AI optical interconnect supply chain: Expert call sees OCS demand accelerating through 2027 alongside expanding 800G/1.6T optical-transceiver demand

Overview

This Nomura expert-call note examines the developing AI optical-interconnect market, covering OCS demand and technology choices, OCS component costs, 800G and 1.6T transceiver demand and pricing, and the expected adoption path for NPO, CPO, and VCSEL solutions.

Core views

The expert characterizes OCS as an increasingly important architecture for large AI clusters because of its potential total-cost-of-ownership, latency, and failure-rate advantages. Google currently uses OCS chiefly for scale-out at the spine layer and for rack-to-rack scale-up interconnection in its 3D Torus architecture; other cloud service providers use it for rack-to-rack connections. The technology could also extend to scale-across networks. Google's OCS demand was about 12-15k units in 2025, is estimated at 18-20k in 2026E, and could reach 35-40k in 2027F. Other CSPs remain at the sample stage in 2026, but major CSPs may each demand 1-2k units in 2027, bringing total OCS demand to roughly 40-50k units. MEMS is expected to remain the leading OCS technology near term, with an estimated 80-85% market share in 2026F, falling to 60-70% in 2027E as liquid-crystal and piezoelectric beam-steering alternatives advance. Over the long term, the expert expects MEMS at 40-50%, liquid crystal at 20-30%, and piezoelectric beam steering and silicon-photonics OCS accounting for the remainder. MEMS offers short switching times and fast assembly that may support mass production, but mechanical movement in the array and use of 220V high voltage may reduce lifespan. Liquid-crystal OCS operates at low voltage and may offer longer life, although reliability remains unresolved. Piezoelectric OCS, developed mainly by Huber+Suhner, is expected to enter small-batch validation shipments in 4Q26, while silicon-photonics OCS faces relatively high insertion loss and requires more time to commercialize. The call also outlines the OCS supply chain and economics. Google and Lumentum are developing MEMS OCS, while Accelink, Eoptolink, Innolight, O-Net, and Huawei are described as working mainly on the technology. Google designs its own MEMS chips and uses Silex for foundry services; other domestic manufacturers mainly procure commercial MEMS chips from Preciseley. Advanced Fiber Resources, Focuslight Technologies, and Optowide Technology can supply fiber-array and lens-array components. For a 384-port MEMS OCS, fiber arrays cost USD2,500-3,000 each, lens arrays USD1,700-1,800 each, and two MEMS chips USD17,000-18,000 each; mirrors cost USD200, the calibration system USD5,000, and assembly about USD15,000, resulting in a total OCS price of about USD100-120k. A 384-port liquid-crystal OCS uses two liquid-crystal panels priced at USD6,000-8,000 each, four collimator arrays at USD3,000 each, and four deflection splitters and crystal wedges totaling USD14,000. For optical transceivers, the expert estimates North American 800G shipments of 45-50mn units in 2026, with DR representing about 70% and FR 30%. DR is split roughly evenly between silicon-photonics and EML solutions, while FR is almost entirely EML. Estimated 1.6T demand exceeds 30mn units in 2026, but material shortages could reduce actual shipments to 10-15mn units. A shortage of 3nm DSPs for FR versions may result in DR accounting for 70% of 1.6T transceivers in 2026; 1.6T DR is estimated to use 70% silicon photonics and 30% EML. For 2027, the expert cites 800G DR pricing of about USD350 per unit for EML and USD300 for silicon photonics, versus about USD400 for FR. For 1.6T, cited prices are USD900-950 for EML DR, more than USD1,000 for FR, and USD800-850 for silicon photonics, with annual price declines below 10%. The next lane-speed upgrade faces component constraints. Although 400G InP laser chips have already been released, Serdes remains limited to 200G and is identified as a key bottleneck. Silicon modulators are capped at 200G, so the expert expects modulator upgrades to require thin-film lithium niobate. For packaging architectures, NPO may see small-volume shipments in 2027F and mass production in 2028F, while CPO shipments are expected at 80-100k units in 2027F, mainly for scale-out networks. VCSELs, which use GaAs and suit 30-40m transmission distances, may fit scale-up networks because they are less temperature-sensitive; however, their required 850nm or 1060nm wavelengths would require custom optical fibers because current lasers are mainly at 1310nm.

Analysis framework

Nomura summarizes an expert discussion by tracing AI-cluster networking requirements through OCS deployment demand, competing technology paths, component supply chains and bill-of-material costs, then assessing high-speed transceiver volumes, technology mix, prices, and packaging-roadmap milestones.

Methodology notes

• Industry AnalysisSupply-demand framework

  Supply-demand analysis of OCS and optical-transceiver markets

  The report estimates OCS unit demand by major cloud-service providers and compares transceiver demand with shipment capacity constrained by material shortages.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  Optical-interconnect supply-chain mapping

  The call links AI-cluster adoption to OCS technologies, chips, foundry services, fiber and lens components, transceiver technologies, and packaging solutions.

• Industry AnalysisCost curve analysis

  OCS bill-of-material cost breakdown

  The report uses component-level prices for MEMS and liquid-crystal OCS to explain the approximate cost structure of a 384-port system.

Asset mapping & comparison

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

• Google (GOOGL US)

  Current major OCS user and developer of MEMS OCS technology

  Strengths

  Uses OCS for spine-layer scale-out and rack-to-rack scale-up interconnection; designs its own MEMS chips.

  Comparison

  Its OCS demand is materially larger than other CSPs, which remain at the sample stage in 2026.

• NVIDIA (NVDA US)

  Potential future adopter of OCS together with CPO/NPO solutions for scale-up interconnection

• Lumentum (LITE US)

  Developer of MEMS-based OCS

  Strengths

  Developing a leading near-term OCS technology path.

  Weaknesses

  MEMS technology may face lifespan limitations associated with mechanical movement and high voltage.

  Comparison

  Competes with liquid-crystal, piezoelectric beam-steering, and silicon-photonics OCS approaches.

• Coherent (COHR US)

  Leading producer of liquid-crystal panels used in liquid-crystal OCS

  Strengths

  Identified as the leading player able to produce the core liquid-crystal panel.

  Weaknesses

  Liquid-crystal OCS reliability remains an issue to address.

  Comparison

  Liquid-crystal OCS offers lower-voltage operation and potentially longer life than MEMS OCS.

  Risks

  Reliability of liquid-crystal OCS remains unresolved.

• Huber+Suhner (HUBN SW)

  Developer of piezoelectric OCS

  Strengths

  Expected to make small-batch shipments for validation in 4Q26.

  Weaknesses

  Technology remains in validation-stage shipments.

  Comparison

  An alternative to MEMS and liquid-crystal OCS.

Key data

• Google OCS demand12-15k units in 2025; 18-20k in 2026E; 35-40k in 2027FExpert estimate
• Total OCS demand~40-50k units in 2027EIncludes Google and estimated 1-2k units for each other major CSP
• MEMS OCS market share80-85% in 2026F; 60-70% in 2027EExpected to decline as alternative OCS technologies develop
• 384-port MEMS OCS priceUSD100-120kExpert bill-of-material estimate
• North American 800G transceiver shipments45-50mn units in 2026DR accounts for about 70% and FR about 30%
• 1.6T transceiver demand and shipmentsDemand above 30mn units; actual shipments 10-15mn units in 2026Material shortages may constrain shipments
• 2027 1.6T transceiver pricesEML DR USD900-950; FR USD1,000+; silicon-photonics USD800-850Annual price decline expected to be below 10%
• CPO and NPO rolloutCPO 80-100k units in 2027F; NPO small-volume shipments in 2027F and mass production in 2028FCPO demand is expected mainly in scale-out networks

Impact & implications

The report indicates that AI-network scaling could broaden OCS adoption beyond Google's current deployments, while technology competition may gradually reduce MEMS dominance. It also highlights that transceiver growth is supported by AI infrastructure demand but remains exposed to component shortages and lane-speed bottlenecks, with CPO and NPO still following a staged commercialization path.

Risks

• Material shortages may limit 2026 1.6T transceiver shipments to 10-15mn units despite demand estimated above 30mn.
• A shortage of 3nm DSPs for FR versions may shift the 1.6T technology mix toward DR.
• MEMS OCS lifespan may be lower than alternatives because of mechanical array movement and 220V operation.
• Liquid-crystal OCS reliability remains a technical issue.
• Silicon-photonics OCS has relatively high insertion loss and may require more time for commercialization.
• Serdes limitations at 200G and silicon-modulator limits may constrain the transition to 400G-per-lane architectures.
• VCSEL deployment at required wavelengths would require custom optical fibers.