KubeCon + CloudNativeCon EU 2026 | Jonathan Bryce, The Linux Foundation

Clips
More from KubeCon + CloudNativeCon EU 2026

Jonathan Bryce

Executive Director, Cloud & Infrastructure

The Linux Foundation

play_circle_outline Kubernetes Powers AI at Scale: CNCF Growth, KubeCon Buzz, Conformance Surge, Shared Specs, Bursty Hardware Demands

play_circle_outline Kubernetes for AI Workloads: Abstracting GPUs, Storage, and Network to Balance Fault-Tolerant Training and Real-Time Inference

play_circle_outline Scaling and Securing Production AI with llm-d, vLLM, CNCF Projects, OpenSSF, OpenTelemetry, PyTorch, Kubeflow

play_circle_outline Inference now dominates compute; McKinsey predicts huge data-center inference demand by 2030.

Info
Transcript

Jonathan Bryce, The Linux Foundation

Jonathan Bryce

Executive Director, Cloud & Infrastructure The Linux Foundation

In this interview from KubeCon + CloudNativeCon EU in Amsterdam, Jonathan Bryce, executive director of cloud and infrastructure of the Linux Foundation, joins theCUBE's Rebecca Knight and Rob Strechay to discuss how the cloud-native ecosystem is positioning itself as the operational backbone of the AI inference era. Bryce points to the CNCF's Kubernetes AI Conformance Program, which saw a 70% surge in certified platforms in just four months — growing from 18 to 31 — as a signal of accelerating industry alignment. He explains why AI workloads demand a fundamen... Read more

explore Keep Exploring

What is driving the momentum behind CNCF's growth—specifically the 70% surge in certified platforms in the Kubernetes AI Conformance Program? add

How do AI workloads differ in their infrastructure requirements—particularly training versus inference—and how do Kubernetes and other cloud-native projects address those needs? add

How are you helping organizations move beyond experimentation and proof‑of‑concepts to realize real ROI (i.e., get AI/ML workloads into production)? add

How is AI compute demand shifting between training and inference, and what are the implications for future data center capacity and computing architectures? add

bolt Powered by CUBE AI

Jonathan Bryce, The Linux Foundation

search

Rebecca Knight

>> Good afternoon everyone and welcome back to Amsterdam where we are theCUBE and we are doing live coverage of KubeCon CloudNativeCon EU here in Amsterdam. As I said, I'm your host, Rebecca Knight, alongside of Rob Strechay, the principal analyst here. Rob, let's talk about CNCF. We're here at KubeCon.

Rob Strechay

>> I'm excited because this has really been just a wonderful time this week here. And I think again, the community has been vibrant, to put it mildly, and I've been loving it.

Rebecca Knight

>> And it's still growing.

Rob Strechay

>> It is.

Rebecca Knight

>> So we're here to talk more about that with Jonathan Bryce, Executive Director of Cloud and Infrastructure at the Linux Foundation.

Rob Strechay

>> Hi.

Rebecca Knight

>> Thank you so much for coming on the show yet again. You're a CUBE veteran at this point.

Jonathan Bryce

>> Yeah, thanks for having me again.

Rebecca Knight

>> Yeah. So as Rob was saying, I mean, there's been such a growth, explosive growth in this community. I have a stat here. The CNCF's Kubernetes AI Conformance Program saw 70% surge in certified platforms. That's nearly doubling from 18 to 31. That does not happen by accidents.

Jonathan Bryce

>> Right.

Rebecca Knight

>> Can you talk a little bit about what's driving that momentum?

Jonathan Bryce

>> Well, I think it's something where CNCF projects are really meeting the moment for AI. AI is going to be something that is going to drive the next 10, 20 years of technology the way that cloud did the last 10 or 20 years. But it's also a very different workload than what we've ever run before. It requires specialized hardware. The usage patterns are super unpredictable. When we talk about bursty in the cloud, it's like, oh, somebody got five times as much traffic. When you talk about bursty and AI, you can need a thousand times as much capacity and then it goes away. So it's like the extremes are there. And so I think that cloud native technologies have always been about how do we build the most capable operational systems in a way where we can evolve them and improve them. And then AI came along and it's like, this is perfect. So it's really exciting to see this happening. AI conformance was something that we launched just four months ago. And we had a lot of the big players like Google and AWS and Red Hat and others that were involved in helping to put that spec together and get the first version out. But in open source, we like to iterate quickly and move fast. And so we have been moving the spec along. We've been building new testing tools. And as you do that, you also bring new people in who want to collaborate on it. And that's what's driving that increase in additional platforms that are conformant now.

Rob Strechay

>> It would seem like, again, AI is obviously having its moment, but I think Kubernetes is as well with that. And I like to say the cloud native portion of it, not just the underlying Kubernetes itself. Why do you see Kubernetes and the cloud native projects as the right place to be a sandbox for AI?

Jonathan Bryce

>> I mentioned kind of the unique nature of AI workloads. And what Kubernetes was built to do is to create this API and this abstraction between underlying infrastructure systems. And initially it was underlying infrastructure systems that were just clouds, which clouds were homogenous. It was a bunch of very standard server skews that you could change the size of the memory, the size of the disk, how many processors or whatever. But that abstraction layer has grown to support GPUs, to support diverse storage systems, to support a wide array of networking capabilities. And that's what AI workloads need is they need different things. If you're doing a training run, that has a certain usage pattern and a certain demand on the underlying infrastructure. If you're doing inference, it's another different demand. So we need these flexible components and that's what all of the cloud native projects fit into. I think the other thing that goes along with that, there are a lot of people who have run AI training for large models on top of Kubernetes and cloud native technologies, but those workloads are, they're more bursty and periodic. If you lose a node, like maybe you lose some of the data set, but you can start it again. Because these models are all probabilistic, it's actually just math anyway. So it's like if you get enough math in there, then you're going to get the same result. When we talk about inference, that's where you are running a model and you're sending requests to the model and you're asking it questions and you're asking it to make predictions and those have to happen in real time. And so that's the kind of workload too that when we look at beyond Kubernetes, we look at open telemetry, OTel, that's about observability. How can we get the most detail out of our system so we can understand and resolve failures really quickly? You look at things like platform engineering tooling where you want to enable people to be able to move quickly, but safely by having standard platforms. These are the things that make production environments possible. And as we move from the training phase to the inference phase, I think it's even a better fit for what this community has built.

Rebecca Knight

>> But how are you helping organizations get out of this experimentation phase and realize some real ROI? Because that's what they've said, last year was about the experimentation and the POCs and this year, okay, now show me the money.

Jonathan Bryce

>> Well, I think there's a really, really critical thing that's been happening in the CNCF. It's actually three things. If you look back to what companies were doing during the experimentation phase, most companies were doing the same thing, but in a different way. And so as an industry, we weren't kind of moving together on this path. You were going on this path, I was going on this path, you were going on this path, and we were trying to get to the same spot, but we were all having to cut our own trail, so to speak. The three things that we've been doing in the cloud native community is we've been evolving our existing projects. So if we look at something like Kubernetes, we've added DRA with super strong support for accelerators and differentiated hardware. We've added the inference gateway, which is a sophisticated way to do networking for inference type workloads. And we've been making the existing projects a better fit for these types of workloads. The second thing that we've been doing is adding net new projects to the ecosystem. This week we announced llm-d, which is a way to scale inference horizontally. A lot of times, inference runs inside a single GPU, so you're bound to how much memory, how many cores, what's the kind of back plane that I have inside of a single GPU. Now these GPUs are massive, but we all know that if we want to do true global scale, every single thing we've ever seen in history, you have to go horizontal at some point. So llm-d allows us to go horizontally scaled for inference workloads. And then the third thing is building connections and integrations with other open source technologies like PyTorch, which is now 80% of the share in the training market, runs very well on top of Kubernetes with projects like Kubeflow. vLLM is the AI inference serving engine that runs inside of llm-d. And so we're bringing the best, the innovation that's coming out of the AI experts work and we're turning it into something that we can operate. So we're kind of bringing it out of the lab and turning it into a factory and we're doing it in the open. So rather than all of us cutting our own trail, we're now going to benefit from doing this together. And that's how you actually can move to rock solid, scalable systems really, really quickly.

Rob Strechay

>> I think that was fantastic, especially with vLLM and in the PyTorch. I know it's like an llm-d. We talked about that a little bit earlier. I think there's a lot of stuff happening in the open SSF group as well where they're looking at the security of this. One of the things you also talked about this week, which was a big theme was sovereignty. Now, I mean, I've been saying for a while, sovereignty is not just a European thing, it's not just a regulation thing. What are you seeing around sovereignty? Because you had a pretty unique take on it as well.

Jonathan Bryce

>> Yeah. Our community is super global. We have 13,500 people here this week from over 100 countries. Isn't that crazy? And I travel a lot and I spend a lot of time all over the world and what I think we've seen just in the world in general is increased uncertainty. What can I rely on? Can I rely on a certain business partner of mine not to change its license and pricing in a way that's going to dramatically impact my business? Am I going to be able to rely on a geopolitical region as a place that I can feel safe about is my data protected or is it... do I really own it? And these things create uncertainty. And I think that sovereignty is actually less about the lines on a map and more about how do I control my destiny through resilience to changes that I may not be able to foresee or predict? And also how do I make sure that I have options? Options are a key part of open source and what it enables. And I'm a huge open source advocate, but I have an Android phone, but I have an Apple laptop. So now, if Apple started doing things that I really couldn't support anymore, I have other options and I have open source options. And so I think that to me is the beauty of open source. And that's what sovereignty is really about. How do I have options so that I have resilience and when the world gets uncertain or when there are changes that I can't foresee, I will be able to make the best decisions to protect my data, ensure my business continuity, make sure that I have what I need to continue relying on the tech that runs everything.

Rebecca Knight

>> But more on sovereign AI, how does the CNCF view its role about something that, as you say, it's really a geopolitical question and not just a technical one too. So how do you talk with enterprises about these very sensitive things?

Jonathan Bryce

>> Well, there's an interesting kind of legacy that we are super fortunate to be able to build on, which is open source. And I see so much wisdom in some of the early decisions that the open source community made as we run into these types of problems. The open source community was really, even though it's called open source, it wasn't really about, I can see the source code. It was about freedom, freedom to use it, to modify it, to share it, to redistribute it. And that is something that you see as it plays out in the hard reality of policy and legal frameworks. These licenses that our projects use, they are recognized all around the world. They are based on core fundamental IP law that's actually pretty standard around the world. So this is amazing because it's like you can't make a law that suddenly makes the Apache software license not available to a specific company or country. And if you look at most countries, they have policies where they recognize that open source and publicly available software are a special category, similar to science and knowledge that we actually, we want everybody to be able to benefit from this. So there's some fundamental underpinnings that actually are much better than relying on a single two-party agreement that every contract you sign, it's like, this is going to be interpreted by Texas law or New York law or whatever. Open source is really built in this global framework. And I think that's one of those like real pieces of wisdom that the early pioneers of open source left us with as an awesome legacy.

Rob Strechay

>> Yeah, I think totally agree. And I think what it's helped is that people can rely on the underpinnings not to change. And that there's stability to it. It's also transparency. And I think that's what a lot of companies are looking at and needing to have that. And especially, I mean, not to go back to European law, but the CRA and everything else that's going on, I think open source, understanding adaptation of this, yes, this is secure, this is signed and has been in this community for quite some time. How do you see a lot of the primitives right now, like pod reshaping and resizing and things of that nature really going to help as we go through this year where, like you said, inference has got to be... To get ROI, you need inference and you have to go out to the edge and stuff.

Jonathan Bryce

>> Well, there was a McKinsey report that came out and it talked about kind of like the market around inference. And one of the stats they had that I found super interesting was in 2023, if you look at the compute workload that was dedicated to AI, two thirds of it was training and one third of it was for inference. Now here we are three years later, and that's completely flipped, and only a third of the AI compute out there is being used for training. Two thirds is being used for inference. We know we're at the beginning of the inference build out. The other stat they had is that by 2030, which is not that far away, inference will be 93.3 gigawatts of data center capacity and usage, which is more than all workloads combined right now. So the things that we're talking about, when you're talking about these fundamental primitives, we have to remake them because compute is going to be entirely different in four years. It's going to be entirely different than it's ever been before in four years. That's wild. I mean, that's really insane. And I say that inference is the largest workload in human history and now I have data to back me up. I love that.

Rob Strechay

>> So I mean, again, this is exciting. You still have some more time here and more to do. I'm going to be at MCP Dev next week in New York City. There's a whole bunch. You're taking this on the road over to China and I believe Japan and India.

Jonathan Bryce

>> India.

Rob Strechay

>> So what should people be looking forward to between now and Salt Lake?

Jonathan Bryce

>> Well, so in my keynote on Tuesday, I had just a very brief set of comments on specialized intelligence. And if you are in the AI hype cycle, everybody wants to talk about artificial general intelligence or super intelligence, AGI, ASI. I'm actually like more interested in specialized intelligence right now because when I have talked with companies who are creating specialized models, they're seeing that they're way faster, way more accurate, and way cheaper. So what is the problem that's holding AI back from moving forward right now? It's basically power capacity. It's hard infrastructure stuff. It's not just something that's simple to really change. So if you can increase efficiency, then you can really get to that ROI and you can bring forward AI adoption. So specialized models, I think are going to be a big piece of this. And by the way, specialized models doesn't mean that I think everyone is going to start from scratch and train their own voice model or their own vision model or their own LLM. In a lot of cases, what we're seeing is new training techniques like continuous training, reinforcement learning, these kinds of things where you start with a model and then you tweak it and you get to a better state. We had a keynote today, Jago from Google and he talked about the Gemma models and I hadn't heard this before. I don't know how I missed this, but they retrained one of the Gemma models to communicate with dolphins. Isn't that wild?

Rob Strechay

>> Yeah.

Rebecca Knight

>> Yeah. Wow.

Rob Strechay

>> So that's the kind of stuff that I think we're going to see. And it doesn't necessarily make sense to imagine that our sort of frontier models should do all of that. When I'm talking to Claude or ChatGPT or Gemini, I want it to answer my questions really quickly, not have to spend time skipping over the dolphin communication part. So specialized models make a lot of sense. And so if we look over the next six months, as I said, I've talked with some companies who are doing this in their environments. I've talked with some projects that are in this space. So we don't have anything that's baked and that we're definitely going to do, but this feels to me like one of those things where it's tickling my brain and I'm going, "Huh, I got to see if there's something here that we need to be leaning into."

Rebecca Knight

>> And if we could all talk to our pets, I think we'd be a lot happier too.

Rob Strechay

>> Yes.

Jonathan Bryce

>> Oh, wouldn't that be cool? I don't know if we'd want to know what our cats are thinking though.

Rebecca Knight

>> They're so bored of us.

Rob Strechay

>> They're spicy.

Rebecca Knight

>> Jonathan, thank you so much for coming on a really interesting conversation.

Jonathan Bryce

>> Yeah. Thank you.

Rob Strechay

>> Thank you.

Rebecca Knight

>> I'm Rebecca Knight, for Rob Strechay, stay tuned for more of theCUBE's live coverage of KubeCon CloudNativeCon 2026. You're watching theCUBE, the leader in enterprise tech news and analysis.