Arash Ghazanfari, CxO Advisor, UK & Europe at Dell Technologies, explains why enterprise AI success will depend less on experimentation and more on building the data centre foundations capable of supporting always-on workloads.
Organisations across every sector are investing in AI to drive insight, automation and efficiency. Yet while ambition is accelerating, many remain stuck between experimentation and execution.
The challenge is no longer proving what AI can do. It is making it run continuously, reliably and at scale. That shift is exposing a fundamental issue: many infrastructure environments were never designed for this new class of workload.
Traditional IT environments, built to support transactional applications, are struggling with the demands of modern AI. They were not designed for constant, data-intensive processing or for agentic systems operating autonomously and at scale. At the same time, organisations cannot pause operations to rebuild from the ground up. The real challenge is evolving infrastructure in place, bridging core systems with emerging AI workloads without disruption.
Data as the foundation… and the bottleneck
Data sits at the centre of every successful AI initiative. But it is not just about volume. Data must be accessible, governed and ready to use. In practice, many organisations are constrained by fragmented estates, slow analytics pipelines and manual processes that delay progress.
As AI moves from pilots to production, these inefficiencies become more significant. The challenge is no longer simply accessing data, but preparing and serving it quickly enough to support real-time, always-on workloads. For agentic AI, where systems make multi-step decisions autonomously, the quality and availability of data directly shape outcomes and trust.
This is changing the role of the data centre. It is no longer just a repository for storage and compute. It is becoming the operational backbone that determines whether AI can deliver meaningful outcomes.
From deskside innovation to data centre scale
At the same time, the way AI is developed and deployed is evolving. It is no longer confined to centralised environments. Innovation increasingly starts at the desk, where teams build and test models locally before scaling them into production.
This ‘deskside to data centre’ model is becoming an important consideration for enterprise AI. It allows organisations to experiment quickly, retain control over sensitive data and reduce latency. It can also introduce more predictable economics.
Cloud-based AI can deliver flexibility, but costs often scale unpredictably with usage. Running AI locally – particularly for inference and iterative workloads – can offer a more controlled cost model, alongside stronger data sovereignty and consistent performance.
To enable this, infrastructure must function as a continuum, allowing workloads to move seamlessly from local environments into data centre platforms without re-architecting.
Preventing data starvation in AI systems
As workloads scale, performance depends on the balance between compute, storage and networking. AI systems are only as effective as the data flow that feeds them.
High-performance compute, including GPU acceleration, is essential. But without fast data pipelines and low-latency networks, even the most advanced processors cannot operate efficiently. Data must move continuously between storage, compute and applications to sustain always-on AI.
This raises the bar for infrastructure design. It is no longer enough to optimise individual components. Compute, storage and networking must operate as a coordinated system, delivering consistent performance under sustained demand.
Complexity as the hidden barrier
Beyond performance, operational complexity remains one of the biggest obstacles to scaling AI. Many organisations are still working across fragmented environments, multiple tools and complex integration layers that slow deployment and increase risk.
As a result, time is spent connecting systems rather than delivering outcomes. The priority is shifting away from building bespoke architectures and towards simplifying how infrastructure is deployed and managed.
More standardised, validated approaches are becoming increasingly important. By reducing integration effort and providing more consistent management, they can shorten deployment timelines and allow teams to focus on innovation rather than operational complexity.
Building a foundation for continuous AI
To fully realise AI’s potential, organisations must move beyond siloed approaches and adopt more integrated infrastructure strategies. Environments must support both traditional applications and AI workloads side by side, scale dynamically and adapt as requirements evolve.
This also means considering more flexible, disaggregated models, where compute, storage and networking scale independently while remaining part of a unified system. Combined with automation and consistent control, this creates a foundation that is efficient today and adaptable for the future.
Ultimately, AI success will depend not just on algorithms or models, but on the infrastructure that supports them. Organisations that invest in simplified, integrated and scalable data centre foundations will be better placed to move from isolated experimentation to sustained, enterprise-wide deployment.
