Attendees networking at Secure-X Forum 2.0, Le Méridien Putrajaya

Putrajaya, Malaysia — 24 September 2026, as Malaysia rapidly cements its position as Southeast Asia’s premier digital infrastructure hub, the exponential surge in generative AI workloads and mega data center investments is transforming the computing landscape. Addressing the critical network bottlenecks that govern AI cluster performance, Ruijie Networks Enterprise teamed up with our distributor, KH Datagate, at the Secure-X: AI-Powered Infrastructure & Cyber Defense Forum 2.0.

Attendees networking at Secure-X Forum 2.0, Le Méridien Putrajaya

Organized by KH Datagate in collaboration with the Public Sector ICT Association (PERJASA) at Le Méridien Putrajaya, the forum convened top government ICT leaders, enterprise architects, and technology strategists to discuss the digital foundations required for sovereign AI and next-generation data centers.

The Network Bottleneck: Why "5 GPUs Waiting for 1" Threatens Compute ROI

Delivering the keynote address titled "Building the AI Data Center: Planning, Performance, and Operational Considerations," Muhammad Ariff, Network Solution Architect at Ruijie EBG Malaysia, shifted the conversation away from conventional datasheets toward the operational realities of distributed computing.

While organizations invest heavily in top-tier accelerators, Ariff highlighted that raw GPU count does not automatically guarantee effective computing throughput. Citing industry findings, 16,384-GPU pre-training cluster, he pointed out that the deployment suffered 419 unexpected interruptions across 54 days—averaging a failure every few hours and driving effective cluster utilization down to just 38% to 43%.

"In distributed AI training, computing tasks are strictly synchronized across thousands of nodes. If five GPUs are ready to synchronize while just one is stalled by a congested or failed network path, the entire computing task halts. A minor network latency anomaly or packet drop is no longer just a connectivity issue—it directly degrades overall GPU productivity and burns capital. The network is the true backbone of the AI data center."

Ariff underscored that maximizing AI computing returns requires a full-lifecycle approach: moving from clean physical cabling validation to real-time telemetry capable of identifying silent throughput degradation before entire workloads crash.

Proven at Hyperscale: China’s AI Cluster Pedigree & Broadcom Strategic Alliance

Ruijie and Broadcom partnership and system engineering
Tomahawk-based switch platforms from Ruijie

Ruijie’s perspective is anchored in extensive, production-scale deployments inside some of the world's most demanding hyperscale AI environments. In China, Ruijie has deployed over 350,000 data center switches across top-tier cloud and AI leaders.

Backing this systems engineering is Ruijie’s status as a Broadcom Tier-1 Strategic Partner. Ruijie designs and manufactures its high-density switching platforms around Broadcom’s Tomahawk silicon family—spanning Tomahawk 3 (12.8T), Tomahawk 4 (25.6T), Tomahawk 5 (51.2T), and the groundbreaking Tomahawk 6 (102.4T) with Cognitive Routing 2.0.

By adopting an Open Network Architecture, Ruijie provides a standards-based, lossless Ethernet fabric that connects multi-vendor accelerators and storage platforms without locking enterprises into costly, proprietary interconnects. This architecture was recently validated at regional benchmarks like Thailand's Siam AI Cloud, which achieved 97% fabric bandwidth utilization across 1,024 GPUs at a 25%+ lower capital cost compared to proprietary alternatives.

Flagship Showcase: The RG-S6990-128QC2XS 400GE AIGC Switch

RG-S6990-128QC2XS 400GE data center switch front panel
RG-S6990-128QC2XS fans and power supply units

At the Ruijie and KH Datagate technology showcase, the centerpiece demonstration featured the RG-S6990-128QC2XS, Ruijie’s flagship 128-Port 400GE fixed data center switch purpose-built for AIGC intelligent computing networks:

  • Ultra-High-Density Throughput: Equipped with 128 x 400GE QSFP-DD ports and 2 x 10GE management ports, delivering non-blocking line-rate forwarding in a compact form factor.
  • AI-Fabric & Lossless RoCE: Powered by Ruijie’s AI-Fabric architecture, the platform integrates advanced RoCE adaptive tuning and Ruijie Dynamic Load Balancing (RDLB) to eliminate microburst congestion, optimize packet forwarding paths, and significantly shorten training cycle times.
  • Full-Stack AIDC Operations: Bridged with Ruijie's 4-layer AIDC infrastructure solution, supporting end-to-end computing power scheduling, real-time telemetry, and visual network O&M.

To help local government and enterprise organizations de-risk their upcoming AI infrastructure investments, Ruijie and KH Datagate announced a complimentary 90-minute AI Architecture & Readiness Review. Available to public and private data center operators, the technical audit assesses GPU cluster topologies, power parameters, RoCE congestion profiles, and scaling pathways.

Lucky draw winner at the Ruijie and KH Datagate booth

The event also featured an interactive Kahoot technical challenge and a booth lucky draw, and Ruijie extends warm congratulations to the attendees who walked away with the grand prizes.

With data center developments rapidly escalating across Cyberjaya, Johor, and greater Kuala Lumpur, Ruijie Networks and KH Datagate remain committed to equipping Malaysian enterprises and public sector agencies with battle-tested, high-performance network fabrics that turn raw AI computing capacity into uninterrupted, reliable intelligence.