2 February 19, the first trading day after the Spring Festival of the Year of the Dragon for A-shares, boosted by the openai Wensheng video model sora, A-share AI application, CPO and computing power stocks surged.
wind ai application sector increased
Among them, wind ai application sector, Chinese Online (300364), Xuanya International (300612), Yidiantianxia (301171), Danghong Technology (688039), Haitian Ruisheng (688787), Yinsai Group (300781) and Wanxing Technology (300624), all seven stocks reached daily limit of 20cm, and the entire sector rose by 7.12% throughout the day.
wind The emergence of cpo optical module increase
sora is also a "miracle with great force", which makes the previous estimate of computing power a bit conservative. The AI computing power server cluster requires a large number of CPO optical modules for communication, and its demand is deterministic.
html On February 19, the three CPO giants Zhongji InnoLight (300308), Xinyi Sheng (300502), and Tianfu Communications (300394) collectively hit the daily limit of 20%, and Tianfu Communications reached a record high. The wind cpo optical module sector rose 14.08% throughout the day.
wind Chip Index Increase
wind Chip Index Sector, Cambrian (688256) rose by 16.14%, Haiguang Information (688041) rose by 12.10%, Huatian Technology (002185) rose by 10.01%, Sugon (603019) rose by 10%, the entire sector The increase throughout the day was 3.03%.
html On February 16, openai released the video generation model sora. This model is a text-generated video model. The model can generate high-fidelity videos of up to 60 seconds based on text information, which surpasses previous video generation models and is a milestone.According to reports, sora has strong video generation performance and can generate complex scene videos with multiple characters and specific motion trajectories, and make the physical interactions of the characters in the video conform to the physical laws of the real world. OpenAI positions sora as a video generation model of a world simulator, laying the foundation for future development of models that understand and simulate the real world.
Debon Securities Research Report pointed out that computing power support and large-scale high-quality data training are still the fundamentals of large-scale video models. According to openai, the capabilities demonstrated by sora show that continuing to expand the scale of video models can more effectively simulate humans, animals and objects in the physical and digital world. With the further improvement of video model duration, resolution and other parameters, it will Put forward higher requirements for computing power. In addition, Debon Securities believes that Sora may already have a commercial technology foundation, and the commercialization of AI video is no longer far away.
Debon Securities recommends focusing on two directions: 1) Large models and applications: iFlytek, Yuncong Technology, Wanxing Technology, ArcSoft Technology, Focus Technology, Danghong Technology (688039), etc. 2) Computing power and servers: Haiguang Information, Cambrian, Inspur Information, Sugon, Digital China, Hi-tech Development, etc.
Generally speaking, the emergence of sora video model has put forward further demand for computing power, which is good for computing power chips and cpo optical modules; in addition, the emergence of sora has subverted the previous video model and promoted the further maturity and maturity of AI applications. commercialize. This is highly consistent with the direction of funds in the A-share market on February 19, with AI applications, computing power and CPO sectors all seeing sharp increases.
The Paper reporter also called a number of AI application A-share listed companies to inquire about the impact of the Sora video model on the company. Many listed companies also responded to investors’ questions on the interactive platform.
Hikvision and Dahua Technology: The direction of sora is opposite
Hikvision (002415) and Dahua Technology (002236) both have plans in the field of multi-modal large models, but the two companies All solutions are based on ready-made videos to provide analysis and analysis to help customers determine what happened in the video, instead of Sora generating videos through text.
html On February 19, a reporter called Hikvision as an investor. Relevant people said that the company’s products are not artificial intelligence in the same category as Sora. The company focuses on perceptual intelligence. The company’s large video model is mainly used in the smart IoT industry to reduce costs and increase efficiency for enterprises. “For example, if you shoot a video and then judge whether something happened in the video, there are clear rules. , to identify a certain thing according to regulations. Sora is generative, which is actually the opposite of our direction."A relevant person from Dahua Co., Ltd. said that the company has the capability of large-scale video models and is currently doing research and development in two industries. The difference is that sora is generative, while Dahua is analytical, and the direction is exactly the opposite. Coming here, Dahua mainly uses existing materials or customer content to analyze, and then tells customers what happened in the video, and makes behavioral judgments based on what happened.
2 At the close of trading on February 19, Hikvision increased by 0.45 %, Dahua shares rose by 3.54%.
Danghong Technology and Yinsai Group: Making generative video products
html On February 19, Danghong Technology (688039) hit the 20cm limit. As for whether Danghong Technology has products similar to sora, A reporter from The Paper called Danghong Technology as an investor. A person from the company told The Paper: "It is being done, mainly for b-end customers. The duration and effect of sora are good, and our duration will definitely not reach one minute. , and are gradually making some breakthroughs. Danghong Technology itself is a video company, and we have some technical reserves in video-related generation, compression, editing, and recognition. Overall, it is pretty good. "Danghong Technology is an intelligent video solution and video cloud service provider. On January 5, Danghong Technology stated on the interactive platform that the company has a self-developed aigc tool set and released a solution for generating three-dimensional volumetric videos from static photos. , and achieves up to 800 times visual lossless compression through point cloud model conversion and compression algorithms, enabling switching between different modes. The company's aigc video content intelligent generation system integrates large model technology to help users in news dissemination and video content Create high-quality video content in many fields such as creation.
html On February 19, Yinsai Group (300781) reached the 20cm daily limit. Regarding whether Yinsai Group has products similar to sora, the reporter called Yinsai Group as an investor and the company stated : "Investors have also asked relevant questions on the Huayi platform. We are sorting out the responses and investors can pay attention. Regarding the progress of insightgpt, you can see the investor relations activity record table previously released by the company. "Yinsai Group's January investor relations activity record sheet disclosed that the internal beta version of insightgpt has recently opened the video intelligent editing function for partners, investment institutions, securities analysts, etc. to try out. It is expected that before the end of February 2024, Open to experience the marketing application products related to Tusheng Video, it is expected that the function of Tusheng Video will be developed and implemented before the end of March 2024, and then a public beta version will be launched to officially start commercialization.
Bohui Technology, Yidiantianxia, and Hanwang Technology: have technology accumulation and reserves
Bohui Technology (688004) also hit the daily limit of 20cm on the 19th. The reporter called Bohui Technology as an investor, and relevant people also said that the company is also actively paying attention to this aspect of content information. At present, the main business of the company is The business is still in some aspects such as media security, smart education and intelligent display and control.
This person explained that the media security business is also one of the company's core businesses. It currently provides video operation and maintenance and related content supervision. In terms of supervision, there are some relevant supervision models, which can conduct a dynamic analysis and processing of relevant text information to achieve the purpose of smart supervision.
Yidiantianxia (301171) also has a 20cm daily limit. The Paper reporter called as an investor Relevant people at Yidiantianxia said that they are currently using AIGC technology to provide customers with some creative material generation services. “We hope to benefit from the new model of sora to reduce costs and increase efficiency. The company's AIGC creation platform kreadoai can help enterprises realize an AI-based full-link closed loop of content production from script writing, voice cloning, personalized digital human selection to output of spoken videos. "
Hanwang Technology (002362) hit the 10% limit on February 19. The reporter called Hanwang Technology as an investor. Relevant people said that Hanwang has been working hard in the field of artificial intelligence and has its own core technologies, such as text recognition, image recognition, video Analysis: Hanwang also has its own core technology for this multi-modal recognition technology, and it has been applied in many fields. In the direction of recognition and analysis technology, the company has its own large model, which is mainly used in vertical fields.
Jiadu Technology (600728) said that the company does not have products similar to sora: "No, the company's transcorc-m multi-modal large model is a large model developed for the early stage of our own business. Everyone is very excited about the new concept, but I didn't listen. Who in China can realize such a function (Vinsheng Video)."
Jiadu Technology stated on the interactive platform on December 11, 2023 that the company's transcorc-m multi-modal large model can understand various modes such as text, images, and videos. It can capture dynamic information and describe, prompt or convert it into command codes according to user needs. It is a very "smart" traffic information assistant.
Relevant people from Yuncong Technology (688327) told The Paper that Yuncong Technology has a relatively deep accumulation in visual aspects. At present, Congrong Large Model has realized modeling around character images and character video data using generative technologies such as diffusion and gan. , realize the generation, control and editing of image and video content through text and audio. At present, Yuncong's technology is mainly used in multi-modal bases and some applications for specific customers, and is not targeted at c-end customers.
