China Securities Co., Ltd. has released a research report stating that since the beginning of this year, optical fiber prices have risen significantly, leading to a notable improvement in investment returns and prompting multiple companies to initiate expansion plans for optical fiber preforms. The firm believes that the development of AI is significantly driving demand for optical fiber, and the industry currently remains in a state of supply shortage. Although many companies plan to expand production, most have limited past experience in manufacturing optical preforms, and it remains to be seen whether they can expand as scheduled. Furthermore, even if expansion proceeds as planned, the majority of new capacity will not be released until the second half of next year or later. Therefore, the impact on the industry's supply-demand dynamics for this year and even next year is expected to be limited. Additionally, AI demand may become even stronger, so there is no need for excessive concern in the short term.

Additionally, OpenAI has released its new-generation AI model series, GPT-5.6, which includes three versions: Sol, Terra, and Luna. The firm maintains an optimistic outlook on large AI models. It continues to recommend the AI industry chain, advising a sustained focus on allocating to the AI computing power industry chain. The main views of China Securities Co., Ltd. are as follows:

Optical Preform and Fiber Expansion Wave Arrives

Since the start of the year, the substantial increase in optical fiber prices and significantly improved investment returns have driven several companies to launch expansion plans for optical preforms. For example, Hengtong Optic-Electric Co., Ltd. initiated the second-phase 800-ton optical preform expansion project in Inner Mongolia in April. Far East Co., Ltd. announced in April a fundraising plan of 20 billion yuan to build a new 1,800-ton AIDC-specific optical fiber preform facility. Tongding Interconnection Information Co., Ltd. announced in May plans for a joint venture to build a 600-ton optical preform and 20 million core-kilometer fiber project. Hoshine Silicon Industry Co., Ltd. filed for a 3,200-ton optical preform project in June. Han's Laser Technology Industry Group Co., Ltd. announced on June 25th plans to expand production by 2,000 tons of optical preforms and 60 million core-kilometers of fiber.

The firm believes AI development is clearly boosting demand for optical fiber, and the industry remains undersupplied. While many companies plan to expand, most have limited prior experience in optical preform manufacturing, and their ability to expand on schedule requires observation. Moreover, even if expansion proceeds as planned, most new capacity will be released in the second half of next year or later. Consequently, the impact on supply-demand dynamics for this year and even next year is limited. Furthermore, AI demand may become more robust, so short-term excessive concern is unwarranted.

GPT-5.6 Series Released

Recently, OpenAI officially launched its new-generation AI model series, GPT-5.6, comprising three versions: Sol, Terra, and Luna. The flagship model, Sol, set a new record with a score of 91.9% on the programming benchmark TerminalBench 2.1. Its cybersecurity capabilities are close to those of Claude Mythos 5, while consuming only one-third of the output token resources. Its score on the biomedical benchmark Genebench v1 improved by 17% compared to the previous generation.

GPT-5.6 adopts a tiered, capability-based foundational pricing strategy. The flagship model Sol maintains a price of $5 per million input tokens. The balanced model Terra is priced at $2.5 per million input tokens, with performance on par with GPT-5.5, making it suitable for daily development. The high-speed model Luna offers a price of $1 per million input tokens, with throughput increased threefold, suitable for batch processing.

Anthropic stated that the U.S. government has notified the company that Mythos 5 can be redeployed to a group of U.S. agencies responsible for operating and maintaining critical infrastructure. The company is working to restore access permissions for these agencies as soon as possible. It will also continue collaborating with the U.S. government to further expand the availability scope of Mythos 5.

The firm believes that large models are still within a narrative of continuously rising capability ceilings. Simultaneously, large model vendors are adopting tiered pricing strategies through performance optimization, such as OpenAI's GPT-5.6 balanced model Terra (performance equal to the previous generation but at significantly lower cost) and high-speed model Luna (throughput increase leading to a sharp drop in per-processing cost), alleviating enterprise customers' cost concerns for deploying AI in complex operations.

Furthermore, top-tier large models are beginning to be deeply tied to national security and compliance barriers. Anthropic's Mythos 5, with direct intervention from the U.S. government, has been selectively made available to agencies maintaining critical infrastructure. Meanwhile, OpenAI explicitly highlighted cybersecurity and biomedical benchmark results in its GPT-5.6 capability demonstrations. This indicates that the network and biochemical capabilities of leading AI models have been strategized, and AI models have become national core assets.

The firm remains optimistic about large AI models. Leading vendors are promoting a continuous decline in the marginal cost of large model inference through underlying architecture optimization, thereby comprehensively accelerating the commercialization ecosystem of AI applications in high-frequency B2B scenarios. The firm continues to recommend the AI industry chain. With technological iteration on the model side and an ecosystem explosion on the application side, the demand for expansion and upgrading of underlying computing power infrastructure will be further confirmed. It is advised to maintain a focused allocation to the AI computing power industry chain.

Risk Factors to Consider

Changes in the international environment could impact the security and stability of the supply chain, affecting the overseas expansion progress of relevant companies; tariff impacts exceeding expectations; slower-than-expected development of the artificial intelligence industry, affecting demand for companies in the cloud computing industry chain; intensifying market competition leading to rapid declines in gross margins; exchange rate fluctuations affecting the exchange gains and gross margins of export-oriented enterprises, including those in the ICT equipment, optical module/optical component sectors; slower-than-expected development of the digital economy and Digital China construction; slower-than-expected development of cloud computing business by telecom operators; lower-than-expected capital expenditures by operators; lower-than-expected capital expenditures by cloud service providers; and weaker-than-expected demand in the communication module and smart controller industries.