Why is optical communication continuously rebounding? Can it still hold on?

Recently, due to the escalation of geopolitical conflicts, market risk appetite has continued to decline. Coupled with a style switch in some funds flowing into cyclical sectors, the technology sector has faced pressure and volatility. On March 18, the computing power industry chain experienced a collective explosion. On March 19, despite the pressure on the broader market, computing power leasing strengthened against the trend. On March 20, the computing power sector continued its strength, with the largest communication ETF, China Asset Management (515050), rising over 4% during the day, and the lowest-fee ChiNext AI ETF (159381) rising over 3%. Source Code Technology (688498.SH) reached a new historical high, while stocks such as Zhuosheng Micro (300782.SZ), Guangku Technology (300620.SZ), New Yi Sheng (300502.SZ), Zhongji Xuchuang (300308.SZ), and Tianfu Communication (300394.SZ) also performed well. 【The increase in the secondary market for ETFs does not represent the fund’s net asset value performance. Investing comes with risks; please invest cautiously.】

In the past five days, the communication ETF China Asset Management (515050) has received a net inflow of over 160 million yuan.

After the rapid rebound in computing power, should we lock in profits or continue to hold? Is it a good time to enter, or will chasing higher prices lead to losses? Let’s discuss this in detail today.

What is the background and catalyst for the rebound?

This round of rebound is not driven by short-term speculative funds but is catalyzed by the industry’s fundamentals exceeding expectations.

On one hand, domestic computing power has sparked a wave of price hikes, validating the industry’s high prosperity: Alibaba Cloud announced a maximum price increase of 34% for AI computing power, storage, and other products, primarily due to a surge in Token usage. Alibaba Cloud’s MaaS business achieved its highest growth rate in history from January to March this year. Baidu (BIDU.US)'s intelligent cloud issued a price adjustment announcement for AI computing power, storage, and other products on the same day. AI computing-related product services: raised by about 5% to 30%; parallel file storage, etc.: raised by about 30%. These prices will take effect from April 18, 2026.

The popularization of AI applications and OpenClaw has ignited demand for inference computing power, combined with NVIDIA (NVDA.US)'s limited production capacity, rising hardware costs, and gaps in domestic alternatives, driving the market into a “seller’s market.” The market expects that short-term price increases are likely to continue.

On the other hand, the two major global industry events that opened concurrently released unexpected industry signals, providing solid fundamental catalysts for the sector’s performance.

First is the OFC 2026 Optical Communication Conference, which, as the most authoritative technology and industry barometer in the global optical communication field, core-validated the speed of technological iteration and demand prosperity in the optical communication industry: 1.6T optical modules officially entered their first year of mass production, having completed certification from major cloud providers in North America, with several institutions raising the global shipment forecast for 2026 to over 10 million units, achieving several times growth compared to 2025; 3.2T optical modules completed the release and verification of the full-stack technical solution, with the pace of generational iteration faster than the market’s previous expectations; meanwhile, cutting-edge technologies such as CPO, NPO, and OCS all released commercialization plans, breaking the pessimistic expectation of “peak demand for optical modules and worsening price wars affecting industry profitability,” clearly defining the industry’s long-term growth path.

Second is the NVIDIA GTC 2026 conference, which, as a core barometer of the global AI industry, set a clear tone for the computing power industry chain through Huang Renxun’s keynote speech and technology release. The core industry signals include three points:

First, the demand-side guidance significantly exceeded expectations. Huang Renxun clearly stated that as of now, by 2027, we can expect at least $1 trillion in highly certain computing power-related procurement demand, greatly exceeding market expectations and directly dispelling market concerns about slowing growth in AI computing power demand;

Second, the industry phase has clearly shifted. AI has fully entered the industrial era of inference + intelligent agents + physical AI from the era of large model training. Token, which is the smallest semantic unit for AI large models to understand and generate language, can be simply understood as the “language atom” of the AI world, will become a core commodity in the AI era. The scenarios of inference and intelligent agents will lead to exponential growth in Token consumption, significantly enhancing the sustainability and rigidity of computing power demand;

Third, the technological iteration and interconnection route are clear. The Rubin Ultra architecture GPU, Feynman forward-looking architecture chip, and LPU inference chip integrated with Groq technology have achieved a leap in computing power performance and inference efficiency, while clearly defining the “copper-light dual approach” interconnection route, stating that the industry chain needs more copper cables, optical chips, and CPO production capacity, calming previous market disputes about the technological route. The performance leap of computing power chips will directly drive the upgrade in demand for supporting products such as optical modules and CPO.

It is essential to clarify that domestic optical module manufacturers possess the capability to participate in the global computing power industry chain, deeply binding themselves to global AI giants such as NVIDIA and Microsoft (MSFT.US). The technological upgrades and capital expenditure expansions of overseas giants will directly translate into orders and performance for leading domestic manufacturers, catalyzing solid fundamental support.

What stage is the computing power sector in currently? Is there significant bubble accumulation?

Here is the conclusion: The computing power sector has officially transitioned from the early phase of thematic speculation to the performance realization phase of accelerated prosperity; the driving logic of the market has shifted from valuation elevation to sustained performance release support. It is currently in the mid-term of the AI computing power industry’s prosperity upward trend, rather than at the end of the market.

Reflecting on the three development stages of the computing power market can help clarify the current position:

2023: Expectation speculation period, with ChatGPT igniting enlightenment in the AI industry. The market’s core pricing is based on the long-term potential of the AI industry, with the sector showing a characteristic of widespread increase, having a low correlation with orders and performance, with returns mainly coming from valuation elevation;

2024-2025: Order verification period, the market returns to rationality, with core pricing shifting to companies’ technological barriers and order realization capabilities, leading to significant differentiation in the sector. Leading companies entering the global giant supply chain continue to strengthen, while purely thematic stocks gradually retreat;

We are currently in the performance realization period and prosperity acceleration period, characterized by leading companies in the industry chain realizing scaled profit release, and the sustainability and certainty of performance growth have been further validated in the two major conferences.

From the latest industry data, optical modules, as the first segment in the computing power industry chain to realize performance, are expected to achieve high growth in net profits for domestic leading manufacturers in 2024 and 2025, with growth coming entirely from the simultaneous increase in volume and price of their main business. As 1.6T optical modules enter large-scale mass production in 2026, the proportion of high-end products will rapidly increase, and the revenue and profit growth rates of leading manufacturers will remain high, with the gross profit margin of 1.6T products significantly higher than that of 800G products, indicating a further potential increase in profitability.

In the computing power leasing segment, the previously feared “overcapacity” has not materialized, with a tight balance in supply and demand for high-end computing power. The core reason is that AI demand has shifted from temporary training computing power to continuous and rigid inference computing power, and the demand for inference computing power will grow exponentially with the popularization of AI applications and intelligent agents, providing long-term support for industry prosperity.

Regarding the question of whether “the market has reached its peak,” it is essential to clarify: the sustainability of rising stock prices primarily depends on whether the driving factors are valuation bubbles or performance support. If the valuation is only driven by emotions, the market lacks sustainability; however, if corporate performance continues to grow rapidly, even if valuations remain stable, performance growth will push stock prices up, which is commonly referred to as “performance digesting valuation.” The current rise of the computing power sector has shifted from early emotional drives to being fundamentally driven. As long as there is no systemic turning point in industry prosperity, and the core logic of performance growth is not compromised, the market has a sustainable foundation.

How to seize the computing power price increase trend?

For ordinary investors, the computing power industry chain evolves rapidly, with significant differentiation among individual stocks, facing high non-systematic risks related to technology, orders, and performance falling short of expectations. ETFs, as investment tools comprising a basket of leading stocks, can capture beta returns from rising industry prosperity while effectively diversifying individual stock risks, making them a better choice for ordinary investors to enter the computing power sector.

This introduction presents three computing power-related ETF products, which can be matched according to individual investment needs:

Communication ETF China Asset Management (515050)

Tracks the CSI 5G Communication Theme Index, with holdings covering leading hardware related to optical chips, optical modules, optical devices, optical fibers and cables, PCBs, storage, etc. It comprehensively anchors core beneficiaries in the CPO/optical communication industry chain, with constituent stocks primarily being core manufacturers in the global AI computing power supply chain, highly aligned with the core tracks catalyzed by the two major conferences, ensuring strong performance realization certainty. Suitable for investors focusing on high prosperity in CPO/optical communication hardware and seeking precise positioning. Off-market connections (Class A: 008086; Class C: 008087)

ChiNext AI ETF China Asset Management (159381)

Tracks the ChiNext AI Index, with holdings covering the entire AI industry chain, including CPO, computing power leasing, AI large models, AI agents, etc., balancing computing power hardware with software applications, highly matching the “inference + intelligent agent” industry theme set by this year’s GTC conference, with prominent growth elasticity. Additionally, this product is the lowest-fee AI-themed ETF in the entire market, offering significant cost advantages for long-term holding and regular investment. Suitable for investors optimistic about the long-term development of the entire AI industry, balancing elasticity and holding costs. Off-market connections (Class A: 025505; Class C: 025506)

Cloud Computing ETF China Asset Management (516630)

Tracks the cloud computing index, focusing on the domestic cloud computing software and hardware ecosystem. Its constituent stocks deeply cover core targets related to domestic AI alternatives, DeepSeek, and AI agents, serving as a core tool for capturing the dividends of AI localization. Suitable for investors optimistic about AI independence and domestic cloud computing and AI software application themes. Off-market connections (Class A: 019868; Class C: 019869).

Risk Warning:

The Communication ETF China Asset Management tracks the CSI 5G Communication Theme Index, with index performance from 2021 to 2025 being: 6.05%, -38.04%, 15.85%, 23.37%, 98.42%. Historical index performance does not predict future performance of fund products.

The ChiNext AI ETF China Asset Management tracks the ChiNext AI Index, with index performance from 2021 to 2025 being: 17.57%, -34.52%, 47.83%, 38.44%, 106.35%. Historical index performance does not predict future performance of fund products.

The risk level of the above funds is R4 (medium-high risk), with specific risk rating results based on the ratings provided by the fund manager and sales institution. This material is for informational purposes only and does not constitute any investment advice or commitment, nor does it serve as any legal document. Investors must carefully read the fund’s “Fund Contract,” “Prospectus,” and “Product Information Summary” and fully understand the risk-return characteristics and product features of the fund, making rational decisions based on their own risk tolerance. The operating time of funds in China is relatively short and may not reflect all stages of stock market development. The market carries risks; please invest cautiously.

Important Fee Notice: The ETFs mentioned above do not charge subscription fees, redemption fees, or sales service fees. Subscription and redemption agents may charge a commission not exceeding 0.5%, which includes fees charged by securities exchanges, registration, and settlement agencies. Furthermore, the management and custody fees for the relevant ETFs are deducted from the fund’s assets, with specific rates shown in the accompanying table.