ReportGem ReportGem 中文

REAL-TIME GLOBAL RESEARCH

Innolight (300308.SZ): Optical module mix upgrade and tech migration opportunities to drive growth; revise estimates and TP

Published: 2026-07-17Institution: Goldman SachsPages: 19Original language: EnglishEvidence page: 3

Research evidence excerpt

Innolight (300308.SZ): Optical module mix upgrade and tech migration opportunities to drive growth; revise estimates and TP

Goldman Sachs Innolight (300308.SZ)

Key player in the global optical transceiver market

Innolight is the largest optical interconnect solution provider globally in terms of the

relevant revenue in 2025, per LightCounting and CIC, with a market share of 21.2%. This

is supported by its R&D capabilities, customer trust and collaboration, large scale with

cost advantages, and production & supply chain capabilities. The company is the leading

supplier across all high-speed optical transceiver segments, including 400G, 800G, and

1.6T products.

800G, 1.6T and above optical transceivers as key growth drivers

In 2026/27/28E, we expect the global optical transceiver market to reach

US$51bn/73bn/69bn, with the 800G and above segment increasing at a +31% CAGR

and reaching US$32bn/ 55bn/ 56bn. This is supported by the accelerated specification

migrations in AI data center optical transceivers and the growing number of rack-level AI

servers powered by NVIDIA. We expect the number of racks to increase to 55k / 105k /

163k over 2026-28E, thereby requiring more optical transceivers.

We expect 800G/1.6T shipment volumes to be at 34m/26m units in 2026E, growing to

45m/46m in 2027E, with 3.2T shipments ramping up to 13m in 2027E, as 800G has

become the mainstream specification for large-scale data centers. We expect 1.6T to

see a significant volume ramp-up from 2026 onwards, with adoption of 3.2T starting in

2027E. Migration of AI server networking is ongoing; for example, NVIDIA’s rack-level AI

servers — GB200 uses 400G (GPU layer) and 800G networking (leaf, spine, and core

layers) and GB300 uses 800G (GPU layer) and 1.6T (leaf, spine layer). We expect NVIDIA’s

The English excerpt is extracted automatically from the cited source page and may contain layout or recognition errors. It is never batch translated.

Open report viewer