ニュースルーム

As AI clusters continue to scale, traditional copper interconnects and electrical packet switches (EPS) increasingly encounter power, latency, and thermal constraints. The OCS solution presented by UTStarcom at the event focuses directly on dismantling these barriers by introducing a dynamic, all-optical switching layer into the fabrics of AI data centers. By eliminating power-intensive Optical-Electrical-Optical (O-E-O) conversions and packet processing at the corresponding AI network layers, UTStarcom’s integrated OCS solution addresses these scaling constraints across three dimensions:

• Enhanced AI Cluster Performance: Operating alongside traditional EPS in hybrid AI data center architectures, OCS provides an automated, dynamically reconfigurable optical layer. By offloading high-volume AI traffic onto direct photonic paths while enabling dynamic topology changes, resource pooling, and optical-layer redundancy, it eliminates buffer-related queuing delays and cuts packet loss and latency during large-scale model training and inference. 
• Lower CAPEX and OPEX: Reduces the number of required optical transceivers while lowering power consumption, cooling requirements, and rack footprint.
• Protocol- and Bit-Rate-Agnostic Scalability: Provides protocol- and bit-rate-transparent fiber channels that that can accommodate future increases in optical line rates without replacing the physical switching fabric.

The platform remains in active development and comprises the following elements: 

• UOS64 (Silicon Photonics Platform): A 64x64-radix OCS concept prototype based on Silicon Photonics (SiPho) technology. Engineered around chip-scale Photonic Integrated Circuits (PIC), it’s designed to provide fast optical switching, high-density integration, and scalable optical path automation.
• MOS64 (MEMS Platform): A 64x64-radix OCS concept prototype built on Micro-Electro-Mechanical Systems (MEMS) micro-mirror core technology, designed to deliver low insertion loss, minimal channel crosstalk, and fast optical path switching. 
• OCS platform software & SOO Station OCS management platform: UTStarcom’s SONiC-based OCS platform software, paired with the Company’s SDN controller. Through open Southbound and Northbound Interfaces (SBI/NBI), the software layer allows upper-layer orchestrators to automate optical-path provisioning, implement topology-on-demand and streamline optical fabric management.

Debuting its OCS concept prototypes at CIOE 2026 represents a critical step in UTStarcom's AI DC network strategy. Throughout the event, the high level of interest from prospective clients and optical partners reaffirmed the necessity of transparent Layer 0/1 optical fabrics. UTStarcom looks forward to working closely with global partners to advance product maturation and accelerate real-world AI data center deployments.

For additional information, check out the CIOE2026 PR at link.

For inquiries regarding our OCS solution, please contact us at sales@utstar.com.