share_log

国泰君安:大算力时代必经之路 关注COWOS及HBM投资链

GTJA: The inevitable path of the era of great computing power, focusing on the COWOS and HBM investment chain.

Zhitong Finance ·  07:06

GTJA Securities stated that based on the layout strategy of AI cloud-side collaboration, its demand for AI servers is expected to further increase in the future, and its supply chain will benefit deeply.

According to the Intelligent Financial APP, GTJA Securities released a research report stating that Apple is actively deploying AI cloud servers and using high-end chips such as M2 Ultra to process advanced AI tasks. Based on the layout strategy of AI cloud-side collaboration, its demand for AI servers is expected to further increase in the future, and its supply chain will benefit deeply.

M2 Ultra deploys cloud servers, focusing on empowering AI-side inference on the cloud. According to IT Home, Apple's self-developed M2 Ultra chip will be deployed to cloud computing servers. Through cloud-side generation of AI large models, AI functions such as intelligent email replies, article and note summaries, and image generation for AI-side devices such as iPhone and MacBook will be effectively improved, and the service experience of voice assistant Siri will be enhanced. Meanwhile, cloud servers equipped with M2 Ultra effectively isolate user data through Secure Enclave to prevent privacy leaks and protect user privacy and security.

M2 Ultra has higher computing power than A100, but chip interconnectivity still needs improvement. Computing Power: M2 Ultra's 76-core GPU has an FP32 floating-point computing power of 27 TFLOPS, which is 39.5% higher than A100 and 40.3% of H100; Cost: The M2 Ultra chip has a production cost of about $2,000, which is one-tenth of H100; Chip interconnectivity: M2 Ultra's chip interconnect architecture is the UltraFusion bridge, which has a significant gap in data transmission rate compared to NVLink, limiting the inference capability of M2 Ultra clusters, which are more likely to be used in billions of parameter cloud-side large model inference calculations.

Apple may lay out server chips in the future to drive hardware demand. The deployment of M2 Ultra on cloud servers marks the starting point of Apple's server chip layout. In the future, it may improve its cloud server layout by authorizing the use of high-speed SerDes technology and developing data center-specific chips. Apple's increased investment in cloud servers will spur demand for high-frequency, high-speed PCB, high-end IC carrier boards, and passive components, and the domestic Apple supply chain is expected to benefit deeply.

The translation is provided by third-party software.


The above content is for informational or educational purposes only and does not constitute any investment advice related to Futu. Although we strive to ensure the truthfulness, accuracy, and originality of all such content, we cannot guarantee it.
    Write a comment