As enterprises move from AI pilots to production deployments, the conversation is rapidly shifting beyond large language models and GPUs towards the infrastructure required to run AI securely, efficiently and at scale. While AI applications continue to evolve, technology leaders are increasingly recognising that long-term success depends on building platforms that are portable, governable and resilient enough to adapt as models, hardware and deployment environments change.
At the same time, sovereign AI has emerged as a strategic priority across the world. However, according to Rhys Oxenham, VP & General Manager of AI at SUSE, sovereignty is not simply about where AI workloads run. It is about giving organisations the ability to retain control over their technology choices without becoming locked into a single vendor or architecture.
In an exclusive interaction with Express Computer, Oxenham discusses why open source is becoming central to enterprise AI, how organisations should think about sovereign AI beyond data residency, and why customer choice will define the next generation of AI infrastructure.
AI infrastructure is evolving rather than being reinvented
Although artificial intelligence appears to represent a completely new technology wave, Oxenham believes the underlying infrastructure follows architectural principles enterprises have relied on for decades.
“The infrastructure that is used to build artificial intelligence systems is not brand new. It is actually following a pattern that we have seen evolving over the last few decades. It is no surprise that modern artificial intelligence applications are built on top of Linux and containers.”
What differentiates AI infrastructure today is the scale of performance required.
“These aren’t just standard cloud-native applications. They have huge demands on the underlying infrastructure in terms of throughput, processing capabilities and optimisation. These systems are incredibly expensive and difficult to procure, so organisations want to make sure they are squeezing as much as they possibly can out of every implementation.”
Drawing on SUSE’s experience across Linux, Kubernetes, telecommunications, edge computing and financial services, Oxenham says the same technologies that have supported mission-critical enterprise workloads are now providing the foundation for enterprise AI.
Open source balances innovation with enterprise control
Open source has long accelerated enterprise innovation, but Oxenham believes its role is becoming even more important as organisations seek greater flexibility in AI deployments. “Open source is not only the route to incredible amounts of innovation, but it is also a pathway to reduce vendor lock-in and make sure that you have operational resilience and business continuity,” he says.
He also challenges the assumption that open-source software compromises enterprise security. “Just because something is open source does not make it ungovernable. In fact, many of the capabilities that come from open source are inherently built for security and stability. The code is reviewed by thousands of people, making it much more auditable.”
At the same time, he stresses that openness alone does not remove the need for governance. “When you look at agentic workloads that are able to act autonomously on behalf of users, governance becomes really important because data is often part of that pathway,” he adds.
Rather than evaluating AI platforms solely on model performance, Oxenham believes enterprises will increasingly prioritise governance, operational resilience and architectural flexibility.
Sovereign AI is much broader than data residency
For Oxenham, one of the biggest misconceptions around sovereign AI is that it focuses only on where data is stored. “I think one of the most important things to recognise is that sovereignty should really be thought of as a spectrum.”
He explains that sovereignty spans every layer of the AI stack, from silicon and operating systems to AI frameworks, deployment platforms and operational control. “It is very difficult in today’s world to have a truly sovereign AI stack. Silicon comes from one place, software comes from another and organisations rely on different technologies across the stack.”
Instead of pursuing complete technological independence, organisations should align sovereignty with their operational priorities and risk profile. He points out, “An open infrastructure platform gives organisations the ability to pick and choose the capabilities they need at different layers while recognising that complete end-to-end sovereignty is extremely difficult.”
He also praises India’s approach towards sovereign AI. “I’ve always been incredibly impressed by the amount of innovation that comes out of India. What stands out is not just the rate of innovation but India’s quick adoption and leadership when it comes to business resilience.”
According to Oxenham, initiatives such as IndiaAI, alongside investments in domestic talent and technology capabilities, demonstrate that India is building long-term competitiveness rather than simply localising infrastructure.
AI will increasingly operate wherever the data is generated
Having previously led SUSE’s edge computing business, Oxenham avers enterprises will deploy AI across both centralised infrastructure and distributed edge environments rather than choosing one over the other. “I think it really depends on the workloads and the business that particular organisation is in.”
He cites manufacturing as an example where real-time anomaly detection requires AI to operate directly on the shop floor, while financial services increasingly require intelligence across branch networks as well as central infrastructure. “A huge amount of AI is actually operated at the edge because that is where the data is generated and where low latency is required.”
For Oxenham, the future of enterprise AI lies in supporting diverse deployment models that are determined by business requirements, industry regulations and where data is created rather than by a single infrastructure strategy.
Private enterprise AI gives organisations choice without lock-in
Looking ahead, Oxenham says his immediate priority is enabling organisations to deploy AI according to their own operational, regulatory and business requirements rather than forcing them into predefined technology choices. “I’m really focused on enabling customer choice. We want to make sure customers can operate AI on their terms.”
He explains that this philosophy underpins SUSE’s vision of private enterprise AI. ‘Private’, he says, ‘does not necessarily mean private cloud.’ Instead, it means AI that remains private to the organisation, regardless of whether it runs on-premises, at the edge or in the public cloud.
“We’re really focused on what we call private enterprise AI—private not in the sense that it has to run in a private cloud, but private to the organisation. Customers can choose to run it on-premises or in the public cloud, but it remains under their control.”
That flexibility extends across every layer of the stack. “We’re not locking customers into a particular silicon type. We have strong partnerships with NVIDIA, but many organisations use different silicon. The same applies to Linux distributions, AI frameworks and foundation models. We really want to help customers use AI on their terms,” he says.
According to Oxenham, the ultimate objective is to make enterprise AI easier to consume while preserving flexibility. “My priority is enabling AI to become a consumable, repeatable and solution-orientated capability for our customers.”
For Oxenham, the enterprises that successfully scale AI over the coming years will not necessarily be those with access to the largest models but those that build secure, resilient and open infrastructure capable of evolving alongside the technology. In that journey, open source will remain the architectural foundation that gives organisations genuine choice, operational resilience and the flexibility to build sovereign enterprise AI without sacrificing innovation.