Dell’Oro Group reports on Optical Circuit Switch technology for AI cluster network efficiency

The latest analyst report highlights the rising demand for advanced network infrastructures to support expanding Artificial Intelligence (AI) clusters, emphasizing the role of Optical Circuit Switches (OCS) in enhancing Graphics Processing Unit (GPU) interconnectivity efficiency. This matters to enterprise IT and technical leaders aiming to optimize AI cluster performance and energy consumption.

Market Overview

AI clusters are rapidly increasing in scale, growing from hundreds of thousands to potentially millions of GPUs, which will require substantial networking capacity to handle inter-GPU communications. Analysts predict that about 124 gigawatts of compute capacity, equivalent to over 70 million GPUs, will be deployed within five years, making network performance a critical factor in AI cluster operations.

Key Findings

The report identifies that as AI clusters grow, the number of required interconnects rises exponentially, driving up cost, power use, and latency. The network must operate near full efficiency to ensure optimal GPU utilization. Additionally, AI data center networks see faster refresh cycles than traditional environments, with switch port speeds advancing from 800 Gbps expected in 2025 to 3.2 Tbps by 2030, necessitating frequent electrical layer upgrades.

Technology or Trend Analysis

Optical Circuit Switches, also known as Optical Cross-Connects, deliver direct optical paths between endpoints, bypassing traditional packet routing to reduce latency and power consumption. First deployed by Google nearly ten years ago, Optical Circuit Switch (OCS) technology avoids the need for frequent upgrades tied to increasing link speeds, offering a speed-agnostic alternative to electrical switching fabrics. The elimination of optical-electrical-optical conversions contributes to lower latency and energy costs.

OCS technology, used in carrier networks and wide-area networks for decades, is considered field-proven. The analysis notes that new OCS products are entering the market tailored to AI data center requirements, aiming to optimize network efficiency in large-scale AI and High performance computing (HPC) environments.

Forecast or Analyst Outlook

The report suggests that OCS technology could become a component of future AI network designs, supporting the accelerating growth and complexity of AI cluster infrastructures. The network evolution is projected to outpace GPU development speed due to the increasing need for efficient data transfer and bandwidth.

Enterprises and technical decision-makers are encouraged to consider OCS as a viable option for scaling AI networks given its operational history and efficiency advantages.

Conclusion

The analyst report details how AI infrastructure growth necessitates network designs that can keep pace with GPU expansion. OCS technology presents a proven, energy-efficient approach for AI cluster networks, potentially influencing future infrastructure choices. This Analyst Signals brief reflects a neutral, fact-based summary of the original research note.