NewsNTech
Optical transceivers, the modules that convert electrical signals to light pulses and move data across a data center's fiber links, define the bandwidth ceiling for large AI training clusters.
Zhongji Innolight, a Chinese maker of these components, raised $6.8 billion in its initial public offering and began trading on the Hong Kong Stock Exchange on Thursday.
Where the transceiver sits in the stack The optical transceiver works at the physical layer between compute hardware and network fabric.
Its throughput ceiling sets the rate at which a training cluster can move model weights and intermediate tensors between accelerator nodes.
Keep reading