The rapid expansion of AI infrastructure is forcing a fundamental shift in how data centers manage network congestion and hardware utilization. IP Infusion reports that over 100 service providers and data center operators globally are now utilizing its OcNOS 7.0 software release to address these shifting requirements. This adoption centers on two primary strategic pressures: the urgent demand for lossless networking within AI clusters to prevent GPU underutilization and the industry-wide push to collapse IP and optical layers into a single, unified footprint. By offering an open alternative to vertically integrated, proprietary systems, the company aims to help operators manage the transition toward high-density 800G fabrics and converged transport networks while potentially reducing total cost of ownership by 40–60%.
Scaling GPU Utilization and 800G Fabric Density
As Ethernet gains traction over InfiniBand in AI back-end networks, the efficiency of the underlying network becomes a critical determinant of ROI for expensive GPU clusters. IP Infusion is positioning OcNOS 7.0 as a tool to prevent the network from becoming a "silent thief" of productivity by ensuring training data flows without interruption. The software utilizes Dynamic ECN, a real-time congestion management feature designed to provide zero-stall performance for AI training jobs. This capability is intended to keep GPU clusters at peak utilization, directly impacting the speed of job completion in high-performance environments.
To support the massive scaling requirements of these fabrics, OcNOS 7.0 has expanded its high-density hardware support. The release adds native support for the 800G OSFP platform, specifically including the UfiSpace S9321-64EO (51.2 Tbps), which complements the existing support for 800G QSFP-DD. This dual support is designed to provide operators with greater flexibility in high-density builds. As the global AI data center networking market is projected to reach $12.80 billion by 2026, the company is targeting organizations looking to move away from locked-in proprietary systems toward more scalable, open architectures that can handle the intense throughput demands of modern generative AI workloads.
Converging IP and Optical Layers for Service Providers
For service providers, the transition from legacy OTN/LDP equipment to low-latency 5G services requires a more programmable and resilient network architecture. OcNOS 7.0 addresses this through support for Segment Routing (SR-TE) and Flexible Algorithms, which the company claims allows for a deterministic traffic path and a simplified, BGP-free core. The software is specifically built for Metro Transport and Access Aggregation use cases, introducing support for the UfiSpace S9502-12SM-NT and Edgecore AS5916-54XL platforms to facilitate high-capacity traffic aggregation across converged IP/Optical transport networks.
A significant component of this convergence is the integration of ZR/ZR+ coherent transceivers directly into the routing platform. This IPoDWDM approach allows operators to manage the optical layer from the routing interface, potentially eliminating the need for separate, standalone optical management systems. Furthermore, the inclusion of a native K3S container runtime enables the router to function as a service platform, allowing operators to run custom monitoring or security applications directly on the hardware. This "all-in-one" capability is intended to support 5G Multi-access Edge Computing (MEC) by removing the requirement for additional edge servers. To support proactive operations, the platform has shifted from traditional polling to on-change gNMI streaming telemetry, delivering network updates in milliseconds.
Key Takeaways
- Over 100 service providers and data center operators are currently using OcNOS 7.0 to manage AI infrastructure and IP/Optical convergence.
- The software supports 800G OSFP platforms, including the UfiSpace S9321-64EO (51.2 Tbps), to assist in scaling massive GPU fabrics.
- IP Infusion claims its disaggregated approach can potentially reduce total cost of ownership by 40–60% compared to vertically integrated, proprietary solutions.
TechInsyte's Take
In our view, the adoption of OcNOS 7.0 signals a maturing market for disaggregated networking, particularly as the "AI tax"—the loss of productivity caused by network congestion—becomes an unmanageable cost for enterprises. By focusing on Dynamic ECN and 800G flexibility, IP Infusion is directly targeting the most significant bottleneck in the AI lifecycle: the gap between raw compute power and data delivery. Furthermore, the move toward IPoDWDM and integrated ZR/ZR+ optics is not just a cost-saving measure; it is a strategic necessity for service providers facing the complexity of 5G and edge computing. If operators can successfully collapse these layers, they will gain the architectural agility required to respond to rapid shifts in transceiver pricing and hardware availability. This shift away from proprietary, vertically integrated "black boxes" toward open, programmable software is likely to accelerate as the AI data center market nears its projected $12.80 billion valuation.
Questions & Answers
How does OcNOS 7.0 specifically address the risk of GPU underutilization in AI clusters?
The software employs Dynamic ECN, which provides intelligent, real-time congestion management. This is designed to ensure that AI training data flows without interruption, preventing the network from slowing down training jobs and ensuring that expensive GPU hardware remains at peak utilization.
What are the primary cost-saving implications of adopting this software over proprietary systems?
IP Infusion suggests that by using its open network software, operators can potentially reduce their total cost of ownership (TCO) by 40–60% compared to closed, vertically integrated solutions. This is achieved through increased flexibility in the supply chain and optimized hardware utilization.
In what ways does OcNOS 7.0 simplify the management of converged IP and optical networks?
The platform integrates ZR/ZR+ coherent transceivers directly into the routing interface, allowing operators to manage the optical layer within a unified workflow. This eliminates the need for separate optical management systems and reduces operational complexity across the IP and optical layers.
How does the platform support 5G Multi-access Edge Computing (MEC) requirements?
The software includes a native K3S container runtime, which allows operators to run custom security or monitoring applications directly on the router. This capability enables an "all-in-one" solution that can potentially eliminate the need for additional edge servers at the network edge.
Source: Businesswire
