Asia Pacific CPO supply chain update: GlassBridge has potential but mass production is still far away, and CPO enablers remain favored

Overview

This report focuses on the Asia Pacific CPO supply chain, the differences between GlassBridge and traditional V-Groove FAU, and the revenue, valuation, and risk-return of key supply-chain companies such as FOCI and AllRing. The core conclusion is that GlassBridge has potential advantages including reworkability, high density, passive alignment, and wafer-level manufacturing, but has not yet changed TSMC’s mainstream COUPE path in the short term. The CPO ecosystem still requires wafer foundries, chip designers, FAU suppliers, lasers, and system integrators to validate together, and design migration cannot be completed overnight.

Core views

The report is constructive on the long-term development of CPO, arguing that current investor expectations for CPO stocks are low. In the short to medium term, TSMC mainstream COUPE platform programs and key customers such as NVIDIA, AMD, and Ayar Labs are still more likely to use grating coupling rather than shift to GlassBridge immediately. On the supply side, TSMC PIC capacity expansion and improvements in insertion-test efficiency are key prerequisites for CPO mass production; on the demand side, the NVIDIA Spectrum CPO switch, future Kyber/Rubin Ultra architectures, and AMD MI500/MI550/MI650 projects constitute the main incremental clues.

Analysis framework

The report adopts a combination of supply-chain fieldwork, capacity planning, shipment assumptions, company-level revenue decomposition, valuation modeling, and risk-return scenario analysis, and evaluates commercialization timing for CPO photonic engines, FAU, advanced packaging equipment, and silicon-photonics-related companies separately.

Methodology notes

• supply chain validationSupply chain checks

  Use supply-chain research to assess customer programs, capacity ramp, test efficiency, and mass-production timing.

  The report repeatedly uses supply-chain checks to evaluate TSMC PIC capacity, customer programs at NVIDIA/AMD/Broadcom, the number of FOCI customers, and AllRing's progress in the CPO equipment segment.

• bottom-up analysisBottom-up analysis

  Break down revenue contribution by customer, product, and capacity.

  For FOCI and AllRing, the report separately decomposes 2026-2028 revenue by CPO switch, FAU share, CoWoS, SoIC, and CPO equipment.

• valuation modelResidual income model

  Estimate AllRing’s base-case target price using a residual-income model.

  The target price of NT$1,580 for AllRing is based on an RI model, with key assumptions including 8% cost of equity, 2.0% risk-free rate, 1.0 beta, 6.0% equity risk premium, 16% mid-term growth rate, and 4.5% terminal growth rate.

• risk-rewardBull/Base/Bear case

  Assess valuation range through bull, base, and bear case scenarios.

  The report sets different revenue CAGR, gross margin, market share, and technology-transition assumptions for AllRing and FOCI to evaluate upside and downside risks.

Asset mapping & comparison

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

• AllRing Tech Co. (6187.TWO)

  CPO and advanced packaging equipment beneficiary

  Strengths

  Benefits from multi-year expansion in CoWoS, CPO, SoIC, and CoPoS; is the sole supplier of the SoIC WoW dispenser; CPO equipment gross margin could reach 55-60%.

  Weaknesses

  SoIC equipment pull-in may be slightly delayed by the pace of TSMC AP7 CoWoS capacity expansion, and revenues are sensitive to TSMC and advanced-packaging capex cycles.

  Comparison

  The report believes implied 2027E P/E is around 40x, lower than CPO peers and close to SoIC peers.

  Risks

  COWoS expansion slows, WoS market share declines, and SoIC and silicon-photonics adoption decelerates.

• FOCI Fiber Optic Communications Inc (3363.TWO)

  FAU and CPO mass-production supply-chain beneficiary

  Strengths

  Expected CPO mass-production revenue begins in July, and continues to scale in 2027; assumed FAU share in NVIDIA Spectrum-X CPO switch is around 40%; more than 10 customers are in prototype validation.

  Weaknesses

  In 2026, profitability is affected by capacity migration from Hsinchu to the new Thailand fab, SiPh/CPO readiness for a new production line, destructive testing, and dilution costs, resulting in EPS being revised to a loss.

  Comparison

  Compared with the traditional business, CPO/SiPh has higher growth potential, but the report has not fully incorporated an upside scenario where over 10 customers ramp to mass production.

  Risks

  Customer project delays, downgrades in traditional-business revenue, slower-than-expected ramp of mass production, and weaker-than-expected gross-margin recovery.

• TFC Communication

  High-end FAU supplier

  Strengths

  The report believes TFC has a competitive advantage in high-end FAU and is less likely to be immediately impacted by early-stage GlassBridge substitution.

  Weaknesses

  If GlassBridge expands its application over the long term, FAU TAM may still be affected.

  Comparison

  Relative to lower-end FAU suppliers, TFC is more likely to remain resilient because of its higher-end positioning.

  Risks

  Accelerated penetration of GlassBridge or other interconnect architectures could reduce demand for traditional FAU.

• Corning GlassBridge

  New glass-waveguide interconnect architecture

  Strengths

  It has wafer-level manufacturing, customizable pitch, detachable connections, passive alignment, and high-density system-integration potential.

  Weaknesses

  It currently mainly supports one-dimensional fiber layout, has not been adopted by TSMC COUPE projects, remains far from large-scale mass production, and may not offer lower insertion loss than traditional FAU.

  Comparison

  The difference from traditional FAU is not 'new FAU replacing old FAU,' but an architectural difference between connectorized glass waveguide bridges and direct fiber-array attach.

  Risks

  Mass-production feasibility, ecosystem onboarding, customer design lock-in, and loss-performance may be worse than expected.

• TSMC

  Core platform for CPO/PIC and advanced packaging capacity

  Strengths

  PIC capacity plans rise from 10kwpm to 15kwpm and reach at least 25kwpm by 2028; CoWoS capacity expansion supports equipment suppliers such as AllRing.

  Weaknesses

  Limited PIC resources mean mass-production customers in 2026-2027 may be concentrated in NVIDIA, Broadcom, and AMD.

  Comparison

  The TSMC COUPE platform remains the core path for CPO mass production in the near term.

  Risks

  PIC, CoWoS, or SoIC capacity ramps may fall short of expectations, impacting downstream revenue recognition.

• AMD

  Potential CPO customer and AI/HPC platform participant

  Strengths

  The report expects AMD MI500 series FAU shipments may start from the second half of 2027, and sees AMD exploring scale-up CPO versions for MI550 or MI650 rack systems.

  Weaknesses

  Related projects still require development progress and mass-production signal confirmation.

  Comparison

  Compared with NVIDIA, AMD’s contribution profile to CPO in the report is more weighted toward 2027/2028.

  Risks

  Delays in platform entry or changes in adoption approach.

Key data

• CPO switch shipment forecast23k units in 2026, 59k units in 2027, 200k units in 2030The report states the implied CAGR for 2024-2030 is 144%.
• TSMC PIC capacity plan10kwpm in 2026 Q2, 15kwpm in 2026 Q4, at least 25kwpm in 2028The report states that in 2026-2027, key COUPE mass-production customers may be NVIDIA, Broadcom, and AMD.
• Insertion 2 EPIC test efficiencyImproved from one wafer per day in H2 2025 to around six hours per waferThe target for the next 6-12 months is 3-4 hours per wafer.
• FOCI FAU share assumption in NVIDIA Spectrum-X CPO switchabout 40%The report expects NVIDIA to contribute about 18% of FOCI total revenue in 2026.
• FOCI earnings estimate adjustment2026E EPS adjusted to NT$(0.41), 2027E EPS NT$16.36, 2028E EPS NT$47.682026 is expected to turn to a loss, mainly due to 1Q26/2Q26 performance, capacity migration, SiPh/CPO readiness, and dilution from secondary issuance.
• AllRing CPO revenue share11% in 2026, 19% in 2027, 26% in 2028The report expects AllRing has entered NVIDIA CPO supply-chain segments in FAU-photonic engine coupling and FAU AOI equipment.
• AllRing revenue forecastNT$9,405mn in 2026E, NT$15,605mn in 2027E, NT$17,305mn in 2028EUnderlying drivers include CoWoS, CPO, SoIC, and other advanced packaging opportunities.
• AllRing target priceNT$1,580Based on the residual-income model, key assumptions include 8% CoE, 16% mid-term growth rate, and 4.5% terminal growth rate.
• FOCI target priceNT$708The risk-reward section of the report shows the target price as NT$708.

Impact & implications

The investment implication is that the long-term direction of the CPO ecosystem remains intact, but pace depends on TSMC PIC capacity, insertion-test efficiency, volume-production schedules for customers such as NVIDIA/AMD/Broadcom, and suppliers’ share in FAU, photonic-engine coupling, AOI, dispensing, and advanced packaging equipment. GlassBridge may exert long-term substitution pressure on lower-end FAU, but in the short term is more likely a complementary architecture in high-density system integration and unlikely to immediately disrupt mainstream CPO design.

Risks

• If GlassBridge matures faster, it could long-term compress FAU TAM, especially in lower-end FAU applications.
• If TSMC PIC, CoWoS, or SoIC capacity expansion is slower than expected, CPO mass production and equipment revenue would be affected.
• If CPO testing and packaging flow steps such as Insertion 2/3 do not continue to improve in efficiency, mass production may be delayed.
• If customer projects at NVIDIA, AMD, Broadcom, MediaTek, Marvell, Ayar Labs, etc. are delayed or design changes occur, supply-chain revenue realization will be below expectations.
• FOCI’s capacity migration, SiPh/CPO new-line readiness, destructive testing, and dilution from secondary issuance may continue to pressure profitability.
• AllRing may face risks from slower CoWoS expansion, declines in WoS market share, and slower advanced-packaging technology migration.
• Morgan Stanley has potential business relationships with covered companies, and investors should treat this research as only one of several decision inputs.

What to watch

• Whether TSMC PIC capacity ramps along the 10kwpm, 15kwpm, and 25kwpm path.
• Whether Insertion 2 EPIC wafer test time can be reduced to 3-4 hours within the next 6-12 months.
• Mass-production schedules for NVIDIA Spectrum-X, Kyber/Rubin Ultra CPO racks, and AMD MI500/MI550/MI650.
• FOCI’s CPO mass-production revenue after July, NRE revenue, and conversion signals from over 10 prototype customers into mass production.
• Order and revenue disclosures from AllRing in CPO FAU-photonic engine coupling, AOI, and dispensing steps.
• Whether GlassBridge is adopted by TSMC COUPE or key customer programs, and mass-production validation results in high-density CPO/NPO modules.
• The ramp pace of advanced packaging capacity expansion such as CoWoS, SoIC, CoPoS, and FoPLP.