Bank of China International: Artificial intelligence enters the “physical AI” era, supply chain preparation is expected to accelerate

The Zhitong Finance App learned that Bank of China International released a research report saying that at the 2026 CES exhibition, Huang Renxun announced that AI has entered the “physical AI” era. Robots and autonomous driving are expected to be ideal carriers for “physical AI.” The Rubin platform has entered full mass production. Among them, core raw materials are expected to usher in a critical “from 0 to 1” point, and supply chain preparation is also expected to accelerate. The bank believes that AI has gradually extended from the digital world to the real world. Robots, autonomous driving, and end-side AI hardware are expected to become AI market carriers, and AI is officially commercialized.

Bank of China International's main views are as follows:

Artificial intelligence enters the “physical AI” era

According to a report by TechWeb on January 6, Hwang In-hoon delivered a speech at the 2026 International Consumer Electronics Show (CES), announcing that artificial intelligence has entered the “physical AI” era. The “physical AI” concept emphasizes the need for AI to integrate real physical dynamics such as gravity and friction to accurately perform complex tasks. Its core pillars include the Newton physics engine, the Cosmos basic model platform, and the GPU+LPU hybrid architecture.

Full mass production of the Rubin platform

According to Tencent Technology, Nvidia disclosed complete Rubin data for the first time. Compared with Blackwell, Rubin GPUNVFP4 inference performance increased to 50 PFLOPS (5X), NVFP4 training performance increased to 35 PFLOPS (3.5X), HBM4 memory bandwidth increased to 22TB/s (2.8X), and the NVLink interconnect bandwidth of a single GPU increased to 3.6Tb/s (2X). The Rubin architecture uses Vera CPU, Rubin GPU, NVLink 6, ConnectX-9, BlueField-4, and Spectrum-6 co-designed to achieve a revolutionary reduction in inference costs. According to TechWeb reports, Rubin's order volume has reached 300 billion US dollars. Rubin has fully entered the mass production stage, and products based on this platform will be launched through partners in the second half of 2026.

Robots and autonomous driving are ideal carriers for “physical AI”

According to TechWeb reports, Hwang In-hoon called robots “the ultimate form of AI” and announced deepening cooperation with Tesla to provide AI brains for Optimus humanoid robots. Optimus completed more than 90% of training through the Omniverse digital twin platform, and the autonomous operation ratio reached 85%, far exceeding the industry average. The mass production plan includes: mass production of 50,000 units in 2026Q1, reducing the cost to less than 20,000 US dollars; increasing production capacity to 100,000 units by the end of 2026, covering the entire process of parts sorting, assembly, and quality inspection. At the same time, Hwang In-hoon believes that autonomous driving is a trend in the automotive industry. Nvidia opened the Alpamayo series of AI models to support L4 autonomous driving.

AI core raw materials are expected to usher in a key point of “from 0 to 1”

The bank anticipates that the PCB and CCL solutions corresponding to the Compute Tray/Switch Tray/Midplane/CPX of the Nvidia Rubin server will upgrade the M8/M8.5/M9/M9 solution respectively. Among them, the M9 solution may use a material combination of high-frequency high-speed resin+HVLP4/5 copper foil+Q cloth, while the M8.5 solution may use a material combination of high-frequency high-speed resin+HVLP4 copper foil+low-DK second-generation cloth. The bank believes that the Rubin Ultra server is expected to use an orthogonal backplane solution of M9 resin+high-grade HVLP copper foil+Q cloth. The bank expects the upstream supply chain of Rubin servers to begin a wave of preparation in the first half of 2026. At that time, the core raw materials for M8.5 and M9 PCB/CCL are expected to usher in a critical “from 0 to 1” point.

Supply chain stocking is expected to speed up

The bank believes that AI has gradually extended from the digital world to the real world. Robots, autonomous driving, and end-side AI hardware are expected to become AI market carriers, and AI is officially commercialized. At the same time, “physical AI” requires greater computing power, data and other resource support, which places higher demands on AI computing power infrastructure construction. Mass production of the Rubin platform means more advanced AI computing power performance, data transmission efficiency, and inference costs, and is expected to stimulate the shipping progress of the upstream supply chain.

Investment advice

It is recommended to focus on Nvidia's supply chain: [PCB] Shenghong Technology, Shanghai Electric Power Co., Ltd., Shennan Circuit; [CCL] Shengyi Technology; [Q] Feilihua, Sinoma Technology; [Resin] Dongcai Technology.

The main risks faced by ratings

Overheating demand in the AI market has caused an industry bubble. Long-term supply-side overcapacity has led to a decline in prices. Technological changes have led to the elimination of original products