As Singapore continues to invest in AI infrastructure, data center networks are rapidly evolving from 10G, 25G and 100G connectivity to 400G and even 800G architectures. For organizations building or expanding AI data centers, upgrading switches and optical transceivers is only part of the equation. The underlying cabling infrastructure must also evolve to support higher bandwidth, greater density and long-term scalability.

As AI clusters grow in size and complexity, cabling is no longer just a physical connectivity layer. It has become a critical foundation that directly impacts network performance, deployment efficiency and future expansion.

Why AI Workloads Require Modern Cabling Infrastructure

Unlike traditional data centers, where traffic primarily flows between users and servers, AI training environments generate massive east-west traffic between GPU servers. Large-scale AI clusters require constant parameter synchronization and node-to-node communication, placing greater demands on bandwidth, latency and network reliability.

To support these workloads, many AI data centers are adopting 400G networks, while hyperscale deployments are increasingly exploring 800G architectures. As network speeds increase, cabling infrastructure must scale accordingly. Otherwise, link capacity, cable density and operational complexity can become bottlenecks that limit overall AI performance.

As a result, cabling has evolved from a supporting component into a strategic element of AI data center design.

Key Cabling Changes for 400G and 800G Networks

One of the most significant changes is the growing adoption of OS2 single-mode fiber. While copper cabling and OM3/OM4 multimode fiber remain suitable for some applications, many new AI data center deployments are turning to single-mode fiber to support longer transmission distances, higher link budgets and future network upgrades.

As AI is increasingly adopted in customer service, office collaboration, knowledge retrieval, coding assistance and data analysis, enterprise network boundaries are becoming more complex. For businesses in Singapore, the question is no longer whether to adopt AI, but how to improve efficiency while maintaining data security, access control and compliance. For organisations advancing digital transformation, building a controllable, visible and manageable cybersecurity foundation has become essential to support the stable deployment of AI.

As AI Adoption Grows, Enterprise Networks Face New Governance Challenges

As AI is introduced into enterprise environments, network governance becomes significantly more complex. For Singapore enterprises, these changes are mainly reflected in the following areas:More access targets: Employees are no longer accessing only internal systems, but may also use browsers, APIs or third-party platforms to call AI services.More complex data flows: Documents, knowledge bases, logs and business data may all be involved in model calls or inference processes.A wider variety of endpoints: Office PCs, mobile devices, remote terminals, IoT devices and even branch networks can all become entry points for AI services.Greater sensitivity to local compliance requirements: In environments where hybrid working, cross-border collaboration and multi-cloud architectures coexist, the lack of unified identity recognition, network segmentation and policy control can further increase potential risks.

Why AI Governance Must Start with the Underlying Network

When discussing AI governance, many organisations first think about approval workflows, data classification and usage policies. However, these requirements ultimately need to be enforced through network capabilities. The network determines who can connect, where they connect from, which resources they can access, and whether abnormal traffic can be identified and isolated in time.

For Singapore enterprises, effective AI governance generally depends on the following foundational capabilities:Identifying users, devices and applicationsLogically isolating different business domains and data environmentsVisualising access behaviour and traffic pathsCentrally managing policy deployment, device configuration and network changes

In other words, organisations cannot solve the problem simply by adding another security tool. AI applications need to be incorporated into an overall network governance framework.

From Campus Access to Internal Segmentation, What Key Control Points Should Enterprises Focus On?

In practice, Singapore enterprises can prioritise three areas: access control, network segmentation and unified management, supported by the right network platform and operations capabilities to ensure that security policies are properly enforced.

Access Control: Clearly Define Who Can Access AI Resources

AI-related applications are often used across multiple teams, including R&D, sales, customer service and operations. Different roles, endpoints and office locations require different access permissions. Enterprises therefore need a unified authentication and admission mechanism that supports differentiated access policies based on identity, device and location, preventing uncontrolled use of AI services.

In this process, stable access and policy enforcement capabilities are fundamental. FS PicOS® enterprise switches support deployment across access, aggregation and core layers, providing a stable foundation for campus networks and helping enterprises implement access policies more consistently.

Network Segmentation: Limit the Scope of Risk Propagation

Internal knowledge bases, office endpoints, guest networks, and resources used for AI inference, training or data processing should not remain on a completely flat network. Through proper logical segmentation, enterprises can define business boundaries more clearly, reduce lateral risk propagation and improve the precision of access control.

As AI assistants, intelligent customer service, real-time voice interaction, meeting transcription and video collaboration become part of day-to-day business operations, enterprises are placing new demands on their branch and campus networks. In the past, networks primarily supported office access and business traffic. Today, however, they are becoming critical infrastructure for delivering real-time AI experiences.

For a growing number of businesses in Singapore, the stable operation of AI applications depends not only on model performance, but increasingly on whether the underlying network can provide low latency, high reliability and ongoing scalability.

Why Does Real-Time AI Place Higher Demands on Networks?

Compared with traditional office activities such as web browsing and email, real-time AI applications rely far more heavily on continuous and stable data exchange. When AI assistants, voice-based customer service, meeting transcription and real-time translation become part of everyday workflows, any network jitter, packet loss or sudden congestion can directly affect the user experience, resulting in slower responses, choppy audio or interrupted collaboration.

At the same time, AI applications often run alongside video conferencing, SaaS tools, file synchronisation and cloud access. As more AI PCs, mobile devices and meeting room terminals connect to wireless networks, enterprises need not only greater bandwidth, but also a campus network capable of supporting high concurrency and multiple services running simultaneously.

In the era of real-time AI, enterprise networks need to focus on the following capabilities:Delivering low-latency, low-jitter performance for real-time servicesSupporting high concurrency across voice, video, AI and office applicationsEnhancing wireless access capacity to accommodate more devices online at the same timeEnabling rapid deployment and streamlined operations across multiple branches through unified management

What Capabilities Are Needed to Build a Real-Time AI Network?

To meet the growing demands of real-time AI services, enterprises need to upgrade their access, aggregation and management layers as a whole, rather than simply increasing bandwidth. FS provides complete campus network solutions to help organisations build stable, efficient and scalable AI-ready network infrastructure.

Ensure Real-Time AI Performance with Intelligent Traffic Scheduling

Real-time AI services place much higher demands on latency and stability than ordinary office traffic. Enterprises need to use QoS (Quality of Service) to prioritise critical services such as AI assistants, voice customer service and video conferencing, ensuring a smooth experience even during peak usage periods.

FS enterprise switches support flexible QoS policy configuration, allowing traffic to be intelligently scheduled according to application type. By prioritising real-time services and reducing the impact of congestion on critical workloads, they provide a stable network experience for AI-powered office environments and intelligent customer service.

Build a High-Performance Campus Architecture to Increase Overall Capacity

As the number of Wi-Fi 6/7 devices, AI PCs and smart terminals continues to grow, upgrading only the access layer is no longer enough. Enterprises also need to improve switching capacity, high-speed uplinks and core network performance to prevent bottlenecks from shifting from the edge to the aggregation and core layers.

FS provides integrated switching solutions covering the access, aggregation and core layers, combined with 2.5G/10G access, 25G/100G high-speed uplinks and high-performance campus switching architecture. This helps enterprises build high-bandwidth networks ready for future business growth, while reliably supporting AI, video collaboration and cloud applications.