Brokerage Morning Meeting Highlights | Tesla is expected to drive the smart driving industry chain into a high-volume period.

Yesterday, the market fluctuated and rebounded, with the Shenzhen Component Index rising by over 1% and the ChiNext Index rising by over 2%. The trading volume of the Shanghai and Shenzhen stock markets reached 2.77 trillion yuan. In terms of sectors, CPO, optical fibers, PCB, MLCC, Siasun Robot & Automation and other concepts performed well. In terms of declines, the sports concept collectively adjusted. As of the close, the Shanghai Composite Index rose by 0.43%, the Shenzhen Component Index rose by 1.63%, and the ChiNext Index rose by 2.66%. At the morning meeting of securities firms today, CICC believes that Tesla is expected to drive the intelligent driving industry chain into a period of high volume; China Securities Co., Ltd. believes that Siasun Robot & Automation is an important direction for AI applications, and sector volatility focuses on quality aspects; Zhongtai believes that reasoning is the core of future AI computing power, and CPUs will undergo a value reassessment. CICC: Tesla is expected to drive the intelligent driving industry chain into a period of high volume Recently, Tesla announced in its 1Q26 earnings conference call that it aims to obtain approval for FSD to enter the Chinese market in 3Q26. On May 21, Tesla announced the landing of FSD in China in its global FSD map, and the mass production of Tesla Cybercab started in April 2026. The expansion speed of Tesla's global FSD map is accelerating, which may accelerate the global penetration rate of intelligent driving. At the same time, Tesla's advanced driving and Robotaxi bring a multiplier effect, further increasing consumer acceptance of intelligent driving and driving the industry chain into a period of high volume. China Securities Co., Ltd.: Siasun Robot & Automation is an important direction for AI applications, and sector volatility focuses on quality aspects Physical AI is the next wave of artificial intelligence, and Siasun Robot & Automation is one of the best physical carriers for AI, with billions of autonomous systems and Siasun Robot & Automation systems expected to operate in the physical world in the future. Recently, the Siasun Robot & Automation sector has undergone some adjustments, mainly due to fluctuations in market sentiment, but the narrative of physical AI is a real industry trend that deserves high attention. In addition, the gradual mass production of Optimus is approaching, with clear guidelines for the mass production scale of the supply chain becoming increasingly clear, and its volume rhythm gradually being verified. The future release and mass production of V3 products are still worth paying close attention to. In addition, the continuous progress of the IPO of domestic Siasun Robot & Automation companies is expected to bring about a revaluation of their intrinsic value, with catalytic effects in the sector continuing to focus on quality aspects. Zhongtai: Reasoning is the core of future AI computing power, and CPUs will undergo a value reassessment The transformation of computing power structure is shifting from being mainly focused on training to having reasoning at its core currently, over 70% of computing power is used for centralized training, and in the future, over 70% of computing power will be used for distributed reasoning, with the demand for reasoning expected to reach 5-10 times that of the training stage. The underlying differences between training and reasoning determine that CPUs are "auxiliary" in training scenarios and can become the "main force" in reasoning scenarios 1) Based on Little's Law (Throughput = Concurrency / Latency), CPUs and GPUs have chosen different routes of "reducing latency" and "increasing concurrency." 2) Training is mainly based on large-scale dense matrix operations, with GPUs handling most of the computations, while CPUs only handle data transfer and cluster scheduling, accounting for 10-30% of the time; Reasoning has characteristics such as fragmentation, long tail, and sensitivity to latency, with CPUs being more competitive in tasks such as decoding, sparse computing, long context management, and embedding, capable of handling over 70% of running loads. This article is reprinted from "Cai Lianshe", GMTEight editor: Liu Jiayin.