Optical Circuit Switching Market – 2Q26

This report is an update to Cignal AI’s OCS Report, last published in 4Q25 with a forecast update published in 1Q26. For this update, Cignal AI has included a deeper analysis of both the evolving Scale Out (TPU) application and the emerging Scale Up (GPU) application, expected to have significant contributions in 2028/29, including OCS port to XPU ratios. This analysis includes an updated overall market size and market size by application. Additionally, the list of OCS vendors has been refreshed and updated.

OCS shipments and forecasts are incorporated in the Optical Component report Excel file and are updated quarterly. As with all Cignal AI reports, interim updates are available via the Live Excel file if something changes between reports. An Interactive OCS Market Sizing Tool is also included below as the projections of TPU and GPU shipments in later years is a matter of opinion, not fact. Adjustments can be made to see the impact on market size.

For a take on OCSs from a variety of OCS vendors, subscribers are encouraged to watch the webinar Optical Circuit Switching for AI Scaling and Datacenter Automation from July 2025.

Clients login to access

Key Takeaways:

• The total market for OCSs will surpass $8 billion by 2030 – a sharp increase over prior forecasts, driven almost entirely by new visibility into the Scale Up (GPU) opportunity.
• The only volume application today is Scale Out for TPU-based (Google) networks (formerly AI Cluster Reconfiguration), which will remain the dominant part of the market for the next 2-3 years. Next-generation Google TPU architecture will use fewer OCS ports, not more, per TPU.
• Near-term revenue remains highly concentrated in Google, whose OCS development has been largely internal to date. Cignal AI’s forecast assumes that development shifts to third-party vendors over the next year, which – together with the emergence of Scale Up – broadens the market to the merchant supply base.
• Nvidia will integrate OCSs into its coming NVL576 architecture. With the introduction of optics in the Scale Up part of the GPU-based (Nvidia) network, OCS for Scale Up will quickly rise in revenue, rivaling and ultimately exceeding Scale Out by the end of the forecast.
• The number of OCS vendors continues to expand. This report includes more than 20 vendors, and more are expected to emerge from labs and startup incubators over the next year.
• There are currently multiple approaches to OCS technology, including MEMS, liquid crystal, piezoelectric, robotic, and silicon photonics. Each one has a different set of capabilities, and there are optimal applications for each inside the datacenter.
• Reliability is the key characteristic of an OCS inside the datacenter, especially for AI applications for which even transitory failures can cause massive disruptions. The value of other features (radix, speed, loss) varies by application.
• OCS deployment will impact datacom transceiver requirements. Most notably, FR optics and future standards such as the OCI MSA, which use a single fiber, will dominate over DR optics (which use multiple fibers). Also, customizations like higher power optics and circulators for bi-directional operation will be more common.
• Coherent Lite could become a factor at 2.4T, as Google is driving an interim solution until 400G/lane (3.2T) is available in volume. (see Coherent Lite: Technology, Applications and Forecast – 2Q26)

Changes from the Previous OCS report:

• Adjustments to the Scale Out (TPU) application (formerly AI Cluster Reconfiguration), including analysis of the effect of Google’s upcoming Boardfly configuration and lower OCS port to TPU ratio.
• Significant improvements to the Scale Up (GPU) model for multi-rack optical interconnects, including OCS Port to GPU ratios and total market size through 2030.
• Updates to the OCS vendor list, including the latest information from established vendors and more startup additions.
• As most readers are now familiar with OCSs, the explainer sections (OCS Definition, OCS Key Characteristics, OCS Technologies) have been moved to the end of the report for reference.

Table of Contents

• Introduction
• Key Takeaways
• Changes from the Previous Report
• Overview
• Datacenter Applications for OCSs
  • Scale Out (TPU Networks)
  • Spine Layer Replacement (aka Restriping)
  • Scale Up (GPU Networks)
  • Multi-Layer Resiliency
  • Mixed Network Bypass
  • Pooled Resources / Datacenter Disaggregation
  • Virtual Meet-Me Rooms
  • Campus Datacenter Interconnect
  • Telecom Applications
• OCS Market Size
  • Scale Out (TPU Networks)
  • Scale Up (GPU Networks)
  • Spine Layer Replacement
  • Other Applications
  • Revenue Summary
  • Ports Shipped Summary
• Vendors Offering OCSs
  • Google
  • Lumentum
  • Coherent
  • Huber+Suhner
  • Telescent
  • iPronics
  • DiCon Fiberoptics
  • Triple-Stone
  • nEye
  • Eoptolink
  • Molex
  • Accelink
  • Terahop (Innolight)
  • Salience Labs
  • Calient
  • Corespan Systems
  • Others
• The Definition of an Optical Circuit Switch
• OCS Key Characteristics
  • Reliability
  • Radix
  • Loss
  • Speed
• OCS Technologies
  • MEMS-Based Switches
  • Liquid Crystal-Based Switches
  • Robotic Switches
  • Piezoelectric Switches
  • Silicon Photonics
• Datacom Transceiver Impact
  • Higher Optical Power
  • Bi-Directionality
  • Wavelength Division Multiplexing (WDM)

Overview

Optical Circuit Switching is not a new concept. In the boom days of the late 90s, a plethora of companies were founded based on using optical switching to automate everything from labs to long-haul networks. Few of those entities still survive, except in limited applications such as programmable patch panels for test environments.

However, interest in OCSs for datacenters and AI applications has been reignited by Google, which demonstrated the power savings and flexibility that OCSs can provide in areas such as spine replacement and Scale Out for TPU networks. In public presentations, Google showcased 40% power savings and 30% cost savings with 50 times better long-term uptime for its switching network – numbers that are difficult to ignore. As a result, other service providers have tasked their engineering teams with investigating the viability of OCSs for use within their networks.

Meanwhile, the Google engineers who worked on those projects have, via career transitions and poaching, cross-pollinated many other service providers who would like to emulate Google’s results. This has had the effect of jump-starting efforts at multiple service providers who can now build on experience from the early efforts at Google. These engineers also bring a more intimate knowledge of the benefits of OCSs with them and have catalyzed demand across a larger number of potential customers.

Optical transceiver vendors are now seeing new specifications placed on high speed (400GbE and faster) datacenter optics that are directly a result of OCS deployment. Higher output power, bi-directionality, and multi-wavelength options are required for seamless interworking with OCSs. Google has even initiated work on a 2.4T Coherent Lite project specifically to increase the speed per port in its OCS network.

For years, the OCS TAM was entirely based on lab automation and network monitoring, with total annual revenue in the low tens of millions of dollars. However, Google has spent well over $1 billion on OCSs to date. The TAM for OCSs is still in flux, with new applications yet to be proven in, but it will certainly be in the multiple billions of dollars in 2026 to high single-digit billions of dollars by 2030. The old technology and technologists are being dusted off and joined by a new set of products aimed at putting optical switching inside lucrative and fast-growing AI-centered datacenters.

Interactive Market Sizing Tool

The total OCS market size is very dependent on several highly variable assumptions ($/port, TPU shipments, etc.). Cignal AI does not have high conviction estimates of Google TPU or Nvidia GPU shipments in 2028+. Users who wish to adjust our assumptions are welcome to use Cignal AI’s interactive OCS Market Sizing Tool to change parameters in the Scale Up (GPU) and Scale Out (TPU) applications and view the impact on the market.