IT Brief Asia - Technology news for CIOs & IT decision-makers
32port optical switch ai datacenter rack silicon photonics beams

Salience Labs debuts all-optical switch for AI hubs

Wed, 11th Mar 2026

Salience Labs has launched an all-optical 32-port switch for AI data centres, as operators face growing congestion from larger clusters and rising rack power density.

The Oxford-based photonics company says the product is available now. It is the first in an optical circuit switch portfolio that will also include 64-port and 128-port variants. Salience Labs positions the line for scale-up and scale-out AI networking, where traffic patterns can create bottlenecks between compute nodes and storage systems.

Analysts expect infrastructure demand to keep rising. ABI Research forecasts global data centre capacity will grow from 24.4 gigawatts to 147.1 gigawatts over the next decade, driven by enterprise AI adoption, hyperscale build-outs, and higher rack-level power density.

Salience Labs is targeting the networking layer inside those facilities, where many deployments still rely on optical-electrical-optical (OEO) conversion and electronic packet routing. Optical circuit switching routes signals through optical paths and is used in designs that favour predictable connectivity patterns and reduced switching overhead for specific traffic types.

Switch details

Salience Labs describes the 32-port unit as an all-optical, fully integrated switch architecture. It is designed to be compatible with existing transceivers and infrastructure, potentially lowering integration risk for operators running mixed-vendor networks.

The switch has been tested at data rates up to 200G using 100 Gbaud PAM4 encoding. Salience Labs also points to a compact footprint, saying it fits within a fraction of 1RU-an increasingly important consideration as operators allocate more rack space to compute and cooling.

On performance, the company says the design reduces overall network latency and removes tail latency. It links that claim to "Tokens per Second / User", a metric used to describe responsiveness in some AI inference workloads, and says its approach can deliver "up to 80% improvement" in that measure.

Power consumption is central to the product's positioning. Salience Labs says its optical architecture can eliminate the need for optical transceivers in certain switching scenarios and deliver "savings up to 8X on power versus current OEO switching solutions". It also says it reduces system costs by removing transceivers associated with traditional electronic packet switching (EPS).

Data centre operators are exploring several ways to relieve network pressure created by distributed training and inference, including faster electrical switching silicon, changes in topology, and greater use of optics. Salience Labs is among suppliers focused on photonic integration, where switching functions and optical routing are implemented with silicon photonics.

Market backdrop

Spending on networking equipment in AI-focused environments is expected to increase. Dell'Oro Group forecasts data centre switch spending in AI back-end networks will exceed USD $100 billion by 2030, driven by deployments across scale-up, scale-out, and other domains.

Salience Labs is also leaning on partnerships that combine photonics manufacturing and test. It is collaborating with Tower Semiconductor and Keysight Technologies as part of an industry ecosystem around optical circuit switching for AI infrastructure.

"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, Salience Labs.

Tower Semiconductor framed its work with Salience Labs around manufacturing readiness for photonic integrated circuits on a silicon photonics platform.

"As the adoption of AI technologies increases across industries, innovation in networking, such as optical circuit switching, is critical to meet the rapidly growing demands of AI workloads," said Dr. Ed Preisler, Vice President and General Manager of RF Business Unit, Tower Semiconductor. "Our partnership with Salience to develop advanced photonic integrated circuits (PIC)-based optical OCS for AI infrastructure built on Tower's market-leading Silicon Photonics platform is set to support customers in confidently scaling from development to high-volume production."

Keysight highlighted testing and validation for optical circuit switching in AI environments, referencing its AI Data Centre Builder tooling as part of a demonstration workflow.

"The unprecedented growth of AI is driving the industry to advance new optical technologies capable of meeting the massive bandwidth demands of next-generation AI workloads," said Ram Periakaruppan. "Through our collaboration with Salience Labs, we are showcasing an optical circuit switch test using Keysight AI Data Centre Builder that demonstrates how these innovations can improve bandwidth efficiency and reduce latency for AI workloads."

Salience Labs says it will demonstrate the 32-port optical circuit switch and outline future products at the OFC industry event alongside partners.