Computing power: NVIDIA's GB300 AI server is expected to be unveiled in Q2 2025, and Shanghai has introduced a blockbuster plan to make great strides towards a smart computing city

　　　 According to a report by Digitimes on the 2nd, NVIDIA's next-generation GB300 server is currently in the intensive design stage, and it is expected to be released in Q2 2025 and trial production in Q3. The GB300 has stronger cooling requirements and fewer motherboard fans, which means that its liquid cooling needs will be stronger. The server will be designed and performed in all aspects, with the main goal of dramatically increasing AI computing power to meet the needs of today's rapidly evolving AI market.

On the chip side, the GB300 superchip will be based on the updated B300 GPU with stronger FP4 performance. The GPU power consumption will be further increased from 1000W on the B200 to 1400W, which is twice that of the original B100. HBM's memory specifications will also be upgraded to an 8-stack 12Hi HBM3E with a total of 288GB.

In terms of connectivity, the GB300 server will be equipped with NVIDIA's latest ConnectX-8 SuperNIC network interface card, plus an optical module with a theoretical bandwidth of 1.6Tbps, which greatly improves the data transmission capacity. This means that the GB300 server will be able to provide smoother and more efficient performance in application scenarios such as massive data exchange and real-time AI inference.

In terms of overall design, the GB300 AI server will also be equipped with a capacitor tray and an optional BBU battery backup unit to improve the ability to cope with power supply anomalies. It is estimated that five BBU modules will need to be installed on each GB300 system, and the unit price of these modules will be about US$300 after mass production, which is about 2,188 yuan. The unit price of a GB300 NVL72 rack is expected to be US$20 to US$25, or about RMB 146 to 182, after mass production. These design optimizations not only improve the reliability of the product, but also provide an important guarantee for the overall experience of data center users.

NVIDIA's GB300 server, with its hardware configuration and design details, fully reflects its expansion of the boundaries in the field of AI computing solutions, which is expected to help enterprises efficiently cope with the growing demand for data processing.

Recently, the General Office of the Shanghai Municipal People's Government issued the "Implementation Plan for Artificial Intelligence "Molding Shencheng". The plan points out that by the end of 2025, Shanghai plans to build a world-class artificial intelligence industry ecology, strive to exceed 100EFLOPS of intelligent computing power in the city, form about 50 industry open corpus demonstration and application results with remarkable results, build 3 to 5 large model innovation acceleration incubators, and build a number of upstream and downstream collaborative empowerment centers and vertical model training fields. The specific implementation is as follows:

Strengthen the empowerment of the basic base: It is required to build a super-large-scale independent intelligent computing cluster, build a multi-level corpus supply system, build a super-large training field for the integration of virtual and real, and accelerate the innovation of the industry base large model system.

Accelerate the creation of key productivity tools: focus on the development of "artificial intelligence + intelligent terminals", "artificial intelligence + scientific intelligence", "artificial intelligence + online new economy", "artificial intelligence + autonomous driving" and "artificial intelligence + embodied intelligence".

Promote the application of key vertical fields: Actively expand the application of six key vertical fields: "artificial intelligence + finance", "artificial intelligence + manufacturing", "artificial intelligence + education", "artificial intelligence + medical care", "artificial intelligence + cultural tourism" and "artificial intelligence + urban governance".

Accelerate the construction of an innovative application ecosystem: Strive to build a large-scale model innovation and development ecosystem, a high-level open source and open ecosystem, a collaborative development ecology of key industries, an industrial incubation and cultivation ecology, and an artificial intelligence testing and evaluation system, and strengthen the guarantee of ecological elements.

Since 2022, Shanghai has issued a number of documents to clearly support the construction of a smart computing city, from the target computing power of 15EFLOPS proposed in the "New Data Center "Computing Power Pujiang" Action Plan (2022-2024)" to the target computing power of 100EFLOPS in the "Implementation Plan for Artificial Intelligence "Molding Shencheng", which is about 5 times the original plan. The 100EFLOPS target means that Shanghai will usher in a qualitative change in technology. At present, most of the world's supercomputers are concentrated in high-performance computing, climate prediction, national defense and scientific research. The realization of the Shanghai plan will significantly enhance its competitiveness in these areas and set a benchmark for other cities and countries. With the implementation of the Plan, Shanghai is expected to attract innovative resources and outstanding enterprises from all over the world, forming an agglomeration effect of the computing industry, thereby accelerating the development of the computing industry in the entire Yangtze River Delta region.

AI traffic tracking: Character.AI +8.25% month-on-month visit duration in this period (2024.12.30-2025.1.3) AI-related website traffic data: the top three visits are ChatGPT (716.2M), Bing (414.4M) and Discord (144.7M), and Discord (5.85%) is the first month-on-month growth rate of visits; The top three average length of stay were Character.AI (00:14:39), Discord (00:12:12) and NotionAI (00:08:59). The average length of stay increased by Character.AI (8.25%) month-on-month.

AI Funding Trends: KoBold Metals topped AI funding this week, raising RMB 3.9 billion in Series C financing

　　　 KoBold Metals won the first AI funding amount this week, receiving RMB 3.9 billion in Series C financing. Founded in 2018 and headquartered in Berkeley, California, KoBold Metals Co is a developer of machine prospecting technology that facilitates the use of AI technology to explore locations for metals such as copper, lithium, and cobalt, and is committed to advancing a clean energy future by accelerating the discovery of critical metals. The company invests $100 million annually in more than 60 projects and R&D across three continents to advance the frontiers of exploration technology.

KoBold combines geoscience, systematic data integration, and artificial intelligence to accelerate the pace of exploration and reduce the cost of mineral exploration. Their geoscience leaders have been involved in the most significant discoveries of the past generation and have invested heavily in integrating all available information about the physics and chemistry of the Earth's crust. Their proprietary artificial intelligence (AI) tool, based on a concept they call Efficacy of Information (EOI), enables KoBold to determine what data to collect at each exploration step to minimise uncertainty. With a portfolio of projects in major mining sites around the world, KoBold is building the world's largest geoscience information base and using artificial intelligence to simulate subsurface conditions and explore depths beyond conventional technical capabilities. They currently deploy their know-how in more than 60 mineral exploration projects on three continents, in joint ventures and alliances with junior and major mining companies.

Investment Advice The

2025 Consumer Electronics Show (CES 2025) will be held in Las Vegas, USA from January 7 to 10, chip architecture design company Arm released a preview of CES 2025, believing that AI will continue to be the focus of this year's CES, at the same time, Arm focused on outlining several major trends including: autonomous driving innovation, the emergence of in-vehicle AI technology, accelerated automotive software development, AI-enabled smart home devices such as TVs, and Arm architecture With the strong momentum of PCs and laptops, the expansion of XR technology, and the rise of high-performance edge-side AI technology, we believe that the new products of AI devices and chip manufacturers are worth looking forward to.

We are firmly and continuously optimistic about AI software and hardware opportunities, and recommend paying attention to iFLYTEK (002230. SZ), Cambrian (688256. SH), high-speed communication connector business or significantly benefit from GB200 Dingtong Technology (688668. SH), Emdoor Information (001314. SZ) and so on.

Risk Warning:

1) The iteration speed of the underlying AI technology is not as fast as expected. 2) Policy supervision and copyright risks. 3) The implementation effect of AI applications is not as expected. 4) Recommend the risk that the company's performance is less than expected.