Global optical transceivers Report Interpretation

Overview

Goldman Sachs presents a more constructive global optical-transceiver outlook, arguing that AI data-center build-outs, faster network specifications and increased optical-module usage per AI chip will drive a major upgrade in 2026-28E market forecasts.

Core views

Goldman Sachs raises its global optical-transceiver shipment forecast by 21%/31%/31% for 2026E-28E, its 800G+ shipment forecast by 31%/39%/36%, and its 1.6T+ forecast by 29%/61%/50%. The central thesis is that rising AI infrastructure spending is increasing demand for high-speed connectivity, while the product mix shifts from lower-speed modules toward 800G, 1.6T and eventually 3.2T. The institution forecasts total global optical-transceiver demand of 548m/691m/719m units in 2026E-28E; although lower-speed products remain important for general data centers and telecom, their shipment contribution is expected to fall from 87% in 2026E to 77% in 2028E. The report expects global optical-transceiver value TAM to reach US$68bn in 2026E, US$131bn in 2027E and US$148bn in 2028E. It raises those figures by 33%/81%/115% versus prior estimates, reflecting higher projected rack-level and ASIC AI-server shipments and increased optical-transceiver use per NVIDIA GPU in rack-level systems. The revised model also raises assumed 1.6T usage per GPU in Rubin Ultra to two from zero and 3.2T usage to five from three. It adds Google TPU v9 servers to 2027-28E estimates, assuming 12 1.6T transceivers per GPU in 2H27E-1H28E and six 3.2T transceivers per GPU in 2H28E. The forecasts retain a one-quarter component pull-in to account for the delivery-time gap between optical-transceiver suppliers and ODM final-system shipments. High-speed migration is the report's key mix driver. It expects 800G to be the mainstream large-scale-data-center specification, with shipment volume rising from 45m units in 2026E to 49m in both 2027E and 2028E. 1.6T shipments are forecast to rise from 33m units in 2026E to 71m in 2027E before declining to 55m in 2028E as hyperscalers migrate further to 3.2T. The report expects 3.2T shipments to begin in 2027E at 23m units and increase to 68m in 2028E. The resulting 800G-and-above segment is forecast to reach US$45bn/US$108bn/US$130bn in 2026E-28E, growing at a 69% CAGR. The demand logic is anchored in AI-server networking. NVIDIA GB200 is described as using 400G at the GPU layer and 800G across leaf, spine and core layers, while GB300 uses 800G at the GPU layer and 1.6T at leaf and spine layers. Goldman Sachs expects subsequent Rubin and Rubin Ultra platforms to migrate further toward 1.6T and 3.2T. It forecasts NVIDIA-powered rack-level AI-server shipments of 50,000/92,000/148,000 racks in 2026E-28E, alongside AMD shipments of 5,000/13,000/15,000 units. ASIC AI servers are also expected to expand, with ASIC projected to account for 50%/52%/55% of total AI chips in 2026E-28E. The report highlights Google and Meta as aggressive users of 800G-plus networking, and notes its channel checks indicate Meta will use more optical transceivers per ASIC than per GPU server, where GPU systems typically use two to three transceivers per GPU. Silicon photonics is expected to gain share as speeds rise because of a lower laser-content cost than traditional EML modules. Goldman Sachs estimates that a 1.6T SiPh transceiver uses four 70mW CW lasers with US$15-20 of laser content per module, versus eight 200G EMLs and US$160 of laser content for a 1.6T EML transceiver; this creates different gross-margin implications for module makers. The report expects SiPh modules to represent 10%-18% of the total optical-transceiver market in 2026E-28E and 68%-74% of the 800G+ market, with 60%/80%/80% SiPh adoption in 800G/1.6T/3.2T transceivers. It is Buy rated on Eoptolink and FOCI among transceiver or engine makers, Landmark and VPEC among CW-laser or epiwafer suppliers, and RoboTechnik in equipment.

Analysis framework

The report uses a bottom-up server-TAM model, building general-server and AI-server demand from major server brands, ODM-direct shipments and other vendors. It then translates AI rack, GPU and ASIC-server assumptions, network-speed migration and optical-module attach rates into transceiver shipment and value forecasts, while separately assessing speed, material and format mix.

Methodology notes

• Industry AnalysisSupply-demand framework

  Bottom-up optical-transceiver TAM forecast linked to server shipments, AI-chip demand and module attach rates.

  The report estimates end demand from general and AI servers, then converts expected network configurations and optical modules per chip into shipment and value TAM by speed.

• Industry AnalysisVolume-price decomposition

  TAM is analyzed by shipment volume, speed mix and assumed year-on-year ASP declines for like-for-like products.

  This separates the effect of higher unit demand and migration to higher-value specifications from changes in pricing for comparable products.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  AI-server and chip-platform changes are linked to transceiver makers, laser and epiwafer suppliers, and equipment providers.

  The report traces how AI infrastructure demand and specification upgrades flow through the optical-transceiver supply chain.

Asset mapping & comparison

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

• Eoptolink

  Buy-rated optical-transceiver or engine maker positioned for high-speed optical-module demand.

  Strengths

  Exposure to 800G, 1.6T and 3.2T mix upgrade.

  Comparison

  Grouped with FOCI among optical-transceiver or engine makers.

• FOCI

  Buy-rated optical-transceiver or engine maker positioned for high-speed optical-module demand.

  Strengths

  Exposure to 800G, 1.6T and 3.2T mix upgrade.

  Comparison

  Grouped with Eoptolink among optical-transceiver or engine makers.

• Landmark

  Buy-rated CW-laser or epiwafer supplier linked to rising SiPh adoption.

  Strengths

  Potential beneficiary of increasing SiPh content in high-speed modules.

  Comparison

  Grouped with VPEC among CW-laser or epiwafer suppliers.

• VPEC

  Buy-rated CW-laser or epiwafer supplier linked to rising SiPh adoption.

  Strengths

  Potential beneficiary of increasing SiPh content in high-speed modules.

  Comparison

  Grouped with Landmark among CW-laser or epiwafer suppliers.

• RoboTechnik

  Buy-rated equipment provider within the optical-transceiver supply chain.

  Strengths

  Exposure to the projected expansion in high-speed optical-transceiver production.

  Comparison

  Identified as the equipment exposure in the preferred supply chain.

Key data

• Global optical-transceiver TAMUS$68bn / US$131bn / US$148bn in 2026E / 2027E / 2028ERevised up 33% / 81% / 115% versus prior estimates.
• Global optical-transceiver shipments548m / 691m / 719m units in 2026E / 2027E / 2028EForecast revised up 21% / 31% / 31%.
• 800G+ marketUS$45bn / US$108bn / US$130bn in 2026E / 2027E / 2028EExpected to grow at a 69% CAGR.
• 1.6T shipments33m / 71m / 55m units in 2026E / 2027E / 2028EVolume peaks in 2027E as 3.2T migration begins.
• 3.2T shipments23m / 68m units in 2027E / 2028EAdoption is expected to begin in 2027E.
• SiPh penetration10% / 16% / 18% of total market in 2026E / 2027E / 2028ESiPh is projected to account for 68% / 73% / 74% of the 800G+ market.

Impact & implications

The report argues that the industry’s value growth will be driven less by broad-based low-speed volume and more by AI-led demand for high-speed modules. This favors supply-chain participants exposed to transceiver engines, CW lasers, epiwafers and related equipment, according to Goldman Sachs.