Jonathan Bryce, Executive Director of CNCF, discusses cloud-native AI infrastructure, GPU optimisation, digital sovereignty, open source, OpenStack, and why enterprise AI will increasingly rely on proprietary, domain-specific models.
As artificial intelligence rapidly reshapes enterprise technology, the focus is shifting from building larger models to engineering the infrastructure that can run them efficiently, securely and at scale. At the same time, organisations are navigating a new wave of challenges around GPU optimisation, digital sovereignty, private AI, open-source governance and the growing demand for domain-specific AI models. Against this backdrop, the cloud-native ecosystem is evolving beyond container orchestration into the foundational layer powering the next generation of AI infrastructure.
In this exlcusive interview with Jonathan Bryce, Executive Director of the Cloud Native Computing Foundation (CNCF), he discusses how cloud-native projects are addressing the infrastructure demands of AI inference, why open collaboration and standardisation are accelerating enterprise AI adoption, the strategic importance of digital sovereignty, the resurgence of OpenStack in private AI environments, and why purpose-built AI models built on proprietary enterprise data are likely to define the next phase of enterprise AI.
CIO&Leader: AI inference is rapidly changing the economics of infrastructure. We are moving beyond software bottlenecks into hardware constraints around power, GPUs and infrastructure capacity. How are CNCF projects such as HAMI and llm-d helping enterprises fundamentally rearchitect infrastructure to handle these AI workloads without overwhelming existing enterprise environments?
Jonathan Bryce: When you look at projections like 93 gigawatts of power consumption, it’s easy to focus only on the amount of new infrastructure that will need to be built. Certainly, there are hardware constraints today, whether it’s available power generation or simply acquiring enough GPU accelerators. But that’s precisely why cloud native technologies have become so important.
As we demonstrated llm-d, showing that with the exact same four GPUs, intelligent workload deployment and traffic routing can deliver significantly higher utilisation and substantially more inference capacity. That’s the real opportunity.
Cloud-native projects are becoming the software layer that extracts far more AI capacity from the same hardware. The future isn’t just about building more data centres—it’s about making every GPU significantly more efficient.
As AI infrastructure scales, the challenge isn’t simply building more hardware; it’s extracting more performance from the hardware organisations already own. That’s where projects like llm-d, HAMI and several other CNCF initiatives become incredibly valuable. Much of the innovation we’re seeing isn’t happening at the hardware layer but at the orchestration and software layer.
Cloud native technologies were originally developed by companies like Google that had already solved problems around operating distributed systems at enormous scale. Kubernetes, for example, originated at Google. OpenTelemetry also emerged from large-scale production environments before joining CNCF.
We’re now reaching a similar turning point for AI.
Initially, enterprises simply wanted to deploy models on the largest NVIDIA systems they could find. Today the challenge has evolved into orchestrating AI workloads efficiently across fleets of GPUs, intelligently routing traffic, balancing workloads and maximising infrastructure utilisation. That’s where cloud native technologies become foundational to enterprise AI.
CIO&Leader: CNCF recently reported nearly a 70 percent increase in certified platforms under the Kubernetes AI Conformance programme within just a few months. What does this rapid vendor alignment tell you about enterprise urgency around AI infrastructure?
Jonathan Bryce: I think the growth reflects two very important trends. The first is there is enormous market demand. Vendors don’t typically invest in certification programmes unless customers are asking for them, so the increase clearly reflects accelerating enterprise adoption of AI infrastructure.
The second trend is even more significant. It demonstrates that vendors are already willing to collaborate around the foundational layers of AI infrastructure.
Companies collaborate when they believe they can collectively create a much larger market. They don’t collaborate when they believe they’re competing for a fixed opportunity. To me, that level of collaboration suggests we’re still in the very early stages of AI infrastructure maturity.
Companies recognise that establishing common standards today will ultimately accelerate innovation tomorrow. We’ve seen this before. The Internet itself became successful because organisations collaborated around standards such as HTTP, TCP/IP and JavaScript. Those common building blocks unlocked innovation across the entire ecosystem.
I believe AI infrastructure is entering a very similar phase today. Standardisation is becoming a mechanism for expanding the overall market rather than limiting competition.
CIO&Leader: The Open Infrastructure Foundation has now joined the Linux Foundation ecosystem alongside CNCF. How does this broader Linux Foundation ecosystem help protect open source developers while providing stronger governance and legal support?
Jonathan Bryce: Each foundation continues to have its own governance, board, budget and technical direction. The Linux Foundation essentially acts as a network that also includes organisations such as the PyTorch Foundation and the Agentic AI Foundation.
What this structure provides is access to a much broader set of shared services. The Linux Foundation offers strong legal support, global policy engagement and expertise around increasingly important issues such as data sovereignty, digital regulation and intellectual property management.
For developers, one of the biggest strengths of open source remains the licensing model itself. Most projects across these foundations use the Apache Software License, which is globally recognised and well understood. That licensing structure hasn’t changed.
Projects continue to operate under the same governance model and intellectual property framework they had before, while benefiting from the additional support that the Linux Foundation ecosystem provides.
CIO&Leader: With OpenStack and CNCF now operating within the same Linux Foundation ecosystem, how does this change enterprise conversations around building sovereign private AI infrastructure outside public hyperscalers?
Jonathan Bryce: OpenStack has experienced remarkable momentum over the past couple of years, driven primarily by three factors.
The first is digital sovereignty. Organisations increasingly want greater control over where their workloads run, where their data resides and who has access to it. We first saw significant momentum around this in Europe during 2022, and that trend continues globally today.
The second driver is AI itself. Unlike traditional cloud computing, AI introduces highly differentiated infrastructure requirements. Organisations don’t simply need identical virtual machines anymore. They require specialised GPU clusters, TPUs and heterogeneous hardware environments that can be managed efficiently.
Interestingly, many of today’s AI-native cloud providers—the so-called neo clouds—are themselves built on OpenStack.
The third factor has been licensing changes across parts of the infrastructure industry. When organisations experience licensing uncertainty or commercial changes, such as those introduced around VMware licensing, they begin looking for infrastructure that provides greater long-term control and flexibility.
OpenStack gives organisations complete ownership of their infrastructure, while Kubernetes provides workload portability across OpenStack, Google Cloud, AWS or virtually any other environment.
That’s always been one of the greatest strengths of open source. It gives organisations choice—not just today, but the ability to change direction later if business requirements evolve.
CIO&Leader: Digital sovereignty has become one of the defining themes of enterprise technology. How do open source licensing frameworks provide enterprises with a stronger long-term foundation than traditional commercial contracts?
Jonathan Bryce: One of the most powerful aspects of open source licences such as Apache is that they’re perpetual. Once software has been released under that licence and an organisation begins using it, those rights cannot simply be taken away.
The people who created these licences deliberately designed them to provide rights that remain permanent. From a digital sovereignty perspective, that’s incredibly valuable.
If organisations truly want long-term control over their infrastructure, software and digital assets, having rights that cannot later be revoked becomes an essential part of that strategy.
More broadly, I view sovereignty through the lens of choice. Some people argue that sovereignty requires deploying infrastructure in one very specific way. I don’t necessarily agree. To me, sovereignty means having credible alternatives.
When viable alternatives exist, they naturally create competitive pressure across the market. That encourages vendors to improve interoperability, support hybrid deployment models, simplify data portability and reduce customer lock-in.
CIO&Leader: Across industries, enterprises are increasingly investing in purpose-built AI models trained on proprietary data rather than relying exclusively on general-purpose foundation models. Why do you believe specialised enterprise AI models will become increasingly important?
Jonathan Bryce: I think we’re still incredibly early in the AI journey. It feels like AI is moving very quickly, but in many respects we’ve only just begun.
The ChatGPT moment captured people’s imagination because, for the first time, conversations with AI felt remarkably natural.Since then, models have become increasingly capable. However, most of today’s large language models are still trained primarily on publicly available information.
Most of the world’s valuable data isn’t public. It’s inside enterprises. It’s contained within proprietary datasets, operational knowledge and domain-specific expertise. That’s where I believe the next wave of AI innovation will emerge.
Every organisation possesses unique information that no one else has. The opportunity is to transform that proprietary knowledge into specialised intelligence.
Instead of searching through every publicly available document to answer an insurance claim or manufacturing question, organisations will develop models optimised specifically for their own business.
Those models can respond faster, operate more efficiently and, most importantly, leverage knowledge that competitors simply don’t possess. That’s where purpose-built enterprise AI becomes strategically valuable.
CIO&Leader: You have witnessed both the rise of cloud computing and today’s AI revolution. Looking at the developer energy across the CNCF ecosystem, does today’s AI infrastructure boom feel like a continuation of cloud computing, or are we entering an entirely different technological era?
Jonathan Bryce: I actually think it’s both. AI is only possible because of the digital infrastructure built during the cloud era.
Cloud computing created massive global data centre capacity, high-performance networking and distributed computing platforms that now power modern AI systems. Without those investments over the past two decades, today’s AI breakthroughs simply wouldn’t exist.
At the same time, I believe AI will ultimately transform technology in ways that go far beyond cloud computing. It reminds me a great deal of the early evolution of the Internet. Initially, the Internet consisted largely of email and static web pages.
Then came dynamic websites where people could shop online, manage accounts and access services. Later, smartphones fundamentally changed how we communicate and live every day. None of those later innovations would have been possible without the earlier phases.
We’re still in the earliest phase of AI. Just as the internet evolved from static web pages to smartphones, today’s AI infrastructure is laying the foundation for entirely new ways of working and living.
I think AI is currently in that foundational stage. We’re still building the basic infrastructure, standards and platforms. Over the coming years, we’ll likely see entirely new ways of interacting with technology, communicating and working that extend well beyond our current understanding of cloud computing.
That’s what makes this moment so exciting. AI isn’t simply another infrastructure upgrade. It’s laying the foundation for the next major evolution of computing.