Larch Networks and Dorado have joined forces to deliver a fully managed, low-power, low-latency AI fabric built entirely on open Ethernet components. This commercially available solution, demonstrated at the OCP Global Summit, offers enterprises a clear alternative to the proprietary and often inaccessible InfiniBand networks that have dominated AI infrastructure. By combining Larch Networks’ hardened SONiC-based switches with Dorado’s Cruz orchestration platform, organizations can now deploy high-performance AI clusters without the extreme supply shortages, multi-year lead times, or complex end-use authorizations that have plagued traditional proprietary networking.
The joint solution directly addresses the availability crisis in AI networking. While hyperscalers have long enjoyed access to advanced AI infrastructure, mainstream enterprises and service providers have struggled to remain competitive. Larch Networks and Dorado Cruz change that by leveraging existing Ethernet fabrics powered by the Marvell TeraLynx 10 switch architecture. This approach not only ensures immediate availability but also delivers lower power consumption and ultra-low latency, reducing total cost of ownership while making advanced AI capabilities accessible to real users and real markets.
What is the Larch Networks and Dorado Cruz AI Fabric?
The Larch Networks and Dorado Cruz AI fabric is a fully integrated networking solution designed to support demanding AI workloads. At its core are TeraLynx 10 switches arranged in a spine-leaf topology, running Larch Hardened SONiC—a customized, enterprise-grade network operating system that provides stability, performance, and flexibility. These switches seamlessly connect GPU servers equipped with ConnectX-7 network adapters, creating a high-speed, low-latency data path essential for distributed AI training and inference.
Also worth a look: Intelligent Memory Review: Reliable DDR4 & eMMC for Telecom and Innovorbs Review: Affordable VMware Managed Services for Enterprises.
Complementing the hardware is Dorado Cruz, a centralized orchestration and monitoring platform. Cruz automates device discovery, configuration, and provisioning, while offering real-time health monitoring, telemetry, and workload-aware tuning from a single pane of glass. This combination allows operators to run a high-performance, open AI network with the standard operational tooling they expect from traditional enterprise systems, eliminating the complexity often associated with open networking environments.
What stands out
Open Ethernet architecture
Unlike proprietary InfiniBand networks, this solution uses standard Ethernet components that are widely available and interoperable. This openness gives operators flexibility and freedom from vendor lock-in, a critical advantage for long-term infrastructure planning.
Immediate availability
While InfiniBand suffers from extreme supply shortages and multi-year lead times, all components of this AI fabric are commercially available off-the-shelf. Enterprises can deploy AI clusters now, not years from now.
Low power and ultra-low latency
The Marvell TeraLynx 10 switches are designed for power efficiency and low latency, making them ideal for the demanding, data-intensive workloads of AI. This reduces operational costs and environmental impact.
Centralized orchestration and observability
Dorado Cruz provides a single pane of glass for managing the entire fabric. Automated device discovery, configuration, health monitoring, and workload-aware tuning simplify operations and reduce the need for specialized networking skills.
Flexible accelerator support
The open architecture allows organizations to choose from a range of AI accelerators, not being tied to a single vendor. This further democratizes AI infrastructure and enables cost optimization.
Who it’s for
This solution is designed for mainstream enterprises and service providers that need to deploy AI clusters to remain competitive but have been hindered by the exclusivity and scarcity of proprietary AI networking technologies. It is particularly suited for organizations that value open standards, want to avoid vendor lock-in, and require immediate availability without the typical lead times of InfiniBand.
It also appeals to operators who are already familiar with Ethernet networking and want to leverage their existing skills and infrastructure. The combination of Larch Hardened SONiC and Dorado Cruz makes the transition to AI networking smoother, as the operational tooling mirrors that of traditional enterprise systems.
What makes it worth considering
The Larch Networks and Dorado Cruz AI fabric stands out because it solves a critical pain point: the inaccessibility of AI infrastructure for non-hyperscaler organizations. By using commercially available Ethernet components, it bypasses the supply chain bottlenecks that have made proprietary AI networking nearly impossible to obtain for many enterprises. The lower power consumption and reduced total cost of ownership further strengthen the business case.
Additionally, the solution’s emphasis on ease of deployment and operation is a significant advantage. Dorado Cruz’s automated orchestration and observability features mean that teams do not need deep specialized expertise to manage high-performance AI networks. This lowers the barrier to entry and allows organizations to focus on their AI workloads rather than on network complexity. For any enterprise looking to build or expand AI capabilities without the constraints of proprietary systems, this solution is a compelling choice.
Verdict
Larch Networks and Dorado Cruz deliver a compelling open Ethernet AI fabric that bypasses proprietary bottlenecks, offering immediate availability and lower TCO for enterprises. With its hardened SONiC switches, centralized orchestration, and flexible accelerator support, it democratizes AI infrastructure and brings advanced capabilities within reach of mainstream organizations. This is a strong recommendation for any business seeking a practical, scalable path to production AI.
Frequently asked questions
It is an enterprise-grade, custom network operating system based on SONiC, optimized for stability and performance in data center environments.
Cruz provides centralized network orchestration, observability, automation, and tuning for AI fabrics, including device discovery, configuration, health monitoring, and workload-aware tuning.
The fabric uses Marvell TeraLynx 10 switches in a spine-leaf topology, running Larch Hardened SONiC and connecting GPU servers with ConnectX-7 adapters.
Yes, all components are commercially available off-the-shelf, offering immediate availability without the long lead times of proprietary AI networks.
It targets mainstream enterprises and service providers that need to deploy AI clusters without relying on hyperscaler-only or proprietary InfiniBand infrastructure.
Yes, the open architecture allows flexible choice of alternate AI accelerators, further democratizing AI infrastructure.
It was showcased at the OCP Global Summit, booth #61, as a live, fully operational AI fabric.
