Breaking Free From Proprietary Hardware: How Open Networks Are Powering the AI Boom
San Jose, Monday, 12 October 2026.
Larch Networks and Dorado are unveiling an open Ethernet system, providing businesses with a lower-cost, flexible alternative to proprietary hardware for demanding artificial intelligence workloads.
Strategic Shift Toward Open Networking Architecture
On October 9, 2026, network solutions developer Larch Networks and software management provider Dorado Software announced a joint demonstration of a commercially available Ethernet artificial intelligence fabric and orchestration system [2]. The live demonstration is scheduled to take place from October 13 to October 15, 2026, at booth #61 during the Open Compute Project (OCP) Global Summit [1][2]. Designed to address persistent hardware supply shortages and eliminate reliance on proprietary protocols such as InfiniBand, the architecture provides enterprises with an open, low-latency Ethernet alternative for high-performance computing workloads [1].
Technical Architecture and Fabric Orchestration
The core infrastructure utilizes a spine-leaf network topology constructed with ten switches operating on Marvell TeraLynx 10 architecture and running Larch Hardened SONiC open-source network software [1][2]. This setup connects directly to graphics processing unit (GPU) servers equipped with ConnectX-7 network adapters [1][2]. Orchestration and monitoring are handled through Dorado’s Cruz AI platform, which integrates automated device discovery, provisioning, real-time health telemetry, and workload-aware tuning to manage operational complexity across complex data center environments [1][2].
Autonomous Management and Operational Evaluation
A key component of Dorado’s Cruz platform is its native agentic AI interface, allowing system operators to query and manage live network infrastructure using natural language while enforcing human-in-the-loop authorization [1]. This system delivers single-pane-of-glass operational visibility across fabric switches, compute clusters, and rack-scale hardware [1]. However, adoption across enterprise IT environments requires rigorous acceptance testing; analysts highlight that prospective buyers must evaluate real-world evidence covering failover state reporting, dependency-aware configuration rollbacks, and clear support boundaries for underlying open-source components [2].