Salience Labs unveiled a 32-port all-optical silicon photonic switch designed to improve performance and efficiency in AI data center networks. The Oxford-based company said the device represents the first commercial product in a roadmap that will scale to 64- and 128-port optical circuit switching (OCS) systems aimed at supporting next-generation AI clusters. The switch replaces traditional optical-electrical-optical (OEO) switching elements with a fully optical architecture designed to reduce latency, power consumption, and system complexity.
The company positions the technology as a networking layer optimized for AI workloads that require predictable, low-latency communication across thousands of GPUs. Salience Labs said its integrated optical switching architecture can connect large GPU clusters across racks while remaining compatible with existing optical transceivers and data center infrastructure. The switch supports high-speed signaling tested at up to 200G (100 Gbaud PAM4) while fitting within a compact footprint significantly smaller than a 1RU device.
Salience Labs also highlighted a growing ecosystem of industry collaborators supporting development and validation of its optical switching platform. The company is working with Tower Semiconductor on photonic integrated circuits based on Tower’s silicon photonics platform, while Keysight Technologies is providing validation tools through its AI Data Center Builder environment. Salience Labs plans to demonstrate its 32-port optical circuit switch and discuss its roadmap at OFC 2026 in Los Angeles.
• 32-port optical circuit switch designed for AI data center networking fabrics
• Optical architecture eliminates OEO conversions to reduce latency and power consumption
• Up to 8× power savings compared with conventional electronic switching approaches
• Up to 80% improvement in tokens-per-second performance by reducing tail latency
• Supports data rates tested at 200G (100 Gbaud PAM4)
• Compact footprint designed to fit within a fraction of a 1RU form factor
• Product roadmap includes future 64- and 128-port optical switching systems
• Partnerships include Tower Semiconductor and Keysight Technologies
“Optical switching is moving networks from electronic packet routing to highly predictable, energy-efficient optical connectivity. We are transforming the networking layer, unlocking the ability to extend scale-up and scale-out networks across the datacenter,” said Vaysh Kewada, CEO and co-founder of Salience Labs.
🌐 Analysis
Optical circuit switching has re-emerged as a candidate architecture for AI clusters where predictable traffic patterns and high-bandwidth GPU fabrics reduce the need for traditional packet switching. Startups and research groups are exploring photonic switching as a way to reduce power consumption and latency inside massive AI systems, particularly as scale-up and scale-out fabrics expand across thousands of accelerators.
Salience Labs joins a growing ecosystem exploring photonic switching technologies alongside developments in co-packaged optics, linear pluggable optics, and silicon photonics interconnects aimed at supporting large-scale AI infrastructure. Industry efforts around optical networking fabrics are increasingly focused on improving bandwidth efficiency, reducing tail latency, and lowering power consumption in GPU-dense data centers.