AI Full Industry Chain In-Depth Practical Research Report | Q2 2026

Research Date: May 9, 2026 [Taogu Ba]
Data Sources: Morgan Stanley 2026.5.7 “AI Profit Cycle”, Goldman Sachs 2026.5.7 “Agentic AI”, A-share institutional fund flows, Q1 2026 financial reports, Northbound / Public fund holdings data
Research Conclusion: Based on global investment bank standards and A-share capital behavior, forming an objective positioning and target rating for the entire industry chain; does not constitute investment advice

1. Core Overall Judgment
Market capital allocation focus is concentrated on the AI hardware chain; electricity and green power are supporting elements for computing power, not independent main lines

Optical communication, high-speed PCB/ABF substrates for AI, computing chips, enterprise storage, advanced packaging and testing, liquid cooling are the six most consensus-driven sectors among institutions in Q2 2026, with the highest performance verification levels
AI investment has entered the performance verification stage, with increased hardware proportion and decreased software proportion. In public fund AI holdings, hardware increased from 42% to 68%, software decreased from 58% to 32%; hardware targets’ first-quarter net profit attributable to parent increased by an average of 85%, software by 12%; the correlation between stock price and performance increased from 0.3 to 0.78

Global institutional consensus:
BlackRock: Overweight AI physical infrastructure and hardware, reduce pure software themes
Morgan Stanley: AI hardware profit cycle supported by platform iteration and architecture upgrades, lasting 2–3 years; power is a constraint, energy storage is a supporting link, not an independent main line
Goldman Sachs: The potential market space for AI optical interconnect increased from $15 billion to $154 billion; the computing architecture shifted from Scale Out to Scale Up, with exponential growth in the value of standalone optical networks and storage
CICC, CITIC: Upgrading from 800G to 1.6T in optical modules, high-speed PCB for AI, HBM / enterprise storage are the most certain incremental directions for 2026

2. Optical Communication
Industry logic: AI computing density doubles every 3 months; copper interconnects reach physical limits; all-optical interconnects are an inevitable trend; Scale Up architecture drives significantly higher elasticity in optical module value than Scale Out; 800G to 1.6T iteration achieves volume and price growth, industry pattern is concentrated, leading companies’ gross margins exceed 45%

Institutional original text
Morgan Stanley (2026.5.7): AI hardware profit is anchored by platform iteration + architecture upgrades, lasting 2–3 years
Goldman Sachs (2026.5.7): TAM for AI optical interconnect expanded from $15 billion to $154 billion, a ninefold increase; Scale Up drives exponential increase in standalone optical network value

Upstream 1: Core raw materials for optical communication
High-purity quartz sand
Stocks: Quartz Shares
Institutional view: CITIC Securities, Huatai Securities, high-purity quartz sand is the core raw material for fiber preform rods, with highly concentrated global supply and clear trend of volume and price increase.

Indium Phosphide (InP) substrates
Stocks: Yunnan Geology
Institutional view: Goldman Sachs, CITIC Securities, InP is an essential substrate for high-speed EML optical chips, with a significant supply gap, domestic substitution accelerating

Thin-film Lithium Niobate (TFLN)
Stocks: Tian Tong Co., Light Storage Technology
Institutional view: Morgan Stanley, CICC, 1.6T + era of CPO demand, demand growth exceeds 120%

Optical crystals
Stocks: Fujing Technology
Institutional view: Huatai Securities, crystals are the core passive components for optical modules, global mainstream suppliers, stable industry pattern

Optical chip substrates / epitaxy
Stocks: San’an Optoelectronics, Huaqiang Silicon Industry
Institutional view: Goldman Sachs, silicon photonics and high-speed optical chips drive substrate demand, domestic substitution entering an accelerated phase

Upstream 2: Optical rods / fibers / chips
Quartz Shares: significant global supply share of high-purity quartz sand, core raw material for optical rods / fibers
Long Fiber Optic: leading global shipment of optical rods / fibers, core manufacturer of hollow-core fibers
Xunfei Technology: leading self-supply rate for high-speed EML optical chips, significant autonomous control barriers
Yuanjie Technology: core target for high-speed EML optical chips, linked with top optical module manufacturers
Institutional view: CITIC Securities, high-speed optical chip gap of 25–30%, increased self-supply brings sustained performance elasticity

Midstream: Optical devices / modules / optical engines
Zhongji Xuchuang: global leader in 800G/1.6T optical modules, NVLink certified, 217 institutional research visits, heavily held by public funds and northbound funds
Xinyi Sheng: leading silicon photonics technology, high yield for 1.6T products, increasing market share among overseas cloud providers
Tianfu Communication: CPO optical engine packaging supplier, deeply linked with Nvidia
Dongshan Precision: AI computing full-stack hardware leader, driven by PCB + optical chips / modules; acquired Sors to develop 200G EML chips (self-supply rate > 99%), mass supply of 800G/1.6T; linked with Nvidia, Meta, Apple, Tesla, full order capacity; Q1 2026 net profit +143% YoY, performance inflection point, high prosperity and high elasticity target

Institutional view: Goldman Sachs, iteration from 800G to 1.6T boosts single-machine value by 8–12 times, leading companies’ profits continue to exceed expectations

Downstream: Computing interconnect / transmission support
Application: AI server clusters, intelligent computing centers, 800G/1.6T switch interconnects, data center full optical interconnects

3. High-speed PCB/ABF substrates for AI (upstream materials → midstream manufacturing → downstream servers / substrate support)
Industry logic: AI server single-machine PCB value is 5–10 times that of traditional servers; ABF substrates upgrade to 18L and above, global supply gap widens from 2027; high-end electronic fabric and high-speed CCL technology barriers are high, supply-demand gap exceeds 20%, delivery cycle exceeds 9 months

Institutional original text
Goldman Sachs (2026.5.7): Agentic AI drives token demand surge by 24 times, benefiting optical communication, PCB, storage, high-end electronic fabric
Goldman Sachs (2026.4.30): High-end electronic fabric supply is tight, Honghe Technology is the only core domestic supplier, scarcity grants valuation premium

Upstream: Copper-clad laminate / electronic fabric / raw materials
Shengyi Technology: second-largest global CCL manufacturer, core supplier for high-speed copper-clad laminate
Honghe Technology: domestic high-end electronic fabric supplier, 35% market share of ultra-thin electronic fabric globally, gross margin over 61% for low DK / low CTE products, Q1 2026 net profit YoY +354%

Institutional view: Morgan Stanley, high-speed CCL and high-end electronic fabric are essential for AI PCB, with oligopoly pricing power

Midstream: High-speed PCB/ABF substrate manufacturing
Shu Dian Co.: Nvidia backplane supplier, leading global market share for 800G switch PCBs, orders locked until 2027
Shennan Circuit: leading domestic technology for AI high-level PCBs and ABF substrates, certified by Nvidia and Huawei
Shenghong Technology: domestic core supplier for AI server PCBs, market share continuously rising

Institutional view: CICC, ABF substrates will face global shortages starting 2027, leading firms will benefit first

Downstream: AI servers / switches / computing hardware
Application: Nvidia Blackwell servers, domestic Manka clusters, high-speed switches

4. Computing chips + storage (upstream materials / equipment → midstream chips / storage → downstream modules / applications)
Industry logic: Global storage industry cycle reversal, HBM/DDR5 / enterprise storage supply-demand tightness drives prices up; AI server storage demand is 5–8 times that of traditional servers; domestic computing chips shift from policy-driven to commercial order-driven

Institutional original text
Morgan Stanley (2026.5.7): Enterprise storage modules are the core of storage chain resilience
Goldman Sachs (2026.5.7): Scale Up revolution drives 29 times increase in standalone optical network and storage value

Upstream: wafers / consumables / equipment
Supporting links: large silicon wafers, wet electronic chemicals, semiconductor equipment, lithography materials, special gases
Institutional view: Goldman Sachs, storage capacity expansion drives rigid demand for upstream consumables, domestic substitution accelerates

Midstream: Computing chips / storage chips
Huaiguang Information: core target for domestic computing chips, 189 research visits, heavily held by public funds and northbound funds
Cambrian: full-stack self-developed architecture, well-developed high-end computing products
Lankong Technology: core target for HBM and DDR5 industry chain, global supply-demand tension
GigaDevice: storage cycle reversal, driven by AI and automotive-grade dual businesses

Institutional view: Morgan Stanley, supply-demand gap for HBM/DDR5 persists, price increase cycle is clear

Downstream: Storage modules / enterprise applications

Jiangbolong: leading domestic enterprise storage module manufacturer, self-developed controllers and SPU intelligent storage, Q1 2026 net profit +2644% YoY, gross margin 55.53%, deeply linked with Yangtze Memory and storage, overseas revenue accounts for over 70%

Institutional view: Morgan Stanley, enterprise storage modules are core to storage chain resilience, counter-cyclical inventory accumulation yields excess returns.

5. Advanced packaging and testing + semiconductor equipment (upstream components → midstream equipment / testing → downstream chip manufacturing support)
Industry logic: AI chips evolve towards HBM3E stacking and Chiplet, increasing packaging and testing value; domestic substitution of semiconductor equipment accelerates, proportion of commercial orders increases, industry expansion demand is rigid

Institutional original text
Morgan Stanley (2026.5.7): AI hardware profit is anchored by platform iteration + architecture upgrades, lasting 2–3 years

Upstream: Equipment components / packaging materials
Supporting links: semiconductor components, packaging substrates, bonding materials, lithography materials, ceramic parts

Institutional view: CICC, domestic substitution of equipment components and packaging materials is entering realization phase.

Midstream: Semiconductor equipment / advanced packaging and testing
North Micro: supplier of core equipment for AI chip expansion, broad domestic substitution space, 172 research visits
AME: second in global etching equipment for 5/3nm nodes, prominent technical barriers
Changdian Technology: third-largest global packaging and testing manufacturer, domestic leader in HBM3E stacking packaging and testing
Tongfu Microelectronics: deep cooperation with AMD, inflection point in advanced packaging and testing business

Institutional view: Goldman Sachs, the value of advanced packaging will double, equipment manufacturers will benefit fully from the expansion cycle.

Downstream: Wafer manufacturing / chip packaging and testing services
Application: expansion of SMIC and other wafer fabs, HBM stacking and testing, Chiplet integration

6. Computing power leasing + liquid cooling (upstream equipment → midstream operation / heat dissipation → downstream computing services)
Industry logic: global computing power supply-demand gap continues to widen, leasing rates near saturation, rents rising; high-density computing scenarios promote liquid cooling penetration, business mainly long-term contracts, high gross margins.
Upstream: servers / liquid cooling hardware
Supporting links: AI servers, cold plates, liquid cooling units, cabinets, piping, heat exchange materials

Midstream: Computing operation / liquid cooling solutions
Litong Electronics: Q1 2026 net profit +821% YoY, leasing rate near 100%
Runjian Co.: integrated computing and power model, stable business loop, order locked until 2027, Q1 net profit +343% YoY

Shuguang Digital: liquid cooling technology widely used in Manka clusters (token factories?)

Institutional view: CITIC Securities, persistent computing power gap, rising leasing rates and rents, high performance certainty

Downstream: government and enterprise / model manufacturers’ computing procurement
Application: large model training, AI inference, intelligent computing center outsourcing

7. AI servers (upstream components → midstream assembly → downstream computing deployment)
Upstream: CPUs / GPUs / PCB / power supplies / cooling / storage
Supporting links: computing chips, high-speed PCBs, storage modules, cooling systems, power supplies, connectors

Midstream: server assembly
Inspur Information: leading domestic AI server market share, core supplier for Manka clusters
Sugon: dual attributes of trust innovation and AI, stable government orders
Institutional view: Goldman Sachs, AI server shipments continue to grow rapidly, leading companies’ market share is concentrated

Downstream: computing centers / cloud deployment
Application: public cloud, intelligent computing centers, enterprise AI clusters

8. Power / Green Power / Energy Storage
Institutional positioning: power is a constraint for AI computing power; energy storage is a supporting link for flexible power supply, not an independent main line; institutional allocation mainly 5–10% core positions
Morgan Stanley (2026.5.7): Power is a constraint, energy storage is supporting, not an independent main line
Core stocks: CATL, Sungrow, State Grid South Power

9. Downstream applications / robots / quantum
Institutional positioning: these directions are market rotation side lines, without independent main line trends; not core allocation directions

Risk Tips
Overseas cloud providers reduce capital expenditure, affecting optical modules, PCB, storage demand
Fluctuations in semiconductor equipment and optical chip supply chains, impacting delivery schedules
Intensified price competition for high-end products, compressing industry gross margins
Market style shifts causing phased valuation corrections for high prosperity hardware targets

Note: This report is objectively compiled based on public research reports and market data, and does not constitute investment advice!!! Highlighted stocks in bold can be paid more attention to

This is the second public research report from this account; original content is hard-won. If you like it, please give more likes, support, and understanding!!!

Attachment:

P.S.: Everyone can follow @Bingchuan688, whose content is the practical section, a very talented blogger. Below is his May outlook, for reference only!