ReportGem ReportGem 中文

GLOBAL RESEARCH ARCHIVE

China Technology AI optical interconnect series 1: How many optical transceivers do we need?

Published: 2026-05-14Institution: HSBC Global Investment ResearchCompany / ticker: 002475.SZPages: 15Original language: 英语Evidence page: 1

Research evidence excerpt

China Technology AI optical interconnect series 1: How many optical transceivers do we need?

Equity Research Report | 14 May 2026

China Technology EquitiesElectronic Equipment &

Instruments

AI optical interconnect series 1: How many optical

transceivers do we need? China

◆ Optical interconnect TAM: USD34bn/58bn in 2026/27e (c5% Bingyi Zheng* (Reg. No. S1700521060001)

of AI capex), rising to USD120bn by 2030e (c8%) Analyst, A-share Technology Hardware HSBC Qianhai Securities Limited

bingyi.zheng@hsbcqh.com.cn

◆ The GPU-to-transceiver attach rate is the key demand driver: +86 21 5066 2028

c4x on a blended basis in 2026-27e, rising to 6-8x by 2030e Steven Sun*, CFA (Reg. No. S1700517110003)

Head of Research, HSBC Qianhai Securities Limited

HSBC Qianhai Securities Limited

◆ Prefer module assemblers for scale and profitability and stevensun@hsbcqh.com.cn

domestic laser chip suppliers for margin expansion +86 755 8898 3158

Yongzhu Wang* (Reg. No. S1700125070003)

Associate

Three self-reinforcing forces drive demand growth. We see the optical transceiver Shenzhen

market entering a multi-year super-cycle, driven by the convergence of three forces

that are compounding each other. (1) AI infrastructure capex is accelerating at an * Employed by a non-US affiliate of HSBC Securities (USA) Inc, and is

not registered/ qualified pursuant to FINRA regulations

unprecedented pace, with five major US CSPs guiding to combined 2026e capex of

cUSD750bn, almost double that of 2025 levels, providing an expanding funding base

for transceiver procurement. (2) We expect the number of optical transceivers

required per GPU to rise, with attach rates evolving from 1:4 in 2026-27e to 1:6-8 by

2030e as cluster size scales toward 500,000+ GPUs. (3) Exponential growth in AI

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