KubeCon + CloudNativeCon EU 2026 | Roberto Carratala' & Vincent Caldeira, Red Hat

Clips
More from KubeCon + CloudNativeCon EU 2026

Vincent Caldeira

APAC CTO

Red Hat

Roberto Carratalá

Red Hat

play_circle_outline Sovereign AI and the Hardware Imperative: Data Jurisdiction, Cybersecurity, and National Competitiveness

play_circle_outline Regulatory drivers like DORA accelerated sovereignty focus, especially in financial services.

play_circle_outline Regional APAC differences: China can build hardware supply chains; India prioritizes partnerships and local models.

play_circle_outline Defense sector heavily impacted; uses embedded AI and requires strict guardrails and transparency.

Info
Transcript

Roberto Carratala' & Vincent Caldeira, Red Hat

Vincent Caldeira

APAC CTO Red Hat

Roberto Carratalá

Red Hat

In this interview from KubeCon + CloudNativeCon EU, Roberto Carratalá, principal AI platform architect at Red Hat, joins Vincent Caldeira, APAC chief technology officer of Red Hat, to talk with theCUBE's Rebecca Knight and Rob Strechay about why sovereign AI is reshaping how enterprises move AI workloads from experimentation into production. Caldeira defines sovereignty as the ability to exert control over one's digital destiny, identifying three distinct layers — data, technical operations and economic competitiveness — that organizations must now address. D... Read more

explore Keep Exploring

What is digital (or data) sovereignty, and why does it matter for control over data, organizational dependencies, economic competitiveness (e.g., access to GPUs for AI), and cybersecurity? add

How do you see AI driving (or affecting) sovereignty, and in what ways will sovereignty and AI come together—particularly given DORA's influence on financial‑services sovereignty in Europe? add

How do countries in the Asia-Pacific region define and approach "AI sovereignty," and why do their strategies differ (for example between China and India)? add

What challenges do defense organizations and other enterprises face when adopting generative AI models, and what transparency, risk‑management, and guardrail measures are needed? add

bolt Powered by CUBE AI

Roberto Carratala' & Vincent Caldeira, Red Hat

search

Rebecca Knight

>> Good morning everyone and welcome back to theCUBE's live coverage of KubeCon, CloudNativeCon EU 2026. We are live in Amsterdam. I'm your host, Rebecca Knight and I'm alongside the incredible Rob Strechay. Rob, it's day two.

Rob Strechay

>> It is day two and it's been buzzing for this entire week. I think a lot of the conversations going on, not only here, but in the hallways has been fantastic.

Rebecca Knight

>> Indeed, indeed. And we're going to get into something profound, sovereign AI.

Rob Strechay

>> Yes.

Rebecca Knight

>> It's a fraught topic in many corners of the world right now.

Rob Strechay

>> Well, I think it's one of those things that it's not just a EU thing. It's not just a regulation thing. There's a lot going on.

Rebecca Knight

>> Indeed, indeed. Well, I would like to welcome Roberto Carratalá, principal AI platform architect at Red Hat. Roberto, welcome.

Roberto Carratalá

>> Thank you for having us.

Rebecca Knight

>> And, Vincent Caldeira, APAC CTO at Red Hat. Thank you so much for coming on the show.

Vincent Caldeira

>> Thank you. Happy to be here.

Rebecca Knight

>> Yeah. So as I said, we're talking sovereign AI. You, Vincent, are giving a talk about sovereign AI later this week, which is a term that means different things to different people right now. I'd love you to explain how you talk about the term and what it means to an organization that's trying to move AI from experimentation into real production.

Vincent Caldeira

>> Yeah. So I think the way we actually define sovereignty is the ability to exert control over your digital destiny. And I think in Europe and outside of Europe, the control over the data has always been a very key topic. I think everyone agrees that any organization needs to have established security and controls around it. What's happened over the past couple of years or so is the realization that their dependency, as well as are technical and operational. So there's really three distinct layers that now organizations are looking at, which essentially it's taking us beyond just the regulatory compliance and the security discussions to other dimensions such as economic competitiveness. Countries that don't have enough access to GPU, they can't have AI factories, meaning they cannot build their own model, their own AI capability. So that's becoming a big, big dependency for economic growth. And then there's a dimension of risk management cybersecurity as well. Are you really owner of your data if your data is stored in a third party system that is controlled by extraterritorial laws?

Rob Strechay

>> Yeah. I was going to say that from that, DORA really kicked things off here in Europe and really pushed people to look at sovereignty in a different way. I mean, the financial services industry, especially where it came out of France and then grew into an EU, but it's really gone worldwide in that way. What are you seeing from ... AI seems to be that next level of driving sovereignty. And how do you see sovereignty and AI really coming together from that perspective?

Roberto Carratalá

>> That's a very good question. So we need to understand that we need to have this full transparency from these models. And from these close source models, you don't have this provenance and you don't know exactly how has been trained. And also some of our customers, they want to own their own models that you have this cost productivity and also be able to deploy certain models because of regulation. So we saw with DORA and the EU as well, but not only in Europe, it's also all over the world that you need to have this control around your own data, your own model, and your own infrastructure in order to be able to have this control around the data and also productivity around the whole cost. And that's key. And also, a lot of our customers are asking around us and are implementing it.

Rebecca Knight

>> So, Vincent, you oversee APAC and obviously with this topic, different regulatory landscapes, different cultures, different approaches to economic competitiveness, different rules, different expectations. But coming from APAC, how has that shaped how you think about and approach what enterprises actually need from their AI infrastructure?

Vincent Caldeira

>> I would say first and foremost from APAC because of diversity, the biggest learning from my perspective is there's not one single way of defining sovereignty and the relevant priorities. So even if you look at a continent like Asia Pacific, if you are a country like China, you have the possibility to manufacture your own accelerator, which obviously not every country can do. So the way you are approaching building like AI factories will be extremely different. You can control the entire supply chain from the ground up potentially. You can still work obviously with external company. So I mean, NVIDIA still has a huge presence obviously in China, but then you have an alternative with your homegrown economy, but you can basically go all the way deep into the supply chain itself. Now, if you take a country like India, which is also a huge population, obviously very big consumer of technology, but India doesn't really have an opportunity to be its own supply chain all the way down to the hardware. So the kind of focus in a country like India will be how do we partner with other part of the industry, the tech industry for hardware, for models, how do we still manage to build our own diversity with our own Indian data, building Indian language model to reflect our cultural difference or aspirations as well, and obviously still have a way to create some kind of ownership. So, to me, what is fascinating is if you take two or three different countries, you will not find an exactly similar way of approaching the discussion. So when we talk sovereign, it's about more of an approach and a way to look at kind of risk and reward from a conversation point of view.

Rob Strechay

>> I mean, when you look at sovereign AI, I mean, I think it's not just about sovereign as in a country. It's about, like you were saying, Robert, that was really around how you keep your data, how you keep your intellectual property. How are those conversations ... You're always out, you're both out talking to customers all the time. How do those conversations go? And when people look at it, "Hey, I want model optionality, but I need to know it's trusted. I need to know it's signed." And the supply chain aspect of sovereign AI has to be a big piece of it as well.

Roberto Carratalá

>> Exactly. And for that reason, we also trust these valuated models. And in Red Hat, one thing that also we put together are all of these validated models that we know that, and we also check in different AI accelerators and be able to at least have all of the provenance and be able of having the flexibility of deploying everywhere. And that also from our customer perspective, they want to be able of deploying, no matter what, if they have a hyper secure data center or if they wanted to go to the cloud, having this flexibility of deploying their own model and also have their own control of the data, it's key. And a lot of different customers, especially the highly regulated ones, they need to follow this because of the highly compliance structure.

Rob Strechay

>> So just, have you seen an industry strength or ones that are really leaning into sovereign AI, like verticals or industries that either of you, I mean, again, that-

Vincent Caldeira

>> Yes.

Rob Strechay

>> Because industries tend to have different needs for certain things like manufacturing, highly distributed or retail, highly distributed, but financial services can be more consolidated usually. What are you seeing across the world?

Vincent Caldeira

>> So I guess from a vertical, so interesting is the one that's probably the least talked about, but it's actually the most impacted, if you think about it, is defense.

Rob Strechay

>> Right.

Vincent Caldeira

>> Defense use an enormous amount of artificial intelligence into their operating model, so to speak. Most of it is embedded. They have a huge amount of internal research as well that happens on cutting edge technology. And the problem is we reach this point where even for the different industry and the huge amount of means they have is still not practical for them to train a generative AI model from scratch.

Rob Strechay

>> Right.

Vincent Caldeira

>> Because that actually costs billions of dollars. But at the same time, we are basically running critical decision on those technology. They need a huge amount of kind of risk management and guardrails around how they actually use it. So Roberto was totally right in the sense that because you cannot own the entire supply chain end-to-end now, you need to have this extra degree of transparency on the process for all the components you are taking, right? So generally the discussion of now with enterprises, the fantastic things today for technologists is you can go to Hugging Face and you have 2.5 million model.

Rob Strechay

>> Yeah.

Vincent Caldeira

>> but the bad thing for enterprise is, do you actually trust your team to use any of these 2.5 million model in a risk-based way? And the answer is obviously not. You can't actually trust that. So you need to start thinking about guardrails. If you're using it in a healthcare situation or to provide banking advice, you have to start thinking about what do I have to do to get to a degree of control of my entire system?

Rebecca Knight

>> Well, that's what I wanted to get at. And I think you both are talking about this because compliance and regulatory frameworks are moving so much faster than legal teams and engineering teams can keep up with. And so is there an architectural principle that you keep coming back to in terms of how you make sure that these are flexible, as you say, and being able to adapt, but also compliant and also doing what they need to be doing to stay within the bounds?

Roberto Carratalá

>> There are, I think, the three principles that also Vincent mentioned. The first is be able to be flexible and having this interoperability using and be able to deploy everywhere, your own models, your own data, your own infrastructure as well. The second one is having this trust and transparency in order to be able to know exactly with these software bill of materials, with these SBOMs, that you know where are your provenance and be able to trust the models that you are deploying there, but also connecting to your data in a private way. And the third one, it's having this cost-effective because we are using these AI accelerators. And especially for the gen AI, you need to control the cost. And sometimes when you are with these close source third party models, you don't have the control of the own cost. And we already saw that a lot of our customers transitioned from just the exploratory phase to now putting things in production that consumes a lot of different tokens and they are skyrocketing on this consumption. So you need to be able of having these cost-effective models. And for that reason, things like LLMD that we are very happy that has been contributed to the CNTF as well, delivers these fully optimized, be able to scale across the different Kubernetes systems. So that's I think the three keys that we are seeing in this industry and also in this blueprint or architecture.

Rob Strechay

>> Well, I mean, I'm glad you brought up LLMD. I mean, you also are contributing vLLM from the Neural Magic acquisition that happened and all of that in PyTorch.

Vincent Caldeira

>> Yeah.

Rob Strechay

>> And I think what's interesting is, as you're here on the floor having these conversations around AI, some of the people who've traditionally come to KubeCon, CloudNativeCon aren't the AI developers. What are the questions that they're asking now as platform engineers coming in?

Vincent Caldeira

>> I think the main challenge of platform engineers in enterprise now is they have a new consumer who is a data scientist. Exactly.

Rob Strechay

>> And.

Vincent Caldeira

>> It's not a consumer we are familiar with, right?

Rob Strechay

>> Right.

Vincent Caldeira

>> The data scientist kind of pattern of usage is, "I'm going to procure a GPU for a while. I'm going to run a bunch of experiments." It's very hard to know at what point this experiment is kind of ready for primetime and production. The pattern of development of software are changing because of this, right? And as a platform engineering, it's a total different way of thinking about how do I need to provision infrastructure.

Roberto Carratalá

>> Exactly.

Vincent Caldeira

>> How do I now ... Principles that we have, for example, FinOps that are relatively easy for the traditional microservice application, now everything goes out of the window and you have is way more expensive infrastructure now to manage and operate. So it's really the coalition of two words that don't actually really want to work with each other, but we actually need each other to be successful.

Rob Strechay

>> Right.

Rebecca Knight

>> Were you going to say something?

Roberto Carratalá

>> Yeah. It's providing as a service, like for example, Models as a Service. One AI developer, the thing that they want is starting to build these AI applications. And for that reason, they need to have these concept of, for example, Models as a Service in order to be able of sharing one model across different organization, but in a very controlled manner, because you want to be able of selecting which model that you are allowing, for example, to the AI developers. And also it's trying to reduce the cost because, again, we have these AI accelerators that cost a lot of money and we are putting also all of these concept of Models as a Service. The AI developers are very interested in it.

Rebecca Knight

>> To wrap up, I mean, these are not small problems, sovereign AI, compliance at scale. Where are we in terms of ... What makes you hopeful, I guess, about this topic in terms of how we will solve them and going forward?

Vincent Caldeira

>> So, to me, what makes me hopeful is, and I think it was mentioned yesterday after the announcement of L&D contribution, we are in this rare moment where the community has anticipated a lot of this issue and is actually ahead of proprietary system. That doesn't happen a lot.

Rob Strechay

>> No.

Vincent Caldeira

>> Typically, we tend to catch up after the fact. Today, the reality is, I think most enterprises are still trying to get there into production. So a lot of people tell me, "Oh, but L&D is too early, people don't have a scale problem." It's actually a good thing. We are actually anticipating the problem for once. So that's what makes me, I guess, very hopeful because it barely ever happens. Generally, we tend to try to fix things after we already have the problem in front of us. I think here, we are kind of a year ahead or so.

Rob Strechay

>> Excellent.

Rebecca Knight

>> That's good to hear. A little self-awareness from the industry.

Vincent Caldeira

>> Yeah.

Rebecca Knight

>> Excellent. Well, Vincent, Roberto, a really fascinating conversation. Thank you.

Vincent Caldeira

>> Thank you so much.

Roberto Carratalá

>> Thank you very much.

Rebecca Knight

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