In the south, take a double long position on Nvidia (07788), which rose nearly 3%, and as of the time of writing, rose 2.83% to 161.9 Hong Kong dollars, with a turnover of 1.8381 million Hong Kong dollars. On the news front, on April 20, South Korea's SK Hynix announced that it has begun mass production of the next generation memory module designed for Nvidia's next-generation Vera Rubin AI chip. The company stated that it has started mass production of the 192GB SOCAMM2, which is a next-generation memory module designed to provide functionality and lower energy consumption for AI servers. SK Hynix emphasized that the mass-produced SOCAMM2 product has more than doubled the bandwidth and improved energy efficiency by more than 75% compared to traditional RDIMM2, providing an optimized solution for high-performance AI operations. Oriental Securities believes that in the era of inference AI, the core challenge of computing power has gradually shifted from computing power to storage and bandwidth efficiency. KV cache presents new requirements for storage architecture, and the innovation of storage architecture brings new opportunities. Nvidia previously launched the ICMSAI native storage platform at the 2026 CES, adding G3.5 flash layer to optimize KV cache management, relying on BlueField-4 DPU to achieve large capacity and high-speed transmission, with multiple storage vendors participating in the construction. It is also advancing the SCADA plan to improve storage efficiency, with cuDF and cuVS data acceleration libraries being adopted by leading foreign cloud platforms, while domestic Xinghuan Technology is also collaborating with Nvidia on the development of GPU databases to jointly promote cost reduction and efficiency improvement in AI inference and data processing.