Red Hat Summit 2024 | Chris Wright | Red Hat

Clips
News
More from Red Hat Summit 2024

Chris Wright

Chief Technology Officer and Senior Vice President, Global Engineering

Red Hat

Three insights you might have missed from theCUBE’s coverage of Red Hat Summit

There’s no shortage of excitement around the possibilities that surround AI and open-source technology. Automation is playing a huge role for AI, including when it comes to bridging the world of hybrid cloud.Exploring the future of hybrid cloud and open-source technology within Red Hat Inc.’s ecosystem was a central focus of the recent Red Hat Summit. This was also the second year that AnsibleFest was held in the same place as the summit.The overall takeaway from the event was that it represented a journey, according to Bob Laliberte, principal analyst at theCUBE Research, in an analysis of the event. There’s no one product that’s going to be a destination, he noted

The strategic impact of technological partnerships: Salesforce and Red Hat’s collaboration on enterprise innovation

In the evolving landscape of technological partnerships and infrastructure management, organizations are continuously seeking strategic advantages through collaboration with industry leaders.This is exemplified in the recent decision by Salesforce Inc. to strengthen and extend its existing relationships with Red Hat Inc., particularly in the transition to Red Hat Enterprise Linux 9

play_circle_outline Red Hat's partnership with Salesforce and their decision to stick with Red Hat and move to RHEL 9

play_circle_outline Resources and automation tools for customers moving to RHEL 9

play_circle_outline The role of AI in RHEL, OpenShift, and Ansible for productivity gains

play_circle_outline The Kepler project for measuring power demands and sustainability in cloud-native applications

Info

Chris Wright | Red Hat

Chris Wright

Chief Technology Officer and Senior Vice President, Global Engineering Red Hat

search

>> Hello, and welcome back to Red Hat Summit and AnsibleFest day two are where you're going to watch the cubes wall-to-wall coverage of the entire event. We're going to bring on the most interesting speakers, really dive into the announcements and really peel back what is interesting about those announcements to help you understand. I'm Rob Strechay. I'm joined by Paul Gillin. Rebecca Knight is here floating around. She'll be on real soon too. But right now, we're so excited to have Red Hat's CTO and SVP of Global Engineering, Chris Wright joining us on the show, right fresh from your keynote. Basically, a little water, a little hydration in between, the hike between the two, always getting your exercise. So thank you for joining us.

Chris Wright

>> Glad to be here.

>> So I think really one of the interesting things that watching the keynote was that, again, Salesforce coming up. I mean, again, I know we were talking a little bit in the lead-up. That was just an amazing client going through and their infrastructure is just massive.

Chris Wright

>> It is. Yeah.

>> And they have SLAs on everything. It's got to be up, it's got to be really tight. CentOS customer, they got to a point where they had to move. So they're looking at, okay, do we stay with Red Hat, do we move off. And they chose to stay with Red Hat and move to RHEL 9. Help us understand really where they saw the value of that and the value of sticking it out with Red Hat and really moving into RHEL.

Chris Wright

>> Well, I might be a little biased-

>> Yes, that's okay.

Chris Wright

>> Since I got a lot of the engineering organization with me, but a big part of that, the way they framed it was strategic partnership. And if you peel that back, the next piece under that was connection to the engineering experts that not only can support something like a complex tool like an operating system, but are actually the authors of a lot of the content in the operating system. And so they're looking at where do we get that level of expertise so we don't have to build it in-house. It's like a perfect partnership, a perfect relationship, and their view was as important as the operating system is, and they're operating at massive scale and doing a mind-numbing number of updates and managing this huge fleet. Actually, what they want to focus on is the higher-level value that they're delivering to their customers so that when they can partner with us, they can just rest easy at night knowing that we're right there and focus on the things that are most important to them.

Paul Gillin

>> What kind of resources is Red Hat throwing at this project? I mean, you've got 20,000 VMs, they're moving over.

Chris Wright

>> Well, for us it's a pretty natural relationship. We have a very mature platform. So it doesn't take a huge number of resources. We have a lot of automation and tools that helps customers move content or move workloads from one place to another. You can do in-place upgrades for things like CentOS to RHEL 9. So it is mostly about that continued long-term relationship and looking where are we going together into the future? They're really intentional about their technology adoption and think about cloud-native technologies, and they love this idea about immutable images and image mode for RHEL. It really matches their workflow and their needs. So it's not only about what can we do today to help them move, but it's that whole big picture.

>> So I think that was a great segue into one of my other questions that we kind of chatted about, which is where does image mode fit in the stack for customers who are out there trying to understand, okay, do I go bare metal, do I go something else in between that and OpenShift. Where does image mode fit and what are the use cases that organizations will thrive on with image mode?

Chris Wright

>> If you think about an operating system and traditional OS, it's RPM-based, you update packages. That model fits anywhere. Image mode fits anywhere. It's just a different workflow and a different model. So think about building content. You want to deliver a bootable image to a server. That content will start with a trusted base, the operating system. You're always going to add something in there. It's an agent. It's something that's unique needs from your enterprise point of view. If you have a truly immutable image, you can't add that agent in there. And if you're using something like an RPM-based workflow somewhere, somebody's going to log in, do some kind of a package update, and you've now got some kind of drift or shift from the primary golden image. So that's useful down to the bare metal, which is really critical. If you're trying to get direct access to accelerators, you're pushing this out to the edge, but that same image can, of course, boot in a VM, and that same image can be the base upon which you build your OpenShift cluster. So I like to say it's turtles all the way down it. It's a useful tool, independent, the use case. Now it's a shift in mindset. So you also have to think about the best application is thinking about how you're cloud-native, how you want to get to the most simplified and rapid and consistent methodology for deploying content.

Paul Gillin

>> One thing that's been themed throughout the show is AI everywhere, and you're driving AI throughout all aspects of RHEL and OpenShift, and Ansible, now where do you think the customers are going to see the greatest productivity gains from these innovations?

Chris Wright

>> AI is touching everything. Now. The way we look at it is it's a critical workload that we want to support on top of our platforms, hence the need to build support for the accelerators. And at the bottom of the stack and something like image mode, you can build this accelerator-adapted image, so it works well on NVIDIA GPUs or a different image, which will work well on Intel accelerators or AMD accelerators with all the application content going up the stack. So there's that workload. It's a workload point of view. Then there's infuse it into our portfolio for helping our customers be more productive with our tools. And then that productivity side, are you helping developers write code? Are you helping content authors create other content? Are you bringing debugging skills to do a better job of root-cause analysis? I mean, the use cases are innumerable, and what we're going through right now is an industry is sorting where are the best starting points so you show early wins and then kind of keep growing from there.

>> Yeah, I think what was really interesting in all of the keynotes, and Ashesh had his and where he talked about some of the AI being used, and we had some really neat discussions yesterday as well around conveyor and app modernization, and you kind of touched on that as well. And I guess we're told you kind of have a theory about how that's going to evolve from an app perspective and a model perspective going forward. Why don't you kind of delve into that a little bit about what ... Is it going to be a hundred percent app modernization or is it, hey, some, I guess you could say we have the concept of building what we call data apps. And these data apps will include models inside them that will be part of larger applications and it's kind of a tiering of these as people go to these cloud native applications? How do you see that and how are organizations thinking about it when you're talking to them?

Chris Wright

>> Well, first of all, early stages in the technology. So everybody's at a different point. Some people are just trying to figure out which way is up. Some people are pretty deep into some deployments and really learning. The way I see it is everything we've done to date for enterprises is supporting the development and delivery of applications. And enterprises are technology companies, they're software companies. They're differentiating themselves as software. Rolling into the future, AI models are a critical part of an enterprise, and I'd say it's a 50/50 split of applications and models that power the enterprise. And when you look at modernizing applications, which generally means following cloud-native principles landing in a container, you may be doing app modernization and discover that you have a critical dependency. It could be a database or some portion of your application that doesn't naturally easily modernize. You can leave it in a VM and bring that over to a modern platform and run your containerized workloads, your virtualized workloads, your AI workloads that need access to accelerators. The application will have integration to the model through just simple APIs. So developers are building intelligence into their applications, and that's the future of the enterprise. So we're trying to make sure that can all fit on the same platform so you have the operational experiences and target the same platform for development.

Paul Gillin

>> That's a pretty big shift. You're going from all applications to half applications, half models. How should customers prepare for that world?

Chris Wright

>> I love the question. So part of it, and it's what I was talking about in the keynote, is one way you can prepare is make sure you lay the right foundation. So build the right platform support for all those different workloads. And then the other way is start bringing in experimentation. You can start as small as you need to learn what's valuable for the enterprise in the AI space. And as you discover something that's useful, you got to figure out how to scale it into production. So you start with the experimentation, you figure out how to put that into a repeatable pipeline so you can deliver the model, repeat the model evolution, and make sure that you're maintaining that integration with the application. So it's like start small, find those wins, and then just keep evolving.

>> Yeah, I think we delved into this at length yesterday and what I liked, and actually Chris Morgan's demo yesterday with Mo and the others on stage was fantastic ... I'm like, "You should package that up." Because it explained it sounds in a way that I thought, Hey, here's where you start with Podman and InstructLab. Here's how you then go to RHEL. Here's then how you go and deploy into production on OpenShift AI and how you take it across RHEL AI and all of these different pieces. What I also like about it, and we're kind of open-source wonks, right? We believe open source is the way, and we loved the Granite open-sourcing announcements that were done of the model and of the code assistant. Where do you see it fitting? I mean, I know our perspective, but I want to get your perspective, and I have a funny feeling open is the way, but how do you see the two working together? Because I think there are ... There's the open source alliance or the AI alliance, which is really pushing on the open source, but then there's government regulations and other things coming. How do you see this all playing out with open source and AI?

Chris Wright

>> More open, more better. So we're definitely going to keep pushing on that open direction. And as you think about what that enables, the broader community support and activity you have around a common piece of technology, the more ubiquitous it's accepted across the industry. So think about something like Linux. It's a de facto standard. Kubernetes, it's a de facto standard. And for one company to try to build a technology that would compete with those, it's just not really possible. The same thing's going to play out in this AI model space. And the other thing that's really important is big large models are difficult to create. They're expensive to run, and they give you more data, more parameters, you get better results. You can also fine tune a model, keep it small, efficient to run produce as good or better results because it's trained on a very specific use case. And I think you'll see that in open source and that proliferation of activity.

>> I think, in fact, we actually have ... We back in September, we put out from the research side of things, myself, Dave Vellante, John Furrier got together and kind of started to look at what was a power law distribution of models. And it had nothing to do with revenue or anything like that. It was size of models and addressability to verticals. And so really big high models in the front end here where you had your foundation models, but then stretching out over, Hey, I want a customized model for my customer service and in this industry and getting that exact with my intellectual property, are you seeing customers? I mean, InstructLab, to me, the light went off when I learned about it. I actually got a little preview, so I learned about it a week and a half ago, and I'm like, wow. I'm like, "This can help you." And there's some other people doing some very interesting thing in synthetic data and stuff like that to help. Do you see that as an enabler to have people more simply fine-tune models? Because I'm with you, and I think we're all with you that nobody's going to go compete with Llama 3 and Granite and build their own model. I mean, that just to me is-

Chris Wright

>> It's cost-prohibitive. It's very difficult.

>> Unless you're building a service that's going to compete with ChatGPT, which I don't know why you would do that. I mean, that's good for them, but I look at it as if your service is servicing your customers and your financial services company, do you see that this ... I mean, I know it came out of the paper from IBM Labs and it was really pulled together very quickly. We had Matt on talking about how he got so excited about how quickly it came together yesterday. Do you see this as a key piece in getting people to models everywhere to that 50% that you're talking about?

Chris Wright

>> Yeah, exactly. And that's why I put it at that number. There's not going to be one big model that runs the enterprise. There's not one big application that runs the enterprise. There's all these little pieces that you compose to run the enterprise. It's the same thing with models. And the difficulty has been how do you specializ, how do you fine tune. RAG is one tool, but infusing the data directly into the model, reducing the size of the model, it's a really effective way. And it's been hard to do that fine-tuning. So making it accessible, that's how you get that proliferation. And that's where you'll see thousands of models each embedded in an application providing that application's intelligence.

Paul Gillin

>> I want to ask you about a project you announced last year, Kepler, an open-source project for measuring the power demands of cloud-native applications. Of course, huge power demands for AI/ML. How are you evolving Kepler for those new needs?

Chris Wright

>> I love that question. Sustainability, it's a big topic. It's especially large in places like Europe where there's regulations and requirements and massive change in the energy dynamics in the market. So we have a lot of requests for how can you support visibility into understanding the power consumption of an application. So you can look at the watts consumed at a server level, but that doesn't give you a lot of insight to where those watts are going. So you take an application down view, Kepler gives you that insight. What we've been doing is working over the last year to expand that. So it's not just CPU-bound workloads like application and business logic, but it's also AI models, which themselves are power-hungry, sitting on GPUs. So now you can combine the insights from the application side, the insights from the AI model side to get visibility into your power consumption. And with visibility, you can think of that as performance. You have a performance SLA, you put it through your CI pipeline and make sure you don't regress. You have an SLA for power, you put it through your pipeline, you make sure you don't regress as you're making changes in your models or your applications. So getting that visibility is key to building more sustainable computing as we go forward.

>> I think that to me is a great place to leave it because I think to your point, we could sit here and talk about sustainability. I think it's the elephant that's not quite in the room in the US yet, but it is definitely in Europe. And I think with the AI Act and some of the other things that are over in Europe, and obviously, there's a lot more on the sustainability side that they've done there that hasn't quite gotten here. You have the two in California and Virginia, but other than that, really not there yet. But I think that's a great place for us to leave it until next time when we get you on.

Chris Wright

>> All right.

>> But thank you for coming on board, Chris. It's always great to get your insights, especially fresh off the keynote.

Chris Wright

>> Glad to be here. Thanks for having me.

>> Yes, and thank you for starting your day with day two at Red Hat Summit and AnsibleFest Live from sunny Denver. Stay tuned for more. Coming here from theCUBE, the leader in high-tech analysis and news.

Chris Wright | Red Hat

search

>> Hello, and welcome back to Red Hat Summit and AnsibleFest day two are where you're going to watch the cubes wall-to-wall coverage of the entire event. We're going to bring on the most interesting speakers, really dive into the announcements and really peel back what is interesting about those announcements to help you understand. I'm Rob Strechay. I'm joined by Paul Gillin. Rebecca Knight is here floating around. She'll be on real soon too. But right now, we're so excited to have Red Hat's CTO and SVP of Global Engineering, Chris Wright joining us on the show, right fresh from your keynote. Basically, a little water, a little hydration in between, the hike between the two, always getting your exercise. So thank you for joining us.

Chris Wright

>> Glad to be here.

>> So I think really one of the interesting things that watching the keynote was that, again, Salesforce coming up. I mean, again, I know we were talking a little bit in the lead-up. That was just an amazing client going through and their infrastructure is just massive.

Chris Wright

>> It is. Yeah.

>> And they have SLAs on everything. It's got to be up, it's got to be really tight. CentOS customer, they got to a point where they had to move. So they're looking at, okay, do we stay with Red Hat, do we move off. And they chose to stay with Red Hat and move to RHEL 9. Help us understand really where they saw the value of that and the value of sticking it out with Red Hat and really moving into RHEL.

Chris Wright

>> Well, I might be a little biased-

>> Yes, that's okay.

Chris Wright

>> Since I got a lot of the engineering organization with me, but a big part of that, the way they framed it was strategic partnership. And if you peel that back, the next piece under that was connection to the engineering experts that not only can support something like a complex tool like an operating system, but are actually the authors of a lot of the content in the operating system. And so they're looking at where do we get that level of expertise so we don't have to build it in-house. It's like a perfect partnership, a perfect relationship, and their view was as important as the operating system is, and they're operating at massive scale and doing a mind-numbing number of updates and managing this huge fleet. Actually, what they want to focus on is the higher-level value that they're delivering to their customers so that when they can partner with us, they can just rest easy at night knowing that we're right there and focus on the things that are most important to them.

Paul Gillin

>> What kind of resources is Red Hat throwing at this project? I mean, you've got 20,000 VMs, they're moving over.

Chris Wright

>> Well, for us it's a pretty natural relationship. We have a very mature platform. So it doesn't take a huge number of resources. We have a lot of automation and tools that helps customers move content or move workloads from one place to another. You can do in-place upgrades for things like CentOS to RHEL 9. So it is mostly about that continued long-term relationship and looking where are we going together into the future? They're really intentional about their technology adoption and think about cloud-native technologies, and they love this idea about immutable images and image mode for RHEL. It really matches their workflow and their needs. So it's not only about what can we do today to help them move, but it's that whole big picture.

>> So I think that was a great segue into one of my other questions that we kind of chatted about, which is where does image mode fit in the stack for customers who are out there trying to understand, okay, do I go bare metal, do I go something else in between that and OpenShift. Where does image mode fit and what are the use cases that organizations will thrive on with image mode?

Chris Wright

>> If you think about an operating system and traditional OS, it's RPM-based, you update packages. That model fits anywhere. Image mode fits anywhere. It's just a different workflow and a different model. So think about building content. You want to deliver a bootable image to a server. That content will start with a trusted base, the operating system. You're always going to add something in there. It's an agent. It's something that's unique needs from your enterprise point of view. If you have a truly immutable image, you can't add that agent in there. And if you're using something like an RPM-based workflow somewhere, somebody's going to log in, do some kind of a package update, and you've now got some kind of drift or shift from the primary golden image. So that's useful down to the bare metal, which is really critical. If you're trying to get direct access to accelerators, you're pushing this out to the edge, but that same image can, of course, boot in a VM, and that same image can be the base upon which you build your OpenShift cluster. So I like to say it's turtles all the way down it. It's a useful tool, independent, the use case. Now it's a shift in mindset. So you also have to think about the best application is thinking about how you're cloud-native, how you want to get to the most simplified and rapid and consistent methodology for deploying content.

Paul Gillin

>> One thing that's been themed throughout the show is AI everywhere, and you're driving AI throughout all aspects of RHEL and OpenShift, and Ansible, now where do you think the customers are going to see the greatest productivity gains from these innovations?

Chris Wright

>> AI is touching everything. Now. The way we look at it is it's a critical workload that we want to support on top of our platforms, hence the need to build support for the accelerators. And at the bottom of the stack and something like image mode, you can build this accelerator-adapted image, so it works well on NVIDIA GPUs or a different image, which will work well on Intel accelerators or AMD accelerators with all the application content going up the stack. So there's that workload. It's a workload point of view. Then there's infuse it into our portfolio for helping our customers be more productive with our tools. And then that productivity side, are you helping developers write code? Are you helping content authors create other content? Are you bringing debugging skills to do a better job of root-cause analysis? I mean, the use cases are innumerable, and what we're going through right now is an industry is sorting where are the best starting points so you show early wins and then kind of keep growing from there.

>> Yeah, I think what was really interesting in all of the keynotes, and Ashesh had his and where he talked about some of the AI being used, and we had some really neat discussions yesterday as well around conveyor and app modernization, and you kind of touched on that as well. And I guess we're told you kind of have a theory about how that's going to evolve from an app perspective and a model perspective going forward. Why don't you kind of delve into that a little bit about what ... Is it going to be a hundred percent app modernization or is it, hey, some, I guess you could say we have the concept of building what we call data apps. And these data apps will include models inside them that will be part of larger applications and it's kind of a tiering of these as people go to these cloud native applications? How do you see that and how are organizations thinking about it when you're talking to them?

Chris Wright

>> Well, first of all, early stages in the technology. So everybody's at a different point. Some people are just trying to figure out which way is up. Some people are pretty deep into some deployments and really learning. The way I see it is everything we've done to date for enterprises is supporting the development and delivery of applications. And enterprises are technology companies, they're software companies. They're differentiating themselves as software. Rolling into the future, AI models are a critical part of an enterprise, and I'd say it's a 50/50 split of applications and models that power the enterprise. And when you look at modernizing applications, which generally means following cloud-native principles landing in a container, you may be doing app modernization and discover that you have a critical dependency. It could be a database or some portion of your application that doesn't naturally easily modernize. You can leave it in a VM and bring that over to a modern platform and run your containerized workloads, your virtualized workloads, your AI workloads that need access to accelerators. The application will have integration to the model through just simple APIs. So developers are building intelligence into their applications, and that's the future of the enterprise. So we're trying to make sure that can all fit on the same platform so you have the operational experiences and target the same platform for development.

Paul Gillin

>> That's a pretty big shift. You're going from all applications to half applications, half models. How should customers prepare for that world?

Chris Wright

>> I love the question. So part of it, and it's what I was talking about in the keynote, is one way you can prepare is make sure you lay the right foundation. So build the right platform support for all those different workloads. And then the other way is start bringing in experimentation. You can start as small as you need to learn what's valuable for the enterprise in the AI space. And as you discover something that's useful, you got to figure out how to scale it into production. So you start with the experimentation, you figure out how to put that into a repeatable pipeline so you can deliver the model, repeat the model evolution, and make sure that you're maintaining that integration with the application. So it's like start small, find those wins, and then just keep evolving.

>> Yeah, I think we delved into this at length yesterday and what I liked, and actually Chris Morgan's demo yesterday with Mo and the others on stage was fantastic ... I'm like, "You should package that up." Because it explained it sounds in a way that I thought, Hey, here's where you start with Podman and InstructLab. Here's how you then go to RHEL. Here's then how you go and deploy into production on OpenShift AI and how you take it across RHEL AI and all of these different pieces. What I also like about it, and we're kind of open-source wonks, right? We believe open source is the way, and we loved the Granite open-sourcing announcements that were done of the model and of the code assistant. Where do you see it fitting? I mean, I know our perspective, but I want to get your perspective, and I have a funny feeling open is the way, but how do you see the two working together? Because I think there are ... There's the open source alliance or the AI alliance, which is really pushing on the open source, but then there's government regulations and other things coming. How do you see this all playing out with open source and AI?

Chris Wright

>> More open, more better. So we're definitely going to keep pushing on that open direction. And as you think about what that enables, the broader community support and activity you have around a common piece of technology, the more ubiquitous it's accepted across the industry. So think about something like Linux. It's a de facto standard. Kubernetes, it's a de facto standard. And for one company to try to build a technology that would compete with those, it's just not really possible. The same thing's going to play out in this AI model space. And the other thing that's really important is big large models are difficult to create. They're expensive to run, and they give you more data, more parameters, you get better results. You can also fine tune a model, keep it small, efficient to run produce as good or better results because it's trained on a very specific use case. And I think you'll see that in open source and that proliferation of activity.

>> I think, in fact, we actually have ... We back in September, we put out from the research side of things, myself, Dave Vellante, John Furrier got together and kind of started to look at what was a power law distribution of models. And it had nothing to do with revenue or anything like that. It was size of models and addressability to verticals. And so really big high models in the front end here where you had your foundation models, but then stretching out over, Hey, I want a customized model for my customer service and in this industry and getting that exact with my intellectual property, are you seeing customers? I mean, InstructLab, to me, the light went off when I learned about it. I actually got a little preview, so I learned about it a week and a half ago, and I'm like, wow. I'm like, "This can help you." And there's some other people doing some very interesting thing in synthetic data and stuff like that to help. Do you see that as an enabler to have people more simply fine-tune models? Because I'm with you, and I think we're all with you that nobody's going to go compete with Llama 3 and Granite and build their own model. I mean, that just to me is-

Chris Wright

>> It's cost-prohibitive. It's very difficult.

>> Unless you're building a service that's going to compete with ChatGPT, which I don't know why you would do that. I mean, that's good for them, but I look at it as if your service is servicing your customers and your financial services company, do you see that this ... I mean, I know it came out of the paper from IBM Labs and it was really pulled together very quickly. We had Matt on talking about how he got so excited about how quickly it came together yesterday. Do you see this as a key piece in getting people to models everywhere to that 50% that you're talking about?

Chris Wright

>> Yeah, exactly. And that's why I put it at that number. There's not going to be one big model that runs the enterprise. There's not one big application that runs the enterprise. There's all these little pieces that you compose to run the enterprise. It's the same thing with models. And the difficulty has been how do you specializ, how do you fine tune. RAG is one tool, but infusing the data directly into the model, reducing the size of the model, it's a really effective way. And it's been hard to do that fine-tuning. So making it accessible, that's how you get that proliferation. And that's where you'll see thousands of models each embedded in an application providing that application's intelligence.

Paul Gillin

>> I want to ask you about a project you announced last year, Kepler, an open-source project for measuring the power demands of cloud-native applications. Of course, huge power demands for AI/ML. How are you evolving Kepler for those new needs?

Chris Wright

>> I love that question. Sustainability, it's a big topic. It's especially large in places like Europe where there's regulations and requirements and massive change in the energy dynamics in the market. So we have a lot of requests for how can you support visibility into understanding the power consumption of an application. So you can look at the watts consumed at a server level, but that doesn't give you a lot of insight to where those watts are going. So you take an application down view, Kepler gives you that insight. What we've been doing is working over the last year to expand that. So it's not just CPU-bound workloads like application and business logic, but it's also AI models, which themselves are power-hungry, sitting on GPUs. So now you can combine the insights from the application side, the insights from the AI model side to get visibility into your power consumption. And with visibility, you can think of that as performance. You have a performance SLA, you put it through your CI pipeline and make sure you don't regress. You have an SLA for power, you put it through your pipeline, you make sure you don't regress as you're making changes in your models or your applications. So getting that visibility is key to building more sustainable computing as we go forward.

>> I think that to me is a great place to leave it because I think to your point, we could sit here and talk about sustainability. I think it's the elephant that's not quite in the room in the US yet, but it is definitely in Europe. And I think with the AI Act and some of the other things that are over in Europe, and obviously, there's a lot more on the sustainability side that they've done there that hasn't quite gotten here. You have the two in California and Virginia, but other than that, really not there yet. But I think that's a great place for us to leave it until next time when we get you on.

Chris Wright

>> All right.

>> But thank you for coming on board, Chris. It's always great to get your insights, especially fresh off the keynote.

Chris Wright

>> Glad to be here. Thanks for having me.

>> Yes, and thank you for starting your day with day two at Red Hat Summit and AnsibleFest Live from sunny Denver. Stay tuned for more. Coming here from theCUBE, the leader in high-tech analysis and news.