Shanghai-based semiconductor designer Biren Technology has unveiled a new generation of AI computing clusters that rely on optical data transmission rather than traditional electrical interconnects.
The company describes these systems as "supernodes," designed to link more than 1,000 processor cards into a single, cohesive unit.
The firm is currently in the midst of a funding round targeting HK$7 billion (approximately US$892.
This architectural shift aims to bypass the bandwidth and latency limits that currently constrain large-scale AI training infrastructure, directly challenging the cluster designs dominated by Nvidia.
The announcement comes as Biren seeks to accelerate the commercialization of its next-generation general-purpose graphics processing units.
The firm is currently in the midst of a funding round targeting HK$7 billion (approximately US$892.5 million) to support the mass production of these new chips.
By integrating near-packaged optics, Biren is attempting to solve the scaling bottlenecks that have become a critical constraint for hyperscalers building massive AI data centers.