According to the Zhitong Finance APP, Guojin Securities released a research report stating that the domestic large model DeepSeek has significantly reduced the training and inference costs of contemporary AI large models by incorporating reinforcement learning during the pre-training phase. It currently ranks first in the mobile application download charts across 140 markets. Major cloud platforms such as Huawei Cloud, Tencent Cloud, Alibaba Cloud, and Baidu Intelligent Cloud have announced their integration of the DeepSeek series models. The demand for distributed and edge-side deployment of AI will increase the need for inter-regional network connectivity, leading to a rise in demand for hardware devices such as switches.
Major domestic chip manufacturers and application companies are accelerating their integration with DeepSeek. On February 5, Huawei announced that the new DeepSeek series models have officially launched in the Ascend community; previously, chip manufacturers such as Haiguang Information and Moore Threads also announced their adaptation and launch. The year 2025 is expected to be a year of large-scale demand for inference, as AI applications spread from centralized cloud services to massive terminals, requiring cost-effective dedicated chips to support this growth, which is favorable for domestic ASIC chip manufacturers.
While enhancing inference and edge computing power, the usage of AIDC is expected to rise simultaneously. Unlike centralized training, retail-type IDC suppliers are expected to be more aligned with the lightweight large model's capacity, suggesting a focus on domestic AIDC suppliers. DeepSeek is accelerating the deployment speed of AI models on the edge side, and high-performance IoT modules are essential hardware for end-side AI applications, with strong growth momentum.
<h2>Guojin Securities' main viewpoints are as follows:</h2>
Domestic AI Chips: DeepSeek is driving the spread of AI applications from centralized cloud services to massive terminals, requiring cost-effective dedicated chips. Currently, major domestic chip manufacturers and application companies are accelerating their integration with DeepSeek. In the context of intensified bans and large-scale computing power construction by domestic internet companies led by ByteDance, it is expected that domestic computing power will see large-scale growth by 2025. It is recommended to pay attention to manufacturers with ASIC chip R&amp;D capabilities and domestic computing power server OEMs such as ZTE Corporation (000063.SZ) and Unisplendour Corporation (000938.SZ).
IoT Modules: Guanghetong will launch the Fibocom AI Stack at the 2025 International Consumer Electronics Show; Quectel Wireless Solutions is collaborating with the Doubao large model team under Volcano Engine to develop companion AI toys and is actively investing in embodied intelligence directions such as robotic pets and robots; Meige Intelligent will innovatively launch AI agent products—AIMO at CES 2025. It is recommended to focus on the layout, cooperation with major manufacturers, and achievement transformation of IoT module manufacturers in the AI field.
IDC: DeepSeek's low pricing strategy and high-performance performance allow small and medium-sized enterprises and developers to develop AI applications at low costs. The usage of AIDC is expected to rise simultaneously, and retail-type IDC suppliers are expected to be more aligned with the lightweight large model's capacity. Runze Technology's main terminal customers are leading internet companies, cloud manufacturers, and AI clients, making it a leading comprehensive computing power center solution provider in China
Switches: The demand for distributed and edge-side deployment of AI will increase the demand for inter-regional network interconnection, leading to a rise in the demand for hardware devices such as switches. Companies capable of mass-producing 800G/1.6T switches/switching chips will benefit significantly in the domestic large model industry by 2025. At the same time, new types of switches such as CPO/OSC entering the commercial stage in the U.S. market will become a strategic high ground for the domestic switch industry.
Optical Modules: DCI is expected to usher in a new round of global resonance prosperity cycle. With the construction of the &#34;Stargate&#34; project, the demand for high-speed DCI optical modules is expected to increase significantly, benefiting manufacturers with relevant product layouts. Dekor continues to increase its R&amp;D efforts for DCI products, and the product gross margin still has room for improvement compared to expectations, with smooth customer expansion currently. DCI products are expected to become a new profit growth point for the company, likely benefiting fully from the construction of the &#34;Stargate&#34; project, and are recommended for attention. Core data update: Operator data maintains steady growth. In December, three operators achieved telecommunications main business revenue of 141.7 billion yuan, a year-on-year increase of 10.10% and a quarter-on-quarter increase of 0.35%; emerging business revenue reached 39.6 billion yuan, a year-on-year increase of 66.39%, with a quarter-on-quarter increase of 7.32%. In December, China's optical module export data continued to grow, with a year-on-year increase of 11.06%; the cumulative year-on-year increase from January to December was 58.24%.
Investment Recommendations and Valuation
It is recommended to pay attention to the domestic AI chip, IoT module, IDC, switch, and optical module sectors driven by DeepSeek's development, as well as the server and optical module sectors driven by overseas AI development.
Risk Warning
Risks include AI commercial value falling short of expectations, technology development speed not meeting expectations, excessive supply chain concentration, increased industry regulation, and intensified market competition

80700

TCTZF

TCEHY

00700

HUAWEI

DPSK

89988

BABA

09988

Guojin Securities released a research report stating that the domestic large model DeepSeek reduces AI model training and inference costs through reinforcement learning, promoting growth in domestic dedicated chips, AIDC, and the Internet of Things. Cloud platforms such as Huawei, Tencent, and Alibaba have already integrated DeepSeek, and it is expected to significantly increase inference demand by 2025, facilitating the development of domestic ASIC chip manufacturers. It is recommended to pay attention to companies with ASIC R&D capabilities and related suppliers

- Guojin Securities reports that the domestic AI model DeepSeek has significantly reduced training and inference costs through reinforcement learning, leading to its top position in mobile app downloads across 140 markets.  
- Major cloud platforms like Huawei Cloud and Tencent Cloud have integrated DeepSeek, driving demand for hardware like switches due to increased distributed and edge deployment needs.  
- The report suggests focusing on domestic AI chip manufacturers and IoT module firms, as 2025 is expected to see a surge in inference demand and the expansion of AI applications from centralized cloud services to numerous endpoints.

Zhitong

TENCENT-R

Tencent Holdings Limited

Tencent

TENCENT

Huawei

DeepSeek

BABA-WR

Alibaba

BABA-W

Range Technology

BIDU-SW

Quectel

MeiG

Fibocom

UNIS

SINOLINK SECURITIES

Guotai Junan Securities: DeepSeek drives the prosperity of domestic special chips, AIDC, Internet of Things and other sectors