Summary

• Salience Labs has completed a $125 million Series B led by Bullhound Capital, with participation from AMD Ventures, Intel Capital and the British Business Bank.
• Its silicon photonic technology routes optical signals without the electrical conversions used by conventional switching architectures.
• The funding will support manufacturing expansion, US operations and future 128-port and 256-port systems.

Oxford-based Salience Labs has raised $125 million to expand manufacturing of optical networking chips, targeting the connections between AI processors that increasingly influence the cost and performance of large computing installations.

The oversubscribed Series B was led by Bullhound Capital, with participation from AMD Ventures, Intel Capital, SBI Group, Hudson River Trading and the British Business Bank. Existing shareholders, including Applied Ventures, Oxford Science Enterprises and Cambridge Innovation Capital, also participated.

Salience will use the funding to increase production capacity, support commercial deployment and expand its presence in Silicon Valley. The investment follows its demonstration of a 64-port silicon photonic optical circuit switch at the ECOC 2026 exhibition in Málaga, while larger 128-port and 256-port systems remain on its development roadmap.

The company is developing an alternative to conventional data centre switching in which optical signals are routed without being converted into electrical form at every switching point. That approach could reduce some of the power consumption, latency and equipment costs associated with connecting large numbers of AI processors.

As AI installations become larger, networking performance has a direct bearing on how effectively expensive computing resources are used. Processors working on a shared task must exchange information, and delays in that exchange can leave computing capacity waiting for data rather than performing useful calculations.

Optical switching changes the data path

Conventional data centre networks use electronic switches to direct traffic between connected devices. Fibre can carry information over substantial distances, but optical signals entering an electronic switch are generally converted into electrical form before being processed and transmitted onwards.

Optical circuit switching establishes paths for light to travel through the switching system without requiring the same conversion at the switching stage. Salience uses silicon photonics to build these capabilities into semiconductor devices, applying established manufacturing techniques to components that guide and control light.

Removing some electrical conversion stages can reduce the amount of supporting equipment required within a suitable network architecture. Salience claims its technology can lower power consumption, switching latency and cost per port compared with conventional electrical switching.

Those comparisons remain supplier performance claims. The results achieved in a complete data centre will depend on network design, workload characteristics, the equipment being replaced and the way optical and electronic systems operate together.

Optical circuit switches also perform a different function from packet switches. A circuit switch establishes connections between endpoints, while packet switching makes forwarding decisions for individual packets of information.

Some AI workloads may benefit from high-capacity optical connections maintained between processors or groups of processors. Others require frequent changes in how traffic is routed, making switching flexibility and the time required to reconfigure a connection important engineering considerations.

Operators must therefore evaluate which portions of their networks could benefit from optical circuits rather than assuming that replacing every electronic switch would produce the best result.

Salience’s technology is also intended to support increasingly dense computing systems, where connections between processors inside and between racks are becoming more demanding. Network architecture can influence how many processors are effectively able to work together on a task.

Manufacturing becomes the next constraint

The company’s 64-port demonstration at ECOC in September provided evidence of the integrated optical switching architecture operating at that scale. It did not establish that the same device has been qualified for unrestricted commercial deployment or manufactured consistently in large volumes.

Moving from a demonstrated chip to volume production requires repeatable performance across manufacturing batches, acceptable production yields and dependable testing procedures.

Photonic devices are sensitive to design and fabrication characteristics that affect optical loss, signal quality and thermal behaviour. As more connections are incorporated into a system, engineers must manage those characteristics while maintaining practical control over switching operations.

Salience’s larger planned devices will increase the number of ports available within the architecture, but the company has not disclosed completed validation results for its proposed 128-port and 256-port products.

The business traces its technology to more than a decade of research at the University of Oxford and Germany’s University of Münster. Established in 2021, it is now investing in the processes needed to manufacture products for commercial networking customers.

Salience works with Keysight Technologies on optical switching testing and with Tower Semiconductor on manufacturing development. These relationships bring relevant engineering and production capabilities, although the latest funding announcement does not specify customer shipment volumes or guaranteed factory output.

The British Business Bank has confirmed a £10 million investment within the Series B, placing public-backed British investment capital alongside specialist venture investors and technology companies.

The additional financing will also support a second West Coast location as Salience expands its engagement with computing infrastructure customers and partners in the United States.

Commercial value depends on workload performance

AI data centre operators have strong financial incentives to reduce networking overhead because processors represent a substantial capital investment and require significant electricity and cooling.

If a network allows processors to exchange information more efficiently, an operator may be able to complete workloads faster or obtain more useful computation from installed hardware. Those improvements could affect the economics of training, inference and the utilisation of expensive computing clusters.

However, the benefit will vary. A workload limited mainly by processor performance may gain relatively little from a faster switch, while one constrained by communication between processors could respond differently.

Deployment also introduces costs beyond the switching device. Optical connectivity, network management, integration, redundancy and maintenance must all be considered when comparing an unfamiliar architecture with established alternatives.

Existing networking suppliers are developing faster electronic switches and optical interconnects, so Salience will compete against technologies that continue to improve rather than a fixed industry baseline.

Its new investors include businesses with substantial interests in processor and semiconductor infrastructure, but their financial participation does not establish customer contracts, purchasing commitments or commercial performance.

The completed funding round gives Salience more resources to industrialise its technology. The evidence needed to assess its commercial progress will include manufacturing yields, qualified production systems, customer adoption and measured performance on representative AI workloads.

The 64-port demonstration establishes a technical milestone, while the next challenge is producing larger and more capable switches consistently enough for deployment in computing environments where equipment reliability and economic returns determine purchasing decisions.