According to the CITIC Construction Investment Research Report, on September 3, OpenAI released the next-generation flagship model GPT-6 Astra, which further enhances complex reasoning, code generation, web browsing, computer operation, and agent autonomous execution capabilities, and supports complex workflows such as documents, tables, and PPTs. The API context window is about 1.05 million tokens. Broadcom FY2026 Q3 achieved revenue of US$29.59 billion, up 86% year on year, of which AI semiconductor revenue was US$16.7 billion, up 221% year over year, mainly driven by demand for customized AI accelerators and high-speed networks; Q4 expects AI semiconductor revenue to be US$21.7 billion, up 236% year over year. We believe that with the continuous upgrading of large model capabilities combined with the expansion of capital expenditure by overseas cloud vendors, demand for inference and infrastructure computing power is expected to remain high, and we will continue to be optimistic about core AI computing power links such as chips, optical modules, switches, and PCBs.

According to the CITIC Construction Investment Research Report, on September 3, OpenAI released the next-generation flagship model GPT-6 Astra, which further enhances complex reasoning, code generation, web browsing, computer operation, and agent autonomous execution capabilities, and supports complex workflows such as documents, tables, and PPTs. The API context window is about 1.05 million tokens. Broadcom FY2026 Q3 achieved revenue of US$29.59 billion, up 86% year on year, of which AI semiconductor revenue was US$16.7 billion, up 221% year over year, mainly driven by demand for customized AI accelerators and high-speed networks; Q4 expects AI semiconductor revenue to be US$21.7 billion, up 236% year over year. We believe that with the continuous upgrading of large model capabilities combined with the expansion of capital expenditure by overseas cloud vendors, demand for inference and infrastructure computing power is expected to remain high, and we will continue to be optimistic about core AI computing power links such as chips, optical modules, switches, and PCBs.