Sponsor Editorial

NADDOD presents a high-performance network architecture designed to scale GPU clusters from 10K to 100K while retaining only leaf and spine switching tiers. By combining 1.6T Ethernet switching with multiplane Fiber Shuffle and 1.6T optical connectivity, the design avoids additional super-spine and core tiers, reducing switch hops, optics, cabling complexity, and associated latency and cost. Explore NADDOD’s approach to building flatter AI fabrics.