日本語 
Hyper Photonix showed up at OFC 2026 with a very clear signal. Current data center networks are not built for what AI is throwing at them. Their new 1.6T silicon photonics transceiver is basically an attempt to deal with that pressure. It runs 200G per lane and hits 1.6 Tbps overall, which sounds like just another speed jump, but that is not the full story here.
The bigger shift is in how they are handling connectivity. Instead of sticking to traditional fiber setups, they are pushing multicore fiber into the mix. That changes things at a practical level. Cabling gets lighter. Space usage improves. Airflow inside racks gets easier to manage. Anyone running large AI clusters knows this is where things quietly fall apart. It is not just compute. It is wiring, heat, and power that become the real problem.
They are also lining up production expansion across China and Mexico, which tells you this is not a lab experiment. There is demand coming, especially from hyperscalers already deep into AI infra builds.
Also Read: Mitsubishi Heavy Industries Unveils 10MW-Class Chiller for Next-Gen AI Data Centers
Zoom out a bit and this fits a larger pattern. Optical networking is being forced to evolve because AI is not slowing down. More bandwidth alone is not enough anymore. Density and efficiency are becoming just as critical. Hyper Photonix is clearly building for that reality, not the old one.
English