how are they operating? What kind of latency are we looking at? Do we need to scale that further? We can see all of that from Splunk.52:21And then, so I, as my last demo here, I have, this is a combination of networking and security demo. So.52:36This is an example of, so we have AI infrastructure deployed, and we are seeing, you know, our Nexus dashboard has noticed that there is a problem here. We have some critical concerns in here.52:52If it's gonna work today, just a moment here. The demo gods are not being friendly to me today.53:13Okay, there we go. My apologies. So, we have, we, we saw that we had a connectivity problem, so we can dive into that. We can pull an analysis, we can have Nexus Dashboard do an analysis.53:29And we see here, we have a, we have, we can see a data flow from our AI pods up through some these are the, these are the smart switches I touched on earlier. These are our top of rack switches. Traffic is going up and over and it's passing through a 3rd party firewall.53:45All here, and we see here traffic getting to the 3rd party firewall is fine, but it's traffic coming out that is showing issues of latency and we're seeing 25 % packet drop. At the same time, we're seeing that there's 98 % CPU usage on this 3rd party firewall.54:05And knowing that we have these intelligence switches, these these smart switches in here, so these smart switches include dpus inside of them. They, these are basically embedded multi core processors in there where we can run relatively arbitrary workloads on them to help.54:22Help, with things like.54:42Distributed security enforce visibility and enforcement. So what we can do here is we can go into secure cloud control. So this is a Cisco hypershield. This is a security product and the idea here is that this is your dashboard into a fully distributed sec.54:44Security portal into your infrastructure. So we can look at our AI pods access rule set, and we see here, here's the rule set, how it's designed and it's passing through this 3rd party firewall. So we click on there.55:04We can say install add install targets. What we're doing is we're actually expanding this, now we're taking this rule set and we're going to expand it to, more architecturally relevant points in our network. So we can expand, we can add targets here. So you'll see here we had that 3rd party firewall where at.55:22Adding two sets of upstream leaves. So these are the leave the sets of leaves that are above our AI pods and above our legacy compute platform. So we're done.55:42We've installed those and now we can deploy those rule sets to those switches. We'll see here now the AI pods access rule set is now deployed on those switches. Now we can go over to Splunk.55:59And we can verify that we see traffic AI pod to our vector database as an example is now flowing through the smart switches. So we see that, you know, we have optimized this data flow. So this is our, you know, what our previous analysis showed us. We can rerun our.56:15Analysis.56:35Now you see that traffic from our AI pods are flowing up through our smart leaves and straight to our legacy apps where vector databases and API gateways are. And you'll actually see here little, little bricks, little brick walls on these leaves, that shows that they are now running a sec security workload. We'll see.56:36Over here that we now have over 800 gig.56:56Of throughput capable. We even dive deeper into this and see, you know, exactly what the protocol distribution is across our two dpus in these smart switches and, and whatnot.56:59So, that's the end of that demo. Again, if you want to learn more about the security portfolio, I recommend you talk to your account team, and they should be able to line something up for you.57:19And with that, to do some networking demos, I'm going to hand it off to Mark. Awesome. Thank you Dave and great demo and representation of the different ways that we can use our, our hardware using intersight.57:36Using the secure factory and then showing through a demo how how it all ties together. So up next we'll focus on more of the the networking piece. So this is where things really start to differentiate compared to standard data center networking.57:56So here we can see an example of a typical AI network. Starting up top, you have the front end network. So this is where your users are coming in, your agents, your telemetry, this is where your, the chat prompts live. This is where we put tokens into a prompt, and then this is where we're sending them in and.58:16To this front end network and.58:36Just as we spoke about earlier, we're seeing this evolution of things going from a chat prompt to now becoming more agentic. So this is where we're starting to see these front end networks are now using 25 gig, 50 gig, hundred gig interfaces to be able to support the amount of load that's coming in through that front end network.58:39Then in the middle here we have our servers, so these could be your C eight 80 fives, e.g., where these are C GPU dense servers that could have up to eight GPUs per server. So this is where all of the GPU magic happens on on the back end network.59:00For each one of these GPUs, they actually each have their own NIC for each GPU.59:19So on the back end network, it needs to support lossless, high throughput, low jitter, low latency, to be able to use things like rocky V two or RDMA over converged Ethernet two, to be able to take what's in memory for these GPUs and then exchange that me.59:25Memory between all of them.59:46So whenever we're sending tokens in and these GPUs have to calculate that, they do something called a collective communication, basically where all of these GPUs will start communicating at the same time. So it's not uncommon to see these uplinks that are in the 400 gig, 800 gig.59:47To hit full capacity. The GPUs have to be able to interact as if they were just one big GPU. So that's why we have to form this lossless network. That way they can all communicate together across all of the different networks.01:00:07And this is where we can use things like Nexus dashboard to manage all of these different networks. So, so far we've only talked about a front end and a back end network. On top of this, there's also going to be a storage network. There's also going to be a management network that's gonna have all of your connections to your, your.01:00:25Sees your BMC ports, everything that's used for management. And with Nexus dashboard is how we can configure, manage all of these pieces from, from one place. There's also options where we can use AI fabric templates that have all of the.01:00:45Best AI practices and QoS settings built into that template, so that way you can use that template once, deploy it many times for support for low latency, intelligent buffering, rocky V two.01:01:04On the backend, additional features such as intelligent packet flow, and these are all built using validated designs, so they've been built, benchmarked, tested. We know what these AI networks are, are capable of handling.01:01:20So from here, let's actually jump over and s into nexus dashboard and see what it looks like alright so we're logged in.01:01:39The Nexus dashboard.01:01:59Now on our homepage, we get an overview of all of the fabrics that have been onboarded. So in the previous demo, Dave showed where we were looking at the Nexus dashboard, analysis piece. Now we're gonna take a look at what's known as the Nexus dashboard fabric controller where we can use Nexus dashboard.01:02:00To manage and monitor all of the devices in a fabric. So at 1st glance, we get a high level overview of anomaly levels, so Nexus dashboard will learn what is healthy in the network. And then as things start to deviate, maybe we get some port flapping or some spikes in the network, we'll tag these as anomaly.01:02:20For you to start investigating. There's also support for advisory levels, so here's where we can find things on P certs.01:02:40Field notices, vves that are specifically tied to the hardware that's being managed by, by Nexus dashboard. And then just a high level overview of how many fabrics are onboarded, number of switches, number of active endpoints.01:02:48So in the topology view, we can see how everything is logically connected. So here we can see I've got two spines connected to four leaves and how they're all connected to each other. We can select any one of these devices and get a high level overview of the health status of that device. We can see.01:03:08The information around IP address, serial numbers, what's the attended role for that device, model number versioning. So when you have these large GPU clusters, Nexus dashboard makes it really simple to look at everything from, from one place.01:03:28You can even double click into one of these leaves and then look at all of the endpoints that are connected to that leaf so we can see IP addresses, MAC addresses that are all connected to this individual leaf here. From here we can go and look at the details.01:03:46Of our AI fabric, and we can see that it was used with a, a template for an AI VXLAN EVPN with IBGP. So this template has all of the settings already built into it to support an AI fabric.01:04:06And we can see the interface statuses of interfaces that are either up, down or possibly admin down. We can also get an overview of hardware resources on the cluster. So for all the devices that are in this fabric, we can look at CPU information from one place. I don't have to log.01:04:21Into individual devices and then try to map out what does that CPU look like across all of those devices. I get all of that in one view here. We can also see things like memory, power supply information, temperature, fan utilization and storage.01:04:41Another nice feature is that we have the option to go back in time. So this is something like a, a network DVR where I can look at EPU utilization from the last 6 h or the last week or specific date and time. So this allows me to go back in time to look at.01:05:01What does resource utilization look like? Maybe not the current state, but what did it look like maybe a day or ago or a week ago so from here.01:05:19You can see some high level overview of connectivity, you can see conformance scores congestion scores. So congestion's gonna tell us, is there anything going on the network that could be dropping packets, if priority flow control taking in to put a pause on traffic? Are we seeing anything around.01:05:34Explicit congestion notification being triggered, then we can even see just the amount of external traffic that's coming in and out of the fabric under inventory, we can see all of the devices that are part of this fabric. So we see a high level of the health for its anomalies, IP information.01:05:54Model numbers and then a feature here called configuration sync status. So when we deploy.01:06:15And manages manage these devices, we use policies and apply these policies to those devices. If there were some type of configuration change or maybe additional information was added to the configuration or maybe even removed from the configuration will get a nice out of sync status here to let us know that something is.01:06:21Change within our configuration. So when you're managing these large AI clusters, you want to ensure that every device, every port, every interface is, is all configured the same way.01:06:41So we can even drill down into a device and get additional information on the status of all the ports. We can see things like anomaly levels, general port status information, if it's in sync with our config status, software versions, serial numbers, etc.01:06:54Uptime. And then we can even drill down into each port and we can get port level stats from here. We can see the operational speed of the port, things like admin status operational status, and then under transit stats here we can see.01:07:15The status of the port over time, but we can see here there's a blue bar here and this is telling us that real time values are being shown to the right of this blue bar here. So using our our ASICs, we're able to pull that telemetry in real time straight into Nexus dashboard.01:07:32You can also see things on the interface error counts, tracking traffic information for flooding traffic multicast, unicast, total traffic, as well as bandwidth utilization, and down below is here where we start to get into that congestion score.01:07:52So here's where it'll tell us if we're starting to see any type of drops on that interface, if priority flow control is kicking in to send out any pause frames is explicit.01:08:12Ingestion notification kicking in to help slow down some of that traffic. So the tools that you need to be able to support things like rocky V two, to be able to have that visibility into all of the metrics to make sure that your, your AI fabric is, is running healthy.01:08:25So here, what we're talking about is with nexus dashboards, this is your your on prem deployment. So these are, these can be physical controllers, it can be virtual devices that can be used for Nexus dashboard. So as Dave mentioned, the other version of that is what's known as Nexus hyper fabric. So what's different about.01:08:47About Nexus hyper fabric is that this is a cloud based controller as opposed to Nexus dashboard, which is your on prem controller.01:09:07So with the hyper fabric, all of your controller management is done in the cloud where you can create different types of blueprints and do all of your configuration directly from Nexus hyper fabric. That's running in the cloud.01:09:17So very similar, we can see different environments that have been onboarded here, three different fabrics, so all using the same cloud based controller here. And we can drill down into each one of these and look at the, the topology, if you hear we've got two spines, four leafs. We can also see how they're.01:09:35All connected from the spine over to the leafs and which physical interfaces they're using.01:09:55We scroll down, we get general inventory information, health status on each device, the model name, the serial number, port counts for, for each device. So what's different about a hyper fabric is.01:10:02When you go to do your configuration, it uses a little bit more of a simpler approach where if I want to configure an interface I can say this interface is gonna be used for a host that's gonna connect down to a server or I can set the port role to maybe a fabric port for another device that's gonna connect to in the network or I can set it to maybe.01:10:18A routed port for a routed device.01:10:38So it's, it's very straightforward on defining what you want that interface or what you want that device to do. One thing that's also different about hyper fabric is typically in a network you want to know maybe when a, a port is down, but with hyperfabric we use things called.01:10:41Assertions where, what if a port came up that it wasn't intended to come up? It will we even capture what we call assertions so that you can identify when something gets plugged into the network, but maybe there wasn't an intention for something to be plugged into that network.01:11:01And then one of the other features you can do with high fabric is that you can start to build out these topologies actually ahead of time, even if you don't have the hardware in front of you. So we can go to add new fabric and we can choose a new fabric from template. And so here we have templates.01:11:19That are already built out for a two tier architectures, three tier architectures, whether if you want two spines for lease, maybe you're just getting started all the way up to a three tier three tier architecture.01:11:39And, and even above that. So we can simply select the template that we wanna start with, give it a name, here we can see everything that's gonna be includ.01:11:52Included with this template here. So we have our, our two spines.01:12:12We have for leaves, we can even modify and add additional tiers to this template so we can select server and let's say we want to include aC8 8885, which is one of those GPU dense eight GPU servers that could be used for, for model training.01:12:17So we simply just select the options here, you can even get a preview of the cabling, so here's what the cabling would look like when you're ready to cable up this hardware once it arrives on site. And then we can save our our fabric blueprint.01:12:39But I forgot something update our our server count here so we can save our our blueprint.01:12:59Then we can actually go to the topology, see how it's cabled up, all of the devices that we created from that template. And then from here we can even start making configurations. So we can go into edit mode, and let's say we're gonna use.01:13:19This 1st port to connect.01:13:39To our server, we can do things where we can set this ahead of time so I can edit the port role and we're gonna set this to host because it's gonna connect down to our server.01:13:43So after you've made your changes, we can do a review configuration of what has changed. And then we can do a comment and push to actually apply this configuration to our topology. And at this point, we don't have any hardware, we're just setting all of this up ahead of time.01:14:02So once the hardware actually arrives on site, we can then take our logical configuration and then bind it to the actual hardware that's on site and then all of the configurations that we have a blueprint gets binded to the physical hardware, so you can pre stage all of this ahead of time using hyper fabric.01:14:24And with that to continue on with the maturity model, I will pass it back to you Dave. Thank you very much.01:14:48So, let me where we are. So just while it's topical, I just wanted to throw this up again. So we have four sessions coming up over the next next four weeks that are gonna dive deeper into some of these topics that we had showed us demo.01:15:06So next week is gonna be networking for AI so deeper into Nexus dashboard and.01:15:26Hydrofabric. The week after that, we're gonna be digging into AI pods and use intersight. And then we get into AI use cases and then we start talking about AI AI. So again that, that.01:15:34QR code can take you right to the registration page. Time is getting short, so I'm just gonna bang on here. So, you know, when we, we have to, you know, as part of this conversation, we have to look at how our enterprises using.01:15:50In AI.01:16:10So, you know, we start here, we've got our user. Now this could be an employee, this could be a customer, it could be just someone in the general public. They're gonna be talking to some kind of agent, and that agent is going to be pulling information from some, you know, databases somewhere, and that agent may also be talking to other agents, and those agents may.01:16:11Might be talking to MCP servers that are pulling up all kinds of.01:16:31Other data, including like, you know, the, the web searches that are retrieving unsanitized data from the internet, that sort of thing. And when we look at this, we have to ask ourselves, where might we be vulnerable?01:16:37And unfortunately, the answer is pretty much everywhere. So this is just kind of an eye chart that just kind of, you know, highlights security risks and safety risks across this entire AI workflow.01:16:54And you you have to ask yourself, how are we going to secure all of that? We can't, you know, it's not just a matter of a firewall blocking ports between say the user and the agent. We want to be able to see the query the user is sending into that agent and we want to be able to see the.01:17:09Response from the agent to make sure that they haven't broken through any guardrails or anything like that. You know, the, if we don't, there can be a considerable risk. Financial damages, reputation harm, intellectual property and personally identifiable and identifiable information.01:17:27Leakage and that sort of thing.01:17:47So this is where, Cisco AI defense comes in. So this is a three step framework for developing secure AI applications. So we can discover your AI assets. We can do, we can actually do detection and look at those AI.01:17:50Assets and see, you know, do they have proper guardrails, that sort of thing. And then regardless of how secure those assets are, we can add an extra layer of protection. So defense at depth. We can add an extra layer of of guardrails if you will.01:18:11To that a, your AI infrastructure. And at the end of the day, all of this is again managed by Cisco Security Cloud control. The same tool that was that I showed off in the hypershield.01:18:29Demo. Little bit more in depth, you know, we, the AI cloud visibility. So your AI models, your agents are going to be calling on lots of other agents possibly, you know, some of your developers have done cloud calls, that sort of thing. We can see all of that and help understand their usage and gauge their.01:18:44Supply chain risk management. So we can, we can actually scan those model files and the repos and the MCP servers. So if you think about, you know, if you're not a multi billion dollar company, you are probably not going to be training your own model, you are going to be downloading a pre trained model from plugging face.01:19:05How sure are you that there is not an embedded executable in that model?01:19:24Their actual videos online where it's been demonstrated that you can embed executable model execution executables inside these models you're downloading from public resources. So how, how can you trust them? This is one of the tools that can get you there.01:19:29AI model and app validation. So one of the, one of the, you know, things that happen, one of the things things we see is every time you you you update an in a model if you do your own fine tuning on a model, how do you know that you have an.01:19:46Actually broken the guardrails that were built into it. How do you know that that thing's not going to teach you how to build a grenade? Typically, every time some every time a company.01:20:05Fine tunes a model or downloads a new model or or builds a new application. They have to hand it off to a red, a red team and they will spend potentially weeks banging away at it, trying to break through, you know, creatively trying to break through the, the guardrails that are built into it. See if you can extract personal information, see.01:20:17If you can make it fall over and possibly even execute code.01:20:37This tool, AI defense can actually algorithmically automate that and bring down that red teaming exercise down to hours if maybe even minutes. And again, runtime protection, actually running your own guardrails on top of all of this.01:20:41This can also help with, say, if your employees are using public AI clouds to do queries from inside your organization, you could actually see those, intercept those and make sure they're not, you know, uploading customer data into a query to review a.01:21:00You know, a PO or something. So we're finally at the end of stage three. So we are in operation, we are in production. This is kind of what it's looking at. All lessons learned that we got from stage two. You'll see here like for instance, we've expanded.01:21:14With the compute cloud. We've got some AI pods in there both for, you know, inferencing. We also have some dense GPU servers in here for fine tuning and possibly training your own models. We've got stacks of servers in here. In this case, this is actually the, you know, one of the things they discovered during stage.01:21:34Too was that their their traditional storage was not fast enough and was not flexible enough to keep those GPUs busy in this infrastructure. So the idea is we've gone to, you know, for, for the AI workloads, we've gone to a vast storage.01:21:55Architecture. Again, this is all available through Cisco even though this is a partner of ours. We've upgraded our networking for the AI pods themselves, we've upgraded the networking to 400 gig and we're automating that through either, you know, the Nexus dashboard or through.01:22:13Hyper fabric.01:22:34And we're automating this all through, you know, Ansible Terraform, Intersight, PowerShell. At the same time, to get faster to realizing the value out of this infrastructure, we've let we can leverage Mint partners and Cisco CX to.01:22:36Help deploy, like how help design architect and deploy these solutions.01:22:56Hey Dave wanna interject here and give you a quick 5 min time check. Yes, thank you. And of course we're still using Splunk to watch and monitor all of this. So we get into systemic. So at this point, if this, you know, AI is being used for everything that we.01:23:03Can all new projects are at least considering the use of AI.01:23:21And from an infrastructure perspective, you know, we are AI optimized. We are now able to just cookie cutter scale our platform up to support more workloads and more use cases. You know, and one of the you know because this is cookie cutter, it's easier for.01:23:26For us to just.01:23:46You know, add functionality, add capacity, we can just start plugging in things like public cloud and AWS, we can just start plugging that into our workloads. You know, we were, we're able to leverage those smart switches that, you know, like with built in dpus in them. We have a new product.01:23:48Unified edge for those, you know, if you have branches or retail outlets, you know, instead of that stack of random hardware, there's print spoolers and managing your atmiss and your point of sale systems.01:24:08We can automate and we we can we can we have a platform to take care of all of that that could be provisioned with as little as a, you know, whomever it is that actually rack and stacks this hardware, they can just swipe their phone.01:24:21Against one of these arms and through NFC, that can be added to your company's inter site portal, and it can automatically start automatically or up through automation, deploying firewall rules, deploying network configurations, configuring deploying operating systems, deploying applications, integrating.01:24:37Creating this into your greater AI infrastructure. That can all be done with the swipe of a phone across one of these arms with the NFC built into it.01:24:57So, at the end of that, we have, we're, we're, we're now transformational. AI is just part of the business now. You know, from an infrastructure, AI capacity is, is tied in with revenue potential.01:25:05One of the biggest constraints is just having revenue generating ideas. So, and with that, I will hand this off to Mark again.01:25:22Riley, do you want to take a minute to do our last question? Yeah, since we're running a little short on time, I think we go ahead and skip our Slido and just jump straight into the session survey and some live Q and A here. So I want to thank everybody once again for joining today, and I want to thank Dave and Mark for the very insightful session. I know I learned.01:25:41It's own.01:26:01But if anyone has the time to fill out a survey, we'd really appreciate any feedback and helps us improve our future offerings. And at this time you can also type in a live Q and A into the chat and our experts are happy to help.01:26:02We'll also say that the session survey results are anonymous, so.01:26:23Don't be shy. Looks like we have a question to the panelists here. I think it's a true six month lead times?01:26:27My advice here would be to work directly with your account team to get more, more specific lead times that they would have the best answer.01:26:49Well, it looks like we have reached our time. Once again I want to thank all of our experts for joining today and everybody for attending. This was our extra insight series session on AI maturity and Cisco AI tech stack. Thanks.01:27:01