Sharon AI required mission-critical fibre infrastructure to support a major GPU deployment powering advanced AI model training and high-performance compute workflows. Codecom designed and delivered specialised MPO16 fibre assemblies, approved for use by NVIS, with the low-loss performance, density and reliability needed for direct device-to-device training at scale. With tight delivery windows and a confirmed ‘GO LIVE’ date, Codecom supplied, tested and delivered all assemblies on time, ensuring seamless deployment into their AI environment.

Challenges

Sharon AI’s rapidly scaling GPU platform required connectivity solutions beyond standard data-centre cabling:

• Ultra-low-latency fibre paths to support GPU-to-GPU AI training
• High-density MPO16 fibre cabling for extreme cluster layouts
• Custom splitter cables for hardware-specific breakout requirements
• Predictable optical performance for mission-critical workloads
• Compatibility and approval under NVIS requirements
• Tight deployment schedule tied to a fixed “GO LIVE” deadline
• High-reliability infrastructure capable of supporting sustained AI workloads

Off-the-shelf cabling could not meet Sharon AI’s density, performance or installation requirements.

“What set Codecom apart from other vendors we engaged with, is their understanding of the unique requirements of GPU infrastructure. They provided not just products, but solutions tailored to our deployment needs. Their MPO fibre cabling has performed flawlessly in our data centre environment.”Nathaniel Marsh, Data Centre Operations Manager, Sharon AI

Solutions

Codecom delivered a fibre connectivity architecture purpose-built for high-density GPU AI training:

• Custom MPO16 assemblies engineered for ultra-low-latency, device-to-device AI workloads
  
• MPO splitter cables designed to match Sharon AI’s GPU layout and breakout requirements
  
• Fibre optic patchcords and CAT6 copper leads supporting mixed-infrastructure interconnects
  
• NVIS-approved cable systems, ensuring compliance with partner and platform standards
  
• Fast-turnaround manufacturing to meet GO LIVE deadlines for a major GPU deployment
  
• Precision lengths and labelling supporting clean routing and simplified long-term operation
  
• High-density, low-loss optical performance suitable for sustained AI training loads
  
• Technical support throughout deployment, ensuring integration remained on schedule
  

This solution provided the reliability, clarity and performance needed to operate GPU clusters at scale and positioned Sharon AI for future expansion across their AI training environment.