)
)
)
15 Sep 2026
Solving the Network Bottleneck in AI Clusters with High-Radix Switching
)
The “network wall” has become one of the primary bottlenecks in large-scale AI training and inference. While compute FLOPS have doubled every 18 months, switch ASIC capacity and radix have historically doubled only every 2.5-3 years. Switch radix in turn limits the single-hop domain size for scale-up and scale-out networks, imposing severe constraints on AI models and increasing costs for both training and inference.
In this presentation from AI Infra Summit 2026, Eridu Co-Founder and CPO Omar Hassen explains how a disruptive new approach is needed--across switching silicon, systems and network architectures--to deliver an order-of-magnitude increase in radix and flatter networks, breaking through the network wall so that AI can continue to scale rapidly and efficiently.