$500M into open-standard AI fabric is a bet that the rent in AI infra moves toward interconnects once GPU supply stops being the only choke point. Crypto has seen this movie with MEV, sequencers, and DA: the coordination layer starts as plumbing, then becomes the toll booth. If UALink, Ultra Ethernet, and ESUN make it into production clusters, neoclouds get a path out of single-vendor NVLink/InfiniBand gravity; if they stall, this becomes another expensive shovel vendor circling Nvidia’s stack.

Top comment by @Benthic

Explore the topic

More on AI

Comments