Goldman raised 800G and above optical module demand forecasts, with AI data centers remaining the key growth driver
AI summary card
Goldman raised 800G and above optical module demand forecasts, with AI data centers remaining the key growth driver
Goldman expects the global optical transceiver market in 2026-2028E to reach US$51bn/73bn/69bn, with the 800G and above segment growing at around 31% CAGR, mainly driven by AI server network upgrades, increasing ASIC penetration, and greater adoption of silicon photonics.
- The report expects global optical transceiver market to reach US$51bn/73bn/69bn in 2026/27/28E, and the 800G and above segment to reach US$32bn/55bn/56bn.
- The report expects 2026E 800G/1.6T shipments of 34m/26m units, rising to 45m/46m units in 2027E. 3.2T is expected to ramp to 13m units in 2027E.
- AI data center upgrading is driving migration to higher-speed specifications. 800G has become the mainstream specification for hyperscale data centers, 1.6T has begun to ramp strongly from 2026, and 3.2T is expected to be adopted starting in 2027.
- The expansion of ASIC AI servers will increase dependence on networking capability, and the report expects ASIC share of AI chips to rise from 43% in 2026E to 55% in 2028E.
- Silicon photonics solutions are expected to accelerate penetration due to cost advantages in high-speed products, reaching 10%-18% of overall market TAM and about 69%-77% of the 800G+ market in 2026-2028E.
Report interpretation
Overview
This report focuses on the global optical transceiver market, especially the demand upgrade for 800G, 1.6T, and 3.2T high-speed optical modules in AI data centers. Goldman believes the expansion of AI server infrastructure, rack-level AI server upgrades by NVIDIA, growth in ASIC AI servers from Google and others, and the substitution of copper cables with fiber will continue to lift high-speed optical transceiver TAM and shipments.
Core views
The key view is that demand for 800G and above optical transceivers will remain in high growth during 2026-2028E. The report raised the 2026-2028E global optical transceiver value TAM by 18%/35%/40% and raised 800G and above transceiver shipment forecasts by 14%/29%/33%. Drivers include higher NVIDIA rack-level AI server and Google ASIC server shipments, higher per-GPU 1.6T+ optical transceiver usage, and platform migration on GB300, Rubin, Rubin Ultra, and Feynman.
Analysis framework
The report uses a combined approach of top-down AI server TAM and bottom-up optical module specification migration: first estimating network connection demand based on AI servers, GPU/ASIC chips, and rack counts, then splitting optical transceiver value TAM and shipments by 800G, 1.6T, 3.2T specifications, ASP, and material technology stack.
Methodology notes
Estimate optical transceiver TAM by speed spec, shipment volume, and ASP
The report splits the global optical transceiver market into Below 400G, 400G, 800G, 1.6T, and 3.2T categories, and estimates 2026-2028E value TAM using annual shipment and ASP forecasts.
Derive high-speed optical module demand from AI server architecture upgrades
Using NVIDIA GB200, GB300, Rubin, and Rubin Ultra rack-level servers as examples, the report analyzes usage growth from migration of GPU-to-leaf-to-spine-to-core networks from 400G to 800G, 1.6T, and 3.2T.
Cost advantage of silicon photonics relative to EML in high-speed optical modules
The report argues that 1.6T silicon photonics transceivers using CW lasers have significantly lower laser-content costs than 1.6T EML solutions, so silicon-photonics penetration rises with higher speeds.
Goldman research frameworks for stock attributes, takeover probability, and financial database
The appendix details Goldman’s research frameworks for growth, cash return, valuation, composite percentiles, M&A ranking, and Quantum database; the main body of this report mainly uses TAM, server, and supply-chain demand frameworks.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- InnolightOptical transceiver / optical engine supply-chain beneficiary
- Strengths
- Listed as Buy in the report, benefiting from demand growth of 800G and above optical modules.
- Weaknesses
- Declining ASPs over time in high-speed products may compress price contribution.
- Comparison
- Along with Eoptolink and TFC Optical, it is an optical transceiver or optical engine manufacturer.
- Risks
- AI server shipments, client capex, or specification migration are below expectations.
- EoptolinkOptical transceiver supply-chain beneficiary
- Strengths
- Listed as Buy in the report, target price RMB496.98.
- Weaknesses
- Demand is highly dependent on the pace of AI data center upgrades.
- Comparison
- Shares exposure to high-speed optical module demand with Innolight and TFC Optical.
- Risks
- Price declines, intensified competition, and customer order volatility.
- TFC OpticalOptical transceiver / optical engine supply-chain beneficiary
- Strengths
- Listed as Buy in the report, target price Rmb330.66.
- Weaknesses
- Continuous technology investment required due to high-speed specification transition.
- Comparison
- Similarly benefits from upgrades of 800G, 1.6T, and 3.2T.
- Risks
- Changes in technology route or slower-than-expected silicon-photonics penetration.
- LandmarkCW laser or epitaxy wafer supply-chain beneficiary
- Strengths
- Listed as Buy in the report, target price NT$2,315.00, benefiting from adoption of silicon-photonics solutions.
- Weaknesses
- Demand depends on the penetration of silicon-photonics routes in high-speed optical modules.
- Comparison
- Shares upstream laser/epitaxy-chain exposure with VPEC.
- Risks
- Silicon-photonics cost advantage not being realized or customer qualification timelines being delayed.
- VPECCW laser or epitaxy wafer supply-chain beneficiary
- Strengths
- Listed as Buy in the report, target price NT$275.00.
- Weaknesses
- Upstream component value may be affected by pricing competition.
- Comparison
- Like Landmark, an early beneficiary of 800G/1.6T upgrades.
- Risks
- ASP declines, order concentration, and technology substitution risks.
- Ruijie NetworksData center switch beneficiary
- Strengths
- Listed as Buy in the report, target price RMB93.52, benefiting from AI data center network upgrades.
- Weaknesses
- Demand is linked to data center networking capex.
- Comparison
- Compared with optical module makers, it is exposed to demand on the switching equipment side.
- Risks
- Data center switch TAM growth below expectations or intensified competition.
- RoboTechnikEquipment-side beneficiary
- Strengths
- Listed as Buy in the report, target price RMB490.76.
- Weaknesses
- Equipment demand may lag and be more volatile than component demand.
- Comparison
- Compared with optical module and upstream materials names, it is more exposure to equipment-stage demand.
- Risks
- Customer expansion pace, order confirmation, and capex volatility.
Key data
- Global optical transceiver market size2026E/2027E/2028E: US$51bn/73bn/69bnReport forecast.
- 800G and above segment size2026E/2027E/2028E: US$32bn/55bn/56bnCorresponding to about 31% CAGR.
- 800G/1.6T shipments2026E: 34m/26m units; 2027E: 45m/46m units3.2T is expected to ramp to 13m units in 2027E.
- Global optical transceiver total demand2026E/2027E/2028E: 452m/528m/556m unitsThe 800G+ premium segment is mainly driven by AI data centers.
- 800G+ demand growth2026-2028E CAGR around 45%, reaching 126m units in 2028EAI data center applications are the main driver.
- Share of shipments at 400G and below87% in 2026E, down to 77% by 2028EReflects migration to higher-speed specifications.
- ASIC share of AI chips2026E/2027E/2028E: 43%/50%/55%ASIC servers are more dependent on networking capability, supporting optical transceiver demand.
- Silicon photonics penetration10%-18% of the overall market in 2026-2028E; 69%-77% in the 800G+ marketThe cost advantage of silicon photonics is more pronounced in higher-speed products.
- Raised global optical transceiver value TAMRaised by 18%/35%/40% in 2026E/2027E/2028EPrimarily due to upward revisions in NVIDIA rack-level AI server, Google ASIC server, and 1.6T+ per-GPU usage forecasts.
Impact & implications
For investors, the report reinforces that expansion of AI compute infrastructure is not only supportive of GPU/ASIC but will also spill over to high-speed networking, optical transceivers, silicon photonics, CW lasers, epitaxy wafers, data center switches, and related equipment vendors. The migration to higher-speed specifications and rising silicon-photonics penetration may reshape supply-chain profit allocation, while a lower share of low-speed products implies relatively weaker growth momentum for traditional 400G and below products.
Risks
- AI server shipments or cloud-provider capex below expectations.
- The specification migration pace of major customers such as NVIDIA, Google, and Meta is slower than assumed.
- Decline in ASPs of 800G, 1.6T, and 3.2T optical modules exceeds expectations.
- Silicon-photonics penetration rises slower than expected, or traditional solutions such as EML retain competitiveness.
- Growth in ASIC AI servers is below expectations, weakening incremental demand for high-speed networking links.
- Supply-chain competition, yield, capex expansion, and customer concentration may affect profitability.
What to watch
- Whether 2026E 1.6T optical modules ramp strongly as expected.
- Whether 2027E 3.2T optical module adoption is delivered to the 13m-unit level.
- NVIDIA GB300, Rubin, Rubin Ultra, and Feynman rack-level server shipments and per-GPU optical transceiver usage.
- Adoption speed of 800G+ connectivity by ASIC users including Google and Meta.
- Whether silicon-photonics penetration in the 800G+ market reaches the 69%-77% range.
- Changes in high-speed optical module ASPs and supply-chain gross margins.