Scaling AI across distributed environments

With workloads spanning across hybrid infrastructure, multi-cloud and the edge, fragmentation is a big challenge for organisations looking to scale AI.

Enterprises are increasingly having to rethink how and where workloads run, as they navigate growing complexity, data sovereignty requirements and real-world deployment of AI. What are the key challenges?

Lim: Enterprises across Asia Pacific are shifting from centralised IT to distributed, AI-driven environments. However, the biggest challenge isn’t where workloads run, it’s how to manage them consistently.

Today, workloads span on-premises, multiple clouds and the edge, yet many organisations are still operating in silos. This fragmentation is what’s preventing AI from scaling beyond pilots.

Data sovereignty is a major pressure point. Governments across Southeast Asia are tightening rules on where data resides, especially in regulated sectors. This forces organisations to be more deliberate about workload placement, balancing compliance with performance and cost.

AI is also changing workload characteristics. While training may happen in the cloud, inference increasingly needs to run at the edge, closer to where data is generated. This creates new demands around latency, orchestration and infrastructure flexibility. We are seeing this play out clearly in the telecom sector. At MWC 2026, Red Hat highlighted how operators are redesigning infrastructure so AI becomes an operational layer across core,

We are seeing this play out clearly in the telecom sector. At MWC 2026, Red Hat highlighted how operators are redesigning infrastructure so AI becomes an operational layer across core, edge, and cloud. In particular, StarHub’s move toward a polycloud architecture reflects this shift, where workloads are dynamically placed based on real-time needs.

The challenge for enterprises is no longer just complexity, but building a consistent operating model that allows them to run and scale AI anywhere.

What are some key considerations now that organisations in Asia Pacific are rethinking their infrastructure for AI workloads?

Lim: A key consideration is architectural flexibility. It’s no longer just about cost – AI workloads need to run across on-premises systems, public clouds and the edge to balance performance, cost and regulatory needs to deliver the best outcomes.

Approaches like polycloud are gaining traction. StarHub’s model, for example, uses real-time intelligence to determine where workloads should run, whether at the edge for low latency, in the cloud for scale, or on-premises for compliance.

Enterprises are also moving away from one-size-fits-all models toward smaller, specialised systems designed for specific use cases. These models are more efficient and practical for production environments, but they also require infrastructure that can support a mix of models, compute types and deployment patterns without adding complexity.

Governance and trust are also key considerations. As AI moves into production, organizations need systems that are auditable, secure, and aligned with local regulations. This is particularly important in APAC, where data sovereignty requirements are evolving quickly and where organisations are placing greater emphasis on control and transparency.

How are modernised platforms helping to simplify infrastructure management, reduce operational overhead and accelerate service delivery?

Lim: Modernized platforms are shifting the problem from managing infrastructure to standardizing how AI is delivered.

A key issue today is fragmentation. Many organizations still rely on separate tools for models, data, and infrastructure. Newer platforms unify these layers, allowing AI to be deployed and governed as a single system rather than across disconnected workflows.

In Japan, SoftBank is integrating distributed AI inference directly into its network operations, dynamically allocating compute resources across environments. This reflects a broader shift toward AI systems that operate in real time rather than as isolated workloads.

The second shift is scale. As AI moves into production, infrastructure needs to support continuous, high-volume inference and increasingly autonomous systems. Platforms like Red Hat’s AI Factory with NVIDIA are designed to provide a stable foundation for these workloads, helping organizations deploy and run AI applications reliably at scale.

Ultimately, modern platforms simplify operations by creating that consistency. They are enabling AI to run as a distributed, real-time system, one that can scale, adapt, and remain controlled across environments.