Open AI Networking

Deploy a high-performance, lossless 800G Ethernet fabric that maximizes the value of your AI investment.

Learn More in OcNOS AI Fabric Solution Brief

Networking Stack for AI Factory

An AI Factory is a high-performance ecosystem where compute, storage, and networking work in perfect sync. The network serves as the central nervous system, crucial for maximizing efficiency and ROI. To accelerate Job Completion Time (JCT) and ensure high GPU utilization, a fine-tuned network fabric is essential.

As the “central nervous system” of your AI Factory, OcNOS provides the intelligent networking foundation you need to handle the most demanding workloads. The right network fabric is key to preventing bottlenecks and ensuring your massive hardware investments are fully utilized.

Accelerate Training Times

The right fabric provides the ultra-low latency and high throughput needed to significantly reduce Job Completion Time for complex AI models.

Enable Scalable Growth

A modern network architecture allows you to seamlessly add compute and storage resources, enabling a pay-as-you-grow model that scales to thousands of nodes.

Optimize Your Return on Investment

Eliminate wasted GPU cycles caused by network bottlenecks, ensuring your high-value hardware is always working at peak performance to maximize your business value.

“Partnering with IP Infusion to deploy OcNOS-DC has revolutionized our ability to deliver high-performance GPU-based services. The AI-optimized features, seamless orchestration integration, and disaggregated architecture allow us to scale efficiently while maintaining the low-latency, lossless connectivity our customers require for their AI/ML workloads.”

George Cvetanovski, Founder and CEO of HYPERSCALERS

The OcNOS Advantage

Proven performance, streamlined automation, and cost-effective scalability for your AI projects.

Engineered for Lossless AI Performance

OcNOS ensures lossless RoCEv2, ultra-low latency, and intelligent traffic prioritization, guaranteeing your AI models converge faster and your training times are reduced.

Your Path to Seamless Automation

With OcNOS comprehensive APIs (Netconf, gNMI, OpenConfig) and dedicated Ansible Collection, you can deploy and orchestrate your entire fabric with speed and consistency.

Carrier-Grade Reliability & Scale

Trusted by global service providers and data centers, OcNOS ensures your AI fabric remains agile and performs consistently as you scale to thousands of GPUs.

Cost Efficiency & Openness

Open architecture gives you the freedom to choose from the industry’s widest ecosystem of whitebox hardware and optics, enabling significant TCO reduction.

Learn More in OcNOS AI Fabric Solution Brief

How OcNOS Creates a Lossless AI Fabric

Simplified diagram of an OcNOS AI Fabric illustrating lossless communication pathways.

OcNOS: Orchestrating an End-to-End Lossless AI Fabric

RoCEv2 (RDMA over Converged Ethernet)

The core protocol for AI networking. RoCEv2 allows the Network Interface Card (NIC) of one server to write data directly into the memory of another, bypassing the CPU and OS kernel. This “zero-copy” approach provides the ultra-low latency essential for AI workloads but requires a guaranteed lossless network.

Priority Flow Control (PFC)

PFC is the ultimate safeguard against packet loss. If a switch’s buffer becomes critically full, PFC sends a pause frame for the specific RoCEv2 traffic class, halting transmission to prevent drops. OcNOS extends this capability over Layer 3 (PFC over L3), which is essential for maintaining a lossless domain across the entire leaf-spine fabric.

Explicit Congestion Notification (ECN)

ECN is the proactive partner to PFC. Instead of waiting for buffers to become full, OcNOS-powered switches monitor queue depth and mark packets as congestion begins to build. This signals the AI servers to gracefully slow their transmission rate, avoiding the need for disruptive PFC pauses in most scenarios.

Dynamic Load Balancing (DLB)

A hardware-accelerated feature essential for AI’s “Elephant Flows.” Unlike static hashing (ECMP), DLB actively monitors the real-time load of all available paths. If congestion is detected, OcNOS and the switch hardware dynamically and non-disruptively reassign flows to less-utilized links, ensuring maximum fabric throughput.

Enhanced Transmission Selection (ETS)

ETS is a Quality of Service (QoS) feature that allocates a guaranteed minimum amount of bandwidth to the high-priority RoCEv2 traffic class. This ensures that critical AI communications are never starved for resources, even in a busy network.

DCB Exchange Protocol (DCBX)

DCBX automates the configuration of Data Center Bridging features. It allows the AI server’s NIC and the OcNOS leaf switch to automatically negotiate and synchronize their PFC, ECN, and ETS settings, simplifying deployment and eliminating configuration errors.

Talk to an Expert

OcNOS 800G Switches

We’ve done the robust testing and validation, so you can choose from the industry’s best hardware with complete confidence.

Vendor	Model	ASIC	Form Factor	800G QSFP-DD Ports	OcNOS SKU
Edgecore	AS9817-64D	Tomahawk 5	2 RU	64	OCNOS-DC-IPBASE-51200-TH5 OCNOS-DC-PLUS-51200-TH5
UfiSpace	S9321-64E	Tomahawk 5	2 RU	64	OCNOS-DC-IPBASE-51200-TH5 OCNOS-DC-PLUS-51200-TH5

See All Qualified Switches

Qualified 800G Optics

OcNOS offers the widest support for optics, ensuring every customer can select their preferred optics supplier.

Vendor	Part Number	Type	Reach	Connector
ATOP	IPI-AT-APCA80-DDC010A-30	ACC	1m	QSFP-DD
ATOP	IPI-AT-APCA80-DDC020A-30	ACC	2m	QSFP-DD
ATOP	IPI-AT-APCA80-DDC030A-30	ACC	3m	QSFP-DD
ATOP	IPI-AT-APCA80-DDC040A-26	ACC	4m	QSFP-DD
ATOP	IPI-AT-APCA80-DDC050A-26	ACC	5m	QSFP-DD
HYPER SCALERS	IPI-HY-QDD-800G-SR8	QSFP-DD800 SR8	OM3 60m & OM4 100m	QSFP-DD \| MPO16
Edgecore	IPI-EC-ET8002-SR8	QSFP-DD800 SR8	OM3 30m & OM4 50m	QSFP-DD \| MPO-16 APC
Smartoptics	IPI-SO-TD8003-SC4C-SO	QSFP-DD800 2FR4	2km	QSFP-DD \| Dual-LC

Resources & Technical Documentation

Solution Briefs & White Papers

Solution Brief: OcNOS for AI Fabric

Discover the technical advantages of building a high-performance, lossless fabric for demanding AI/ML workloads with OcNOS.

Learn More

White Paper: Migrating to Hybrid Cloud-Native Data Center

Learn the strategies and benefits of migrating to a modern, cloud-native architecture with an open networking foundation.

Read Whitepaper

Solution Brief: 400G Upgrade for Data Centers with OcNOS

Discover how 400G switches with OcNOS save power, space, and cost while liberating data centers from scale limitations.

Download Brief

Product & Technical Documentation

OcNOS DC Feature Matrix

Explore the comprehensive list of protocols and features, including DCB, supported by the OcNOS Data Center platform.

View Features

Hardware Compatibility List

Browse the ecosystem of open, whitebox hardware platforms compatible with OcNOS, including 800G Tomahawk 5 switches.

View HCL

Customer Stories & News

Press Release: IP Infusion to Showcase OcNOS Data Center for AI/ML Networking

IP Infusion will demonstrate its OcNOS Data Center networking operating system, optimized for Ethernet-based AI/ML data center fabrics, at the OCP APAC Summit 2025.

Case Study: Scott Data Deploys Network Disaggregation

Learn how Scott Data, a Tier III Certified data center, chose OcNOS for its comprehensive features and exceptional support.

Related Products

OcNOS Data Center

A carrier-grade NOS for your data center, with features for seamless operations and unmatched scalability.