OptimalNets provides ON.AI, a patented software layer that dynamically optimizes network traffic for AI datacenters, integrating with existing Ethernet fabrics, SONiC switches, and RoCEv2 stacks. By continuously monitoring AI workloads and adjusting routing and congestion controls in real time, it delivers 30–50% higher performance, faster job completion, and lower tail latency without requiring any hardware changes.
Funding
Funding not disclosed
Founders
Product
Problem
AI datacenters rely on high‑performance Ethernet fabrics, but network inefficiencies and static configurations limit GPU utilization, increase tail latency, and extend training or inference times. Upgrading hardware to address these issues is costly and disruptive.
Solution
OptimalNets offers ON.AI, a patented, workload‑aware software layer that dynamically optimizes network traffic for AI workloads without requiring new hardware. The platform integrates with existing Ethernet fabrics, SONiC switches, and RoCEv2 stacks, automatically adjusting routing and congestion controls based on real‑time AI job characteristics. By aligning network behavior with the demands of training, inference, and agentic AI tasks, ON.AI delivers 30–50% higher performance, faster job completion, and improved GPU utilization while reducing tail latency. The solution is deployed as a software overlay, allowing datacenter operators to realize gains instantly and maintain compatibility with their current infrastructure.
Target Audience
Primary customers are operators of large‑scale AI datacenters and cloud providers that run GPU‑intensive training, inference, and autonomous AI workloads.
Features
- Patented workload‑aware engine that continuously monitors AI traffic patterns and reconfigures network parameters in real time
- Seamless integration with standard high‑performance Ethernet fabrics, SONiC operating system, and RoCEv2 protocols
- No hardware changes required; the software layer runs on existing switches and servers
- Performance boost of 30–50% across AI training, inference, and agentic AI workloads
- Automatic optimization of routing, flow control, and congestion management to lower tail latency
- Centralized dashboard providing visibility into network efficiency and AI job performance metrics