Optical circuit switches are poised to become a core component of AI networks

Overview

This report focuses on the trend of optical circuit switches (OCS) in global AI networks. OCS routes signals through purely optical paths, reducing the optical-electrical conversion steps in traditional electronic switching, and therefore offers advantages such as low latency, low power consumption, high bandwidth, and topology flexibility. The report believes that as major technology companies such as Google, NVIDIA, Microsoft, and META advance AI cluster network upgrades, the importance of OCS in scale-out, scale-up, and cross-region connectivity scenarios will increase.

Core views

The core view is that OCS is evolving from a traditional optical cross-connect technology used in telecommunications into important infrastructure for AI data center networks. Google's practice in the TPU v4 network shows that OCS can improve large-scale compute resource pooling and network efficiency; NVIDIA's next-generation Dragonfly architecture also mentions that OCS can be used for inter-rack and inter-cluster connectivity. The report believes that OCS, CPO, and pluggable optical modules are not in a simple substitution relationship, but are complementary across different network layers and transmission distances.

Analysis framework

The report analyzes the topic across several dimensions, including technical principles, different OCS technology routes, applicable AI network scenarios, power consumption comparisons, market size forecasts, and supply chain mapping. Technology routes include MEMS, digital liquid crystal, direct light beam steering, and silicon photonics solutions; the industry chain analysis covers upstream optical components and key chips, as well as midstream equipment manufacturers and system integrators.

Methodology notes

• Technology route comparisonComparison of four OCS solution types

  MEMS, DLC, DLBS, and SiPh solutions

  The report compares differences among various OCS solutions in port scale, unit cost, insertion loss, switching time, crosstalk, and reliability to assess maturity and commercialization paths.

• Market sizingOCS TAM forecast

  Market size and CAGR from 2025 to 2029

  Citing Cignal AI forecasts, the report notes that the OCS market is starting with Google as the leading adopter and is expected to exceed USD 2.5 billion by 2029, using this to assess the benefit pool across the industry chain.

• Energy efficiency comparisonPower consumption comparison of network solutions

  Pluggable, Pluggable+OCS, CPO, CPO+OCS

  The report compares per-bit power consumption across different AI network interconnect solutions, emphasizing that the combination of OCS and CPO offers lower energy consumption.

Asset mapping & comparison

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

• Optical circuit switches (OCS)

  Core technology theme of the report

  Strengths

  Low latency, low power consumption, and high bandwidth, supporting dynamic optical paths and AI compute resource pooling.

  Weaknesses

  Link budget, insertion loss, port density, reliability, and total cost still need improvement.

  Comparison

  Compared with traditional electronic switches, it reduces optical-electrical conversion; compared with CPO and pluggable solutions, it is more oriented toward topology reconfiguration and all-optical core-layer connectivity.

  Risks

  Technology standards, cost reduction, and the pace of large-scale deployment may fall short of expectations.

• Lumentum (US.LITE)

  MEMS technology and OCS solution supplier

  Strengths

  The report says it launched a high-density 300x300 array optical switch at OFC and disclosed strong order and revenue outlook.

  Weaknesses

  Commercial deliveries are concentrated in the future, and there remains uncertainty around execution and capacity ramp-up.

  Comparison

  Unlike Coherent's DLC route, Lumentum mainly represents the MEMS route.

  Risks

  Customer adoption pace, competitive dynamics, and risks from shifts in technology routes.

• Coherent (US.COHR)

  Supplier of digital liquid crystal OCS solutions

  Strengths

  The DLC solution has advantages in low voltage, reliability, and service life.

  Weaknesses

  The report mentions that optical path switching time is usually around several hundred milliseconds, and mass production and yield may also be challenging.

  Comparison

  Compared with MEMS, DLC has no mechanical parts; compared with silicon photonics solutions, it involves different trade-offs in maturity and performance.

  Risks

  Switching speed, scaled production, and progress in customer validation.

• TFC Communication (300394.SS)

  Potential beneficiary of key OCS components

  Strengths

  The report believes that as a supplier of key optical communication components, it is likely to benefit from growing demand for MEMS arrays, lenses, and fiber arrays.

  Weaknesses

  Sensitive to demand conditions for optical components and modules.

  Comparison

  Compared with full equipment vendors, TFC Communication is more positioned in the key component supply chain segment.

  Risks

  Optical component/module demand may be weaker than expected, and changes in optical communication technology may negatively affect the business.

• Corning (US.GLW)

  Supply chain participant for fiber collimator arrays and related components

  Strengths

  The report lists Corning as a relevant optical component supplier in the supply chain chart.

  Weaknesses

  The report does not provide specific financial sensitivity for Corning's OCS business.

  Comparison

  More upstream in materials and optical components rather than full OCS equipment.

  Risks

  Slow conversion of OCS demand or pricing pressure on components.

• Microsoft (US.MSFT), Google-A (US.GOOGL), NVIDIA, META

  Potential OCS adopters and drivers of AI cluster networks

  Strengths

  Large AI cluster buildouts can drive demand for higher bandwidth, lower power consumption, and more flexible network architectures.

  Weaknesses

  There are differences in architectural approaches and in-house development pace among cloud vendors.

  Comparison

  Google has already applied OCS in its TPU network, while other large technology companies are still at the stage of interest or small-scale testing.

  Risks

  Changes in capital expenditure pace, architecture choices, and technology validation progress.

Key data

• OCS market sizeAbout USD 400 million in 2025; expected to exceed USD 2.5 billion in 2029Source is Cignal AI forecasts; the report states a four-year CAGR of about 58%.
• OCS latency10 to 100 nanosecondsThe report says this is about one-hundredth of the latency of electronic switches.
• OCS power consumption advantageAbout 1/5 of electrical switches with the same bandwidthBased on the report's description of the all-optical switching path.
• Effect of Google's MEMS Apollo platform30% cost reduction, 40% lower energy consumptionThe report cites Google's description of the MEMS optical chip platform.
• Scale-out network performance40% lower energy consumption, 30% cost reduction, 30% higher throughputThe report cites Google's research findings.
• Network solution power consumptionPluggable 83 pJ/bit; Pluggable+OCS 50 pJ/bit; CPO 48 pJ/bit; CPO+OCS 31 pJ/bitThe report cites data from NVIDIA and Nomura Research.
• Lumentum commercialization progressTotal orders exceed USD 4 billion; annual revenue expected to exceed USD 1 billion in 2027The report says most product deliveries are scheduled for 2H26, with shipment CAGR above 150% from 2025 to 2028.
• TFC Communication rating and target priceBuy; target price RMB 420.00; current price RMB 330.66Current price as of April 15, 2026; target price is based on expected 2026 EPS of RMB 8.40 and a 50x P/E multiple.

Impact & implications

If OCS adoption accelerates in AI data centers, industry chain value may expand from standalone optical modules toward all-optical switching equipment, optical components, MEMS arrays, lens arrays, fiber collimator arrays, and system integration. From an investment perspective, beneficiaries include vendors with full OCS solution capabilities as well as companies supplying key optical components to OCS; however, commercialization still depends on link budget, port density, reliability, and cost reduction.

Risks

• OCS link budget and insertion loss may fail to meet the requirements of standard optical transceivers.
• Port density, connector density, and cabling capability may be insufficient to match the needs of large-scale AI clusters.
• Failure modes, redundancy design, and hot-swap capability are unclear, which may limit deployment in high-availability scenarios.
• Overall cost may be too high, affecting large-scale adoption in AI data centers.
• Demand for optical modules and optical components may come in below expectations, or competition in areas such as 400G and 800G may intensify.
• Product upgrade delays, price wars, and pressure on export business may affect the profitability of related suppliers.

What to watch

• Validation and procurement progress of OCS by major technology companies such as Google, NVIDIA, Microsoft, and META.
• Whether NVIDIA Dragonfly and Feynman-related network architectures actually adopt OCS.
• Changes in port scale, switching speed, insertion loss, and mass-production yield across the four technology routes of MEMS, DLC, DLBS, and SiPh.
• Product launches and order execution by companies such as Lumentum, Coherent, iPronics, Taclink, InnoLight, Eoptolink, and Accelink.
• Prototype, order, capacity, and customer validation progress of Chinese manufacturers at OFC and in subsequent commercialization.
• Actual power consumption and cost performance of CPO+OCS solutions in AI data centers.