OCS may become the core of AI networks, with all-optical switching delivering advantages in low latency, low power consumption, and high bandwidth

Overview

This report focuses on optical circuit switches (OCS) in AI data center networks. OCS is an all-optical switching device that can route and forward data through dynamic optical path adjustment without going through the optical-electrical-optical conversion of traditional electrical switching chips. Nomura believes that as AI cluster scale expands, traditional electrical switching faces bottlenecks in SerDes speed, power consumption, and latency, and OCS is likely to play a more important role in scale-up, scale-out, and scale-across networks.

Core views

The core view is that the OCS market may be approaching an inflection point of accelerated adoption. Google has taken the lead in validating OCS benefits in lowering cost and power consumption while improving throughput in TPU networks; NVIDIA discussed its next-generation Dragonfly architecture at OFC 2026 and indicated that OCS can be used for inter-rack and inter-cluster connectivity. The report believes that, beyond Google, other hyperscale cloud vendors have also entered the testing stage, and OCS system vendors and key optical component suppliers may benefit over the next 2-3 years.

Analysis framework

The report analyzes OCS from several dimensions, including technical principles, solution pathways, application scenarios, comparison with CPO and traditional switches, market opportunity, and supply chain breakdown. It compares four solution types—MEMS, DLC, DLBS, and SiPh—and assesses the commercialization pace in combination with progress by Google, NVIDIA, Cignal AI, and industry chain vendors.

Methodology notes

• Technology route comparisonComparison of four OCS solution types

  MEMS, DLC, DLBS, SiPh-based OCS

  The report compares differences among various OCS solutions in port count, per-port cost, insertion loss, switching time, crosstalk, and reliability, concluding that MEMS is currently the most mature with a more stable supply chain, while liquid crystal solutions have physical-property advantages but face greater challenges in mass production and yield.

• Network architecture analysisscale-up / scale-out / scale-across

  Layered application scenarios in AI cluster networks

  OCS can be used for scale-up resource pool reconfiguration in Google TPU systems, may replace some Layer 1-2 electrical switches in scale-out, and can support long-distance high-bandwidth interconnection across buildings, sites, and clusters in scale-across scenarios.

• Market sizingTAM and CAGR tracking

  OCS market size forecast

  The report cites Cignal AI forecasts that the OCS market will be about USD400mn in 2025 and may exceed USD2.5bn in 2029, corresponding to a four-year compound growth rate of about 58%.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Zhongji InnoLight / Innolight

  Potential beneficiary; the report assigns a Buy rating

  Strengths

  Has optical communication module capabilities and is advancing a SiPh-based OCS solution through subsidiary TeraHop; listed by the report as an early beneficiary.

  Weaknesses

  Faces risks from high-end optical module demand, competition in 400G/800G, product upgrades, and price wars.

  Comparison

  Compared with vendors that only provide components, InnoLight is extending into OCS solutions and system-related capabilities.

  Risks

  High-end optical module demand falling short of expectations, slower-than-expected product upgrades, and price wars affecting export customers.

• Suzhou TFC Optical Communication

  Potential beneficiary among key optical devices; the report assigns a Buy rating

  Strengths

  As a manufacturer of key optical communication components, it is expected to benefit from demand for upstream OCS components such as MEMS arrays, lenses, and FAU.

  Weaknesses

  Sensitive to downstream optical communication demand and changes in customer strategy.

  Comparison

  More focused on the upstream optical component segment, with a different benefit logic from OCS system vendors.

  Risks

  Optical component or optical module demand below expectations, technology route changes weakening competitiveness, changes in core customer strategies, and geopolitical risks.

• Lumentum

  OCS solution participant, not rated

  Strengths

  One of the main developers of the MEMS route associated with Google and a participant in the OCP OCS subproject.

  Weaknesses

  The MEMS solution still needs to address reliability, cost, and large-scale deployment issues.

  Comparison

  The MEMS route currently has relatively high maturity and a relatively stable supply chain.

  Risks

  If other technology routes lead in cost, reliability, or mass production, its share may be affected.

• Coherent

  OCS solution participant, not rated

  Strengths

  Mainly advancing the DLC digital liquid crystal solution, with potential for low voltage and high reliability.

  Weaknesses

  Liquid crystal solutions face challenges in mass production and yield improvement.

  Comparison

  Compared with MEMS, DLC has advantages in non-mechanical structure and stability, but its commercial maturity still needs validation.

  Risks

  Mass production progress, yield, cost, and customer adoption falling short of expectations.

• NVIDIA

  AI network architecture enabler, not rated

  Strengths

  Discussed OCS for inter-rack and inter-cluster all-optical switching in its next-generation Dragonfly architecture and is exploring CPO+OCS combination solutions.

  Weaknesses

  OCS solutions are still in the architecture advancement and deployment validation stage.

  Comparison

  NVIDIA is a demand-side and architecture-side driver, not a traditional OCS component supply chain company.

  Risks

  Uncertainty in AI cluster architecture iteration, technology route selection, and deployment pace.

Key data

• 2025 OCS market sizeabout USD400mnCignal AI forecast; the market is mainly led by Google.
• 2029 OCS market size forecastover USD2.5bnEquivalent to about 58% CAGR for 2025-2029.
• OCS latency advantageabout 10-100nsNo optical-electrical conversion required, about 1/100 of the latency of electrical switches.
• OCS power consumption advantageabout 1/5 that of an electrical switch with the same bandwidthPower consumption mainly comes from the drive module, with no need for switching chips and matching optical modules.
• OCS impact in Google's research40% lower power consumption, 30% lower cost, 30% higher throughputThe report cites Google's research to illustrate potential benefits in networks such as scale-out.
• Power consumption comparison of network solutionsPluggable 83 pJ/bit; pluggable+OCS 50 pJ/bit; CPO 48 pJ/bit; CPO+OCS 31 pJ/bitThe report cites NVIDIA materials, showing that CPO+OCS reduces power consumption by about 2.6x relative to traditional pluggable solutions.
• Zhongji InnoLight valuationTarget price CNY799.00, based on 35x 2026F EPS CNY22.84Current price CNY779.98, date 15-Apr-2026.
• Suzhou TFC valuationTarget price CNY420.00, based on 50x 2026F EPS CNY8.40Benchmark index is CSI300.

Impact & implications

If OCS adoption accelerates in AI cluster networks, value across the industry chain may expand from traditional electrical switching and pluggable modules to all-optical switching systems, MEMS arrays, lenses, FAU, optical chips, and system integration. Chinese vendors would no longer be merely optical component suppliers, but would begin to enter the competitive landscape for OCS system manufacturing, potentially enhancing their value contribution and strategic position in AI data center network upgrades.

Risks

• High insertion loss in OCS may make it difficult to meet the link budget of existing standard optical modules.
• Insufficient port density and connector performance may fail to match the high-density cabling requirements of AI clusters.
• Reliability, failure modes, redundancy design, and hot-swappable capability still need validation.
• Overall cost remains relatively high, which may limit large-scale deployment.
• If demand for high-end optical modules and AI data centers falls short of expectations, it will affect volume growth across the industry chain.
• Technology route changes, customer strategy adjustments, supply chain diversification, and geopolitical risks may affect the performance of related companies.

What to watch

• Progress in small-batch OCS testing by hyperscalers such as Microsoft, Meta, and NVIDIA.
• Whether NVIDIA's next-generation Feynman architecture and Dragonfly network topology adopt OCS.
• Mass production progress of the four solution types—MEMS, DLC, DLBS, and SiPh—in insertion loss, switching speed, reliability, and cost.
• Customer validation and commercial orders for OCS products from Chinese vendors after OFC 2026.
• Whether the division of labor among OCS, CPO, and pluggable solutions across different AI network layers becomes clear.
• Whether the OCS market grows rapidly from USD400mn to over USD2.5bn as forecast by Cignal AI.