Silicon photonics start-up iPronics has raised US$125 million in series B funding to accelerate deployment of its optical circuit-switching technology for artificial-intelligence data centers. 

As AI clusters grow, keeping thousands of GPUs and other accelerators efficiently connected is becoming an increasingly difficult engineering challenge. Moving data among processors can consume substantial power and strain conventional electrical networking. That has sharpened interest in optical approaches that can deliver high bandwidth while reducing the number of electrical conversions along the data path.

iPronics is targeting that problem with optical circuit switching (OCS). Unlike conventional packet switches, which electronically inspect and route individual data packets, an optical circuit switch establishes a direct optical connection between selected ports. Once the path is configured, data can travel through the switching fabric without being converted from light to electricity and back again inside the switch.

That can reduce power consumption and switching overhead. But OCS also comes with challenges, namely, that the optical connections need to be reconfigured as traffic patterns change. In an AI cluster, where different processors may need to communicate intensively at different stages of training or inference, the ability to alter those corrections quickly and under software control is especially attractive.

iPronics rack-mounted technology combines a silicon photonics switching fabric with control software, telemetry and application programming interfaces. The company says the system can dynamically reconfigure connectivity within and between racks, allowing data center operators to adapt the physical network topology to changing workloads.

OCS is not a new concept, but the scale and power requirements of AI computing are giving the technology fresh momentum.

The use of silicon photonics is central to the approach. By integrating optical switching functions on chip, iPronics aims to provide a compact, solid-state alternative to optical switches based on elements moving mechanically, while increasing the density of optical connections that can be accommodated in a rack-scale system.

OCS is not a new concept, but the scale and power requirements of AI computing are giving the technology fresh momentum. The central question is whether new generations of faster, denser and more programmable optical switches can become practical building blocks for large AI clusters.

Against that backdrop, iPronics, which was founded in 2019 as a spin-off from the Universitat Politécnica de València, Spain, will use the funds to support expanded operations and accelerate commercial deployment of its technology, including through its growing US presence.

The funding round was co-led by Maverick Silicon and Light Street Capital and joined by Nvidia and other investors, bringing the company’s total funding to US$177 million.