US National WireUS NATIONAL WIRE
Tech

Nvidia Backs iPronics to Scale Optical Circuit Switching

Portrait of Devon Marsh
Devon MarshSilicon Valley startups & VCSep 3AI
Nvidia Backs iPronics to Scale Optical Circuit Switching

AI-generated image · US National Wire

A new $125 million investment targets silicon photonics to reduce the size and latency of OCS hardware used in massive AI clusters.

Nvidia has joined Maverick Silicon and Light Street Capital in a $125 million investment to support the development of second-generation optical circuit switching (OCS) technology from startup iPronics, as The Register first reported.

Unlike traditional packet-switching ASICs from companies like Marvell or Broadcom, OCS functions as a high-speed optical switchboard. While Google has utilized OCS in its TPU clusters for years to manage large compute domains and hot-swap accelerators, The Register notes that first-generation OCS appliances are often bulky and slow, with reconfiguration times around 100 ms.

iPronics aims to solve these bottlenecks by replacing MEMS and LCD-based systems with a silicon photonics design that lacks moving parts. The company claims this approach enables sub-ms reconfiguration times and significantly higher density. For example, the iPronics One supports 32 ports per chip, with 72- and 144-port chips currently in development. The startup estimates it can fit as many as 720 port pairs into one rack unit, a significant reduction compared to 300-port appliances from Coherent, which generally need eight or more rack units.

This investment follows a broader push by Nvidia into optics; in March, the company invested $6 billion—split as $2 billion each—into Lumentum, Coherent, and Marvell. While OCS is best suited for environments where network paths remain relatively stable, The Register suggests it could be used in hybrid environments, combining OCS-based meshes for rack-to-rack communication with switched fabrics over copper inside the rack.

Sources

More from Devon Marsh