Validação conjunta em campo da inferência de IA distribuída em centros de dados distribuídos baseados em IOWN

The Evolution Toward Network-Centric AI Infrastructure

The rapid expansion of AI-driven applications, such as smart city services, industrial automation, and real-time image analytics, increasingly demands ultra-low latency processing, scalable GPU utilization, and high-bandwidth connectivity. Conventional centralized data center architectures increasingly face limitations in addressing these emerging requirements. To overcome these challenges, a collaborative field validation study was initiated by Accton, Edgecore, NCHC, NIAR, and NTT. This study successfully demonstrated the viability of distributed AI inference over an IOWN-based All-Photonics Network (APN) connecting data centers across Taiwan and Japan.

A Cross-Border Collaborative Ecosystem This multi-domain project was uniquely executed as a practical, field-level deployment rather than a controlled laboratory simulation. Each partner brought specialized expertise to the ecosystem:

• NTT provided the IOWN architecture, APN connectivity technologies, and orchestration control.

• Accton contributed the AI infrastructure architecture and disaggregated hardware platforms.

• Edgecore delivered system integration, optical networking, and operational management platforms.

• NCHC and NIAR supplied the distributed GPU compute environments, validation facilities, and AI workload execution platforms. Working together, they successfully validated a cross-border architecture involving networking, optical transport, AI infrastructure, and operational integration.

Validation Framework and Results The partnership utilized a Smart Traffic Monitoring demonstration as a practical validation framework. The objective was to evaluate whether AI services could be dynamically executed across geographically distributed compute resources. The field validation confirmed that distributed AI inference can be performed across geographically separated sites with stable latency, significantly aided by APN-based connectivity. Furthermore, the project proved that distributed GPU resources could be coordinated independently of physical location. By integrating Data Center Interconnect (DCI) orchestration and optical wavelength switching technologies, the infrastructure supported highly flexible traffic engineering.

The Open Fabric Rack Solution and Future Outlook Building on these positive results, the organizations evaluated a deployable model known as the Open Fabric Rack Solution. This modular, disaggregated architecture integrates application servers, APN connectivity, DCI gateways, and open networking platforms to support next-generation network-centric AI services. The fundamental architectural shift proposed by this whitepaper is the transformation from device-centric computing to a network-centric distributed AI infrastructure.

In conclusion, this joint field validation demonstrates that multi-party ecosystem collaboration is vital for accelerating the evolution of globally deployable AI infrastructure. By leveraging IOWN APN and disaggregated hardware, the technology industry can transition from siloed computing to a highly efficient, future-ready AI framework.