The Zhitong Finance App learned that Dongguan Securities released a research report saying that in the context of continuous iterative development of artificial intelligence applications, computing power resources are the necessary foundation to support application implementation and continuous operation. Parallel computation of large-scale computing power clusters with ultra-massive parameter data places new demands on the optical communication industry, such as large bandwidth and high speed, and drives continuous iteration of products such as optical modules and optical chips. Looking ahead to the future market, demand for computing power at home and abroad will continue to rise. Optical communication device and equipment manufacturers are expected to continue to benefit from this trend. It is recommended to focus on the subsequent development of the optical communication industry chain. Related targets include Zhongji Xuchuang (300308.SZ), Guangxun Technology (002281.SZ), Tianfu Communications (300394.SZ), and Optical Bank Technology (300620.SZ).

Incident: 1) On February 19, the State Council's State-owned Assets Administration Commission held a special promotion conference on artificial intelligence for central enterprises on “AI Empowering Industry Renewal”. The conference emphasized that central enterprises should plan the development of artificial intelligence in overall work, further promote industrial renewal, and speed up the layout and development of the intelligent industry. 2) On February 22, Nvidia disclosed financial results for the fourth quarter and full year of fiscal year 2024 and financial data forecasts such as revenue and GAAP for the first fiscal quarter of fiscal year 2025.

The review of Dongguan Securities is as follows:

The development of artificial intelligence is included in the overall work planning of central enterprises.

The State Council's State-owned Assets Administration Commission held the “AI Empowers Industrial Renewal” Artificial Intelligence Promotion Conference for Central Enterprises and pointed out that it is the functional mission of state-owned central enterprises to accelerate the development of artificial intelligence. Central enterprises should consolidate the foundation for development, concentrate their main resources on the areas where they are most needed and have the greatest advantage, and speed up the construction of a number of intelligent computing power centers. Special AI+ actions will be launched to strengthen demand traction, build a batch of multi-modal high-quality industrial data sets, and create large-scale models from infrastructure, algorithm tools, and intelligent platforms to solutions to empower the industrial ecosystem. This conference has important guiding significance for capital expenditure investment of central state-owned enterprises that have the capacity and responsibility for inclusive infrastructure. The demand-driven construction policy is expected to inject vitality into domestic computing power construction, solve problems such as unbalanced supply and demand for computing power resources and difficulties in matching the computing power of manufacturers' resource pools, and do a better and higher quality construction and deployment in the field of artificial intelligence.

The operator is actively promoting the implementation of a 10,000-card computing power cluster.

As a major domestic telecom operator, China Telecom has continuously promoted the development of industrial digital services and achieved excellent results, continuously optimizing the “2+4+31+X+O” computing power layout. As of the first half of 2023, the company's intelligent computing added 1.8 EFLOPS, reaching 4.7 EFLOPS, an increase of 62%. The company actively assumes the responsibility of central enterprises to promote artificial intelligence infrastructure. According to the IT Times, China Telecom will deploy the country's largest operator-level intelligent computing center in Shanghai, where a single pool will build 10,000 domestic computing power cards. It is the first domestic hyperscale computing power liquid cooling cluster to support a single pool of 10,000 cards.

The data center business of key overseas enterprises grew more than expected, and the optical communication industry chain is expected to continue to benefit.

According to Nvidia, a key overseas manufacturer in the field of artificial intelligence computing, the company's revenue for the full year of the 2024 fiscal year reached a record high of 60.9 billion US dollars, up 126% year on year. Fourth quarter revenue reached 22.1 billion US dollars, up 22% month on month. Among them, data center business revenue reached 18.4 billion US dollars, all higher than previous market expectations and the upper limit of the company's guidance. For the first fiscal quarter of fiscal year 2025, Nvidia predicts that it will continue to grow and achieve a revenue scale of 24 billion US dollars (fluctuating 2% up and down). Nvidia's chief financial officer said demand for its advanced artificial intelligence chips continues to “exceed supply.” Fabrinett, the largest ODM manufacturer in the optical communications industry, achieved revenue of US$713 million in the second fiscal quarter of fiscal year 2024 as an ODM partner for Nvidia optical modules, of which the optical communications business revenue was US$568 million, an increase of 12% over the previous year. LightCounting pointed out that in the first year of deployment, Nvidia manufactured most of its optical modules through Fabrinet, and will allocate most of the demand to optical module giants Coherent and Innotech in the future, and chip and module manufacturers in the optical communications industry chain are expected to continue to benefit.

Risk warning: changes in the external policy environment; downstream applications falling short of expectations; market competition risks, etc.