Organizations across all sectors are developing automation, machine learning, and generative AI proofs of concept at unprecedented speeds. While many perform well in controlled environments, significantly fewer survive the harsh realities of enterprise production, such as strict security policies, complex infrastructure, legacy systems, and budgetary constraints.
During an episode of the Analytics Insight Podcast, host Priya Dialani sat down with Rhys Oxenham, SUSE’s Vice President and General Manager of AI, to discuss the reasons behind the frequent stalling of enterprise AI projects before deployment.
Their conversation addressed the production chasm, open infrastructure, the spectrum of AI sovereignty, the AI factory model, the India AI Mission, and the primary barrier he would eliminate. Below are the key highlights from the interview:
1. What does SUSE do, and what is your role?
SUSE is a worldwide leader in enterprise open-source solutions. For more than 30 years, our focus has been providing resilient, stable platforms tailored for mission-critical operations. Many recognize us through our enterprise Linux offerings and strategic collaborations with entities like SAP. Presently, we also deliver a cloud-native open infrastructure platform leveraging Kubernetes spanning the edge, data centers, and public clouds. Since joining roughly three and a half years ago, my journey has included edge computing and telecommunications before transitioning to lead the AI division earlier this year. My responsibility is evolving our robust open-source foundation into a secure, sovereign base supporting private enterprise AI.
2. What are the biggest challenges when an AI project moves from pilot to production?
Demonstrating a pilot on a single laptop is straightforward. Transitioning to production introduces scale, rigorous security, tooling, sustained operations, and cross-functional teams, a gap we identify as the production chasm. This friction is partly driven by a rapid pace of change; with new frameworks and models emerging nearly every week, organizations fear that current investments will become obsolete within months. Additional hurdles stem from fragmented tools and isolated pilots. Uncontrolled expenses often arise from pay-as-you-go frontier models, while recent shifts in export controls demonstrate that access can be revoked abruptly. Furthermore, shadow AI introduces data vulnerabilities when staff input corporate information into external applications. Regional regulations vary significantly, requiring certain workloads to remain strictly localized within specific jurisdictions.
3. What does open infrastructure mean for enterprise AI, and how does it reduce vendor dependency?
While executive boards demand tangible business outcomes from AI, frontline teams frequently lack authorized tools, driving them toward third-party services. Open infrastructure supplies customers with flexibility, autonomy, and the capacity to pivot—a concept I describe as operational resilience. This entails a transparent, inspectable stack established on open standards across every layer, encompassing application stacks, underlying hardware and silicon, and execution environments. Rather than locking users into a single stack, it empowers businesses to deploy intelligence wherever it serves them best.
4. What does AI sovereignty mean at an operational level?
Sovereignty extends far beyond a simple compliance checkbox or basic data governance. I view it as a continuum stretching from operating systems and silicon to models and runtimes. Because absolute isolation is impractical today, the objective is aligning technological decisions with an organization’s distinct risk profile while maintaining a viable contingency plan. Enterprises should retain the ability to migrate between platforms and safeguard their proprietary intelligence against foreign access limits or shifts in export controls.
5. What is an AI factory, and how does it change the way enterprises build AI?
Rather than physical manufacturing facilities or gigafactories, an AI factory represents the execution of AI workloads at an industrial scale. I characterize it as an assembly line integrating underlying infrastructure with application development. It incorporates zero-trust security, GitOps methodologies, and a continuous pipeline from development to production. Standardized tools eliminate technical debt and accelerate time to value. This framework accommodates MLOps, agentic workloads requiring human oversight and policy governance, and generative AI applications like retrieval-augmented generation and chatbots. Private enterprise AI transcends private cloud concepts; it guarantees that intelligence operates securely within a customer’s absolute control, preventing third-party removal.
6. How can the India AI Mission change compute access for Indian enterprises?
My recent visit to India highlighted this initiative as a substantial catalyst. Historically, Indian businesses have wrestled with steep expenses and GPU procurement bottlenecks. The mission intends to roll out thousands of GPUs through public-private alliances, democratizing compute availability for startups and enterprises alike. It additionally fosters regional cloud infrastructure and local AI gigafactories, liberating companies from sole reliance on external hosting or hyperscalers. The initiative addresses local talent, open datasets optimized for the Indian demographic, and data quality. Nevertheless, hardware capacity represents only half the challenge; enterprises simultaneously require an open infrastructure platform to own, scale, and launch workloads from day one.
7. If you could remove one barrier to production, what would it be?
I would eliminate the false compromise between velocity and oversight. Driven by pressure to implement available technologies rapidly, teams frequently compromise intellectual property protection and security. Although maintaining control preserves ownership and resilience, it can appear sluggish, ultimately fostering governance failures, runaway expenses, and shadow AI. By standardizing on runtimes engineered specifically for private enterprise AI, teams can operate swiftly within their own secure boundaries while retaining total command over their data and intelligence.
Listen to the full discussion on the Analytics Insight Podcast.



