Summary of Key Points
The optical module industry is not experiencing a "complete bubble burst," but rather a severe structural differentiation: there is an overcapacity in the low-end market (price wars leading to losses), a tight balance in the mid-range market (leading manufacturers profiting while second-tier companies struggle), and a shortage of high-end capacity (rich customers unable to secure supplies). The real bottleneck lies not in the assembly of optical modules but in their core components—high-speed EML (Electro-Mechanical-Light) chips, particularly 200G EML chips, which are monopolized by American and Japanese manufacturers, with a supply gap of 60%-70%. Orders for these chips are backlogged until 2028. It is unlikely that domestic high-end optical chips will be able to replace imported ones in the short term, and the industry is likely to become more concentrated among leading players, with profits shifting towards the upstream chip manufacturers.
Detailed Analysis
1. The "True and False Overcapacity" in Optical Modules
Optical modules are categorized by speed, and the supply and demand situations vary greatly:
- Low-end (10G/25G, etc.): True overcapacity, a survival struggle
The technical barrier is low, and many small and medium-sized manufacturers have entered the market in the past two years, leading to an overcapacity rate of over 45%. Prices for 25G optical modules have dropped by 30% in one year, with factory prices approaching cost levels. Many manufacturers are operating at less than 50% capacity. As AI computing power becomes mainstream, demand for traditional data centers is declining, and many small and medium-sized manufacturers will be eliminated in the next 2-3 years.
- Mid-range (800G): Tight balance, with leading manufacturers profiting
This is the current main product for AI data centers, with a global demand of about 45 million units and a supply gap of 10%. However, leading manufacturers (such as NeoPhotonics and Innolux) account for 70% of the market share, operating at full capacity with orders backlogged until the end of the year and enjoying high profit margins. Second-tier manufacturers have lower yields (only around 60%) and cannot pass the certification of major customers like NVIDIA, so they can only sell at low prices with thin profits.
- High-end (1.6T): Severe shortage, with chip supply as the bottleneck
This is the core component for the next generation of AI data centers, with a demand of 25 million units and an effective capacity of only 15 million units (a 40% gap). The issue is not that manufacturers do not want to expand production; rather, 1.6T modules require 8 200G EML chips, which are monopolized by overseas companies. Leading module manufacturers have secured their quotas in advance, leaving second-tier manufacturers unable to obtain them. Additionally, second-tier manufacturers have low yields, and many of their products are substandard and not considered valid supply.
2. The Truth Behind the "Order Cuts" Rumors
The so-called "order cuts" are not a sign of overall demand decline but rather an upgrade in demand:
- The cuts affect low-speed optical modules and the mid-to-low-end orders of second-tier manufacturers. The orders for 800G/1.6T products from leading manufacturers have not decreased; instead, they have increased (for example, Google and Meta have increased their purchases of 1.6T modules by more than 30%).
- Customers are simply adjusting their portfolios: reducing orders for 800G modules and increasing those for 1.6T modules to prepare for the next generation of AI clusters. Manufacturers that cannot keep up with technological advancements will naturally be eliminated.
3. The Real Bottleneck: The 200G EML Chip Shortage
The 200G EML chip is the "heart" of optical modules, as it converts electrical signals into high-speed optical signals. The situation is severe:
- Large gap: Global demand for these chips in 2026 is 150 million units, but effective capacity is only 50-80 million units, resulting in a 60%-70% shortage. Delivery times have increased from 16 weeks to 40 weeks (nearly 10 months), and spot prices are 50% higher.
- Monopoly: Three American and Japanese companies (Lumentum, Coherent, and Sumitomo Electric) control 80% of the high-end market capacity and are cautious about expanding production due to concerns about fluctuations in AI demand.
- Upstream Supply Issues: The phosphide indium substrates used to produce EML chips are monopolized by Japanese and American companies, with prices rising by 200%. Delivery times have increased from 8 weeks to 24 weeks, leaving chip manufacturers without the necessary raw materials to expand production.
4. Domestic Optical Chips
Domestic chips can replace low-end and mid-range products (100G and below), but not high-end 200G EML chips in the short term:
- Current Status: Yuanjie Technology has a yield of 45%-50% for 200G EML chips (under sample verification); Changguang Huaxin plans mass production in Q4 2026; Guangxun Technology supplies a small amount to internal use only.
- Three Major Challenges:
① Yield: Overseas manufacturers have yields of over 90%, while domestic ones are between 45%-50% (doubling the cost difference).
② Certification: It takes 12-18 months for domestic chips to enter the supply chains of major customers.
③ Ecosystem: Overseas manufacturers have had partnerships with Google and NVIDIA for over a decade, making it difficult for domestic companies to gain customer trust.
- Conclusion: Domestic substitution of high-end chips is unlikely until 2027-2028, unable to address immediate needs.
5. Industry Future
- Increasing Differentiation: Leading manufacturers will strengthen their relationships with chip and customer bases, expanding their market share. Second-tier manufacturers, lacking both technology and chips, will gradually be eliminated, leading to a monopoly by 3-5 companies.
- Profit Shift: In the past, module manufacturers profited; in the future, chip and substrate manufacturers will take a larger share of profits, with module manufacturers becoming mere assembly providers.
- Opportunities for Domestic Companies: The barriers for high-end 200G EML chips are high, but with the next generation of 3.2T optical modules (using 400G EML) and CPO (Coherent Photonic Packaging) technologies, domestic companies have a chance to catch up if they can develop and mass-produce these products in time.
Final Summary
The optical module industry is not experiencing a bubble burst but rather a situation where the low-end market is collapsing, the mid-range market is highly competitive, and the high-end market is in short supply. The real bottleneck is the 200G EML chip. Domestic substitution will take time (at least until 2027-2028), and the most valuable players in the industry will be those who control these core components. This analysis uses plain language to explain complex financial and business concepts, making it easy for non-finance professionals to understand.