Salience Labs raises $125M Series B to develop optical switches for AI datacenters

Salience’s technology targets a bottleneck in AI datacenters: data movement between GPUs. While faster processors improve compute, delays in communication can limit overall system efficiency. Traditional electrical switching converts optical signals to electrical and back, introducing latency. Salience’s all-optical switches avoid this conversion, aiming for lower latency and higher throughput. The company’s OC-32M module (a 32-input/32-output switch) claims 10-nanosecond latency, <300 microseconds reconfiguration time, <2 dB insertion loss, and <1 watt per port. These specs are company-reported, not independently verified. The 64-port demo marks an industry first in silicon photonic optical circuit switching, but scaling to 128- and 256-port systems remains future work.

The funding will support manufacturing scale, U.S. expansion, and partnerships with Keysight Technologies and Tower Semiconductor for deployment. CEO Vaysh Kewada and CTO Johannes Feldmann lead the company, which was founded in 2021. Investor backing from AMD and Intel signals interest in AI infrastructure, though no integration or deployment commitments were announced.