The optical module sector experienced a collective surge, with a core group of leading stocks initiating a strong rebound. This propelled the ChiNext Artificial Intelligence ETF Huabao (159363) to a significant gain of 3.26%. Among its top ten holdings, Eoptolink Technology Inc.,Ltd. (SZSE: 300502) soared 11%, Ingenic Semiconductor Co.,Ltd. (SZSE: 300223) advanced 9.21%, Zhongji Innolight Co.,Ltd. (SZSE: 300308) rose 6.86%, and Suzhou Tfc Optical Communication Co.,Ltd. (SZSE: 300394) climbed 5.48%.
In terms of fund flows, the ChiNext Artificial Intelligence ETF Huabao (159363) attracted a net inflow of 466 million yuan as investors bought on weakness.
Catalysts for the Rally
A direct catalyst was the announcement by Meta Platforms, Inc. (Nasdaq: META) of a substantial investment increase, adopting an "energy co-construction" model. The company significantly raised the total investment for its Hyperion data center project in Louisiana from $10 billion to a maximum of $50 billion, with computing capacity scaling to 5GW. Notably, Meta is not only funding the computing infrastructure but is also deeply involved in energy supply, covering the full cost of new power generation facilities and grid upgrades, transforming its data centers from major energy consumers into co-builders of the energy ecosystem.
Furthermore, clarifications from a leading optical module manufacturer effectively boosted market confidence. After market hours on July 12, Zhongji Innolight Co.,Ltd. held an urgent investor meeting to refute market rumors regarding "underwhelming performance" and "deliberate stock price suppression." The company provided clear guidance: current orders already fully cover its production capacity for the entire year of 2026, demand growth for 800G products is significantly exceeding expectations, rumored price cuts are "greatly exaggerated," and the total demand for 1.6T products in 2027 remains largely in line with prior expectations. This statement directly alleviated market concerns about a potential weakening in industry sentiment.
Morgan Stanley Raises Capex Forecasts
On July 13, Morgan Stanley released a thematic report on AI capital expenditure, raising its 2027/2028 Capex forecasts for the five major hyperscalers to $1.23 trillion and $1.4 trillion, respectively. Computing capacity is projected to approach 120GW by 2028, a nearly fourfold expansion from approximately 30GW in 2025.
The most critical aspect of this report is not merely the upward revision of Capex, but the shift in focus from "whether Capex will continue to rise" to "how much effective computing power this Capex translates into, why costs continue to escalate, and whether revenue streams can support it." In other words, the market will now look beyond who spends the most, focusing instead on who can actually deliver computing power and convert it into sustainable AI revenue.
Morgan Stanley's view is that the AI ecosystem remains constrained by computing power supply. Factors such as chips, cabinets, power shells, data center construction cycles, engineering labor, community resistance, and political cycles are all extending the timeline from project initiation to delivery. The period from groundbreaking to operation for some data centers may now stretch to three years. Therefore, this is not a case of "tech giants spending recklessly." It is an infrastructure race where leading companies are forced to start projects early, lock in resources in advance, and secure financing ahead of time to secure computing power options for the next three to five years.
For the optical communication industry chain, more significant than the overall Capex increase is Morgan Stanley's reassessment of unit construction costs. The firm raised the cost per GW of computing power construction by approximately 20%—estimating costs around $35 billion per GW for GB200, $39 billion for GB300, $49 billion for Vera Rubin, and $50 billion for Rubin Ultra.
Brokerage Perspectives on the Sector
GF Securities: Industry Sentiment Continues to Be Validated
The positive sentiment in the optical communication industry continues to be validated, benefiting from accelerated global AI computing infrastructure investment and data center construction. Demand for high-speed optical modules and optical components is growing steadily, with interim performance forecasts from some manufacturers showing significant quarter-on-quarter profit increases. Looking ahead to the second half of the year, tight supply of upstream core materials is expected to ease quarter by quarter. New optical interconnection technologies like NPO/CPO will accelerate their adoption. As one of the steepest growth segments within AI hardware, the optical communication sector will benefit long-term from increased penetration in scenarios like optical connections within cabinets and to storage, continuously raising its market potential.
Guosheng Securities: More Positive and Profound Industry Resonance
Recently, concerns about a peak in AI infrastructure investment and potential computing power oversupply have led to some correction in hardware sectors like optical modules. However, looking beyond short-term sentiment, more positive and profound resonance is occurring within the industry's fundamentals. The current period represents an important window for investing in leading optical communication companies. This judgment is based on three key supports: First, capital expenditure (CapEx) from major North American cloud service providers has not peaked and is expected to continue rising. Second, the pricing dynamics in optical communications are healthier and more growth-oriented compared to traditional cyclical products. Third, within the broader technology sector, optical communications offer strong near-to-medium-term earnings certainty, remaining a crucial pillar for stability in the tech space.
Oriental Securities: Structural Shift in Global Optical Communication
The global optical communication industry is undergoing a structural shift from traditional telecom-driven demand to AI computing-driven demand. The iteration cycle for optical modules has compressed from four years to two to three years, with explosive demand for 800G/1.6T high-speed products. According to LightCounting predictions, the global Ethernet optical module market is expected to reach $26 billion by 2026.
Zheshang Securities: Growing Influence of Switch Chips
The 400G/lane optical DSP represents the next evolution from the 200G/lane architecture, doubling single-lane bandwidth. This can support single-chassis system switching capacities of 102.4T and lays the foundation for 204.8T networks using 3.2T modules. Over a five-year horizon, combined shipments of 1.6T/3.2T optical modules could reach the hundreds of millions, with a significant proportion adopting 400G optics. The influence of switch chips on optical interconnection will continue to strengthen.
Soochow Securities: AI Demand Drives Volume and Price Increases
The explosion in AI computing demand is driving simultaneous increases in both volume and price for AI infrastructure like optical modules. The ongoing speed upgrades (800G→1.6T→3.2T) and form factor evolution (pluggable→CPO) in optical modules continue to push for higher precision and increased value in supporting test equipment. The overall positive direction of the industry has not fundamentally changed.
Accessing the Theme Through an ETF
High-Concentration Exposure ETF
The ChiNext Artificial Intelligence ETF Huabao (159363) offers concentrated exposure, with approximately half of its index weight focused on the optical module/CPO sector. The other half covers domestic computing power and AI software applications, forming a balanced "hardware + application" portfolio. The three leading optical module companies—Eoptolink Technology Inc.,Ltd., Zhongji Innolight Co.,Ltd., and Suzhou Tfc Optical Communication Co.,Ltd.—collectively account for over 39% of the ETF's weight.
To comprehensively capture opportunities in the AI theme, investors may consider the ChiNext Artificial Intelligence ETF Huabao (159363) and its corresponding off-exchange feeder funds (023407.OF, 023408.OF).
Comments