Cisco Bets on Optics as Salience Labs Raises $125M for All-Optical Switching

Cisco and Nvidia Corp. have announced plans for USFP, or Ultra-Dense Small Form Factor Pluggable, an optical module built for 400 gigabits per lane. Current optical form factors top out at 200 gigabits per lane, Chopra said. The companies aim to drive an open ecosystem for high-bandwidth, efficient optics that is about four times denser than today's OSFP 1.6T and about 1.3 times more power-efficient. The optics work rests on Cisco Silicon One, a chip architecture announced in December 2019 that spans about 100 gigabits to 100 terabits. Chopra said the premise that locality defines function is flawed and that what matters is securing and deploying different sizes anywhere needed.

Salience Labs Inc. said it raised $125 million in Series B funding led by Bullhound Capital. Intel Capital and AMD Ventures participated through their investment arms, along with new investors SBI Group, Hudson River Trading and British Business Bank; existing backers Applied Ventures, Oxford Science Enterprises, Cambridge Innovation Capital and Silicon Catalyst Ventures returned. The startup develops all-optical circuit switching technology for AI data centers. CEO Vaysh Kewada said traditional networking fabrics based on electronic switches and repeated optical-to-electrical-to-optical conversions cannot keep up with the throughput demands of AI computing clusters made up of tens of thousands of GPUs and other accelerators.

Salience Labs recently demonstrated what it called the world's first 64-port all-optical photonic switch at the ECOC 2026 conference in September. The company said denser 128-port and 256-port configurations are under development. It cited data from Cignal AI forecasting that the optical circuit switching market will grow to more than $8 billion by the end of the decade. The new funding will go toward volume manufacturing, geographic market expansion and research and development of higher-density switches. Intel Capital's Srini Ananth said optical switching can unlock significant gains in performance and efficiency and that Salience Labs stands out through its integrated silicon photonics architecture and proprietary optical amplification.

Vast Data Inc. is betting that AI pipelines assembled from separate best-of-breed tools cannot keep sensitive data in sync. Vaughn Stewart, vice president of systems engineering at Vast, said in an interview with theCUBE Research's John Furrier at the AI Data Pipeline Forum that such pipelines create too much exposure to data that is not in sync between updates and vector embeddings, or between user access controls on source data and what is provided through an inference query. Vast raised about $1 billion at a $30 billion valuation in April 2026, according to SiliconANGLE. Its disaggregated, shared-everything architecture combines storage systems and data-processing tools, Stewart said.

Stewart estimated that about half of the world's data still sits in corporate data centers or colocation facilities because it is sensitive and in regulated industries, and much of it legally cannot reach a public cloud frontier model. Vast announced DataEnclave in September 2026 with partners including Nvidia Corp. and Cohere Inc. The confidential AI runtime runs models in secured virtual machines, requires remote cryptographic attestation before keys are released and logs all access to a tamper-proof database. Stewart said that if there is no attestation, there is no access to the keys for the model and data.

ScaleFlux Inc. is targeting the memory and storage bottleneck that leaves expensive GPUs waiting for data. Pichai Balaji, vice president of products and marketing at ScaleFlux, said a GPU operates at about 20 to 22 terabytes per second, while DRAM is almost 37 to 40 times slower and storage is almost 800 times slower. He said the bottleneck is created and amplified as data moves from the GPU to memory and then to storage. ScaleFlux builds SSD controllers and Compute Express Link memory controllers, pairing the silicon with firmware and a full reference architecture.

ScaleFlux's AI-optimized SSD platform targets key-value cache offload, the store of conversational context that inference workloads use. Balaji said NAND need not be used only as a storage device and can be used as a memory device, with an intelligent SSD controller or CXL memory controller taking advantage of NAND characteristics. The company has demonstrated more than 200 flexible data placement streams and 7 to 10 drive writes per day on standard triple-level-cell NAND. Because ScaleFlux is NAND agnostic, customers can build around whatever flash they can source in a tight supply market, Balaji said.