At the World Artificial Intelligence Conference in Shanghai (July 17-20, 2026),
Biren Technology unveiled a next‑generation NPO optical interconnect,
distributed-decoupled supernode architecture that scales up to 1024 cards per
supernode. The company said its BR2xx series continues a chiplet architecture,
supports low-precision high-throughput FP8/FP4 compute, increases memory
bandwidth, and natively integrates supernode interconnect. Biren’s BLink2.0
protocol reportedly lets up to 1,024 GPUs share a single memory space. Product
lineup includes a 16-card standard server supernode (electrical interconnect), a
128-card high-density rack supernode (electrical), and a 1,024-card distributed,
decoupled supernode using NPO optical interconnect.
