Delos Data showed a modular server platform at Computex 2026 that gives AI chip startups a shortcut to rack-scale architectures. The chassis has 36 OSFP ports per four OAM sockets, delivering 3.6 TB/s per chip of interconnect. This matches Nvidia's Rubin GPUs while using standard pluggable transceivers.
Tap to vote and see what everyone thinks.