financial sector February 18 news overseas AI accelerated again during the spring festival holiday!

Overseas AI accelerates again

On February 9, OpenAI CEO Sam Altman said that he was in talks with potential investors, including the UAE government, to increase global chip production capacity and expand its ability to support AI performance. The project may need to raise up to US $5 trillion to US $7 trillion. Altman2 spoke in a tweet on the X platform on the 8th: "We believe that the world's demand for AI infrastructure, including chip manufacturing capacity, energy supply, data centers, etc., has far exceeded the current construction plan."

On February 9, NVIDIA announced the creation of a new division focused on designing custom chips for companies such as cloud computing. The new NVIDIA division will be led by Dina McKinney. McKinney, formerly vice president of AMD CPU design and Marvell infrastructure processors, will oversee teams building custom chips for cloud computing, 5G telecommunications, gaming, automotive and more.

On February 9, Microsoft is developing a new version of the Copilot, code-named Deucalion, to provide richer and more powerful functions for employees' office needs. Microsoft has promoted the popularity of Microsoft 365 Copilot among employee teams.

On February 16, OpenAI released the Sora large model, which can generate 1-minute high-definition video according to user instructions, and can generate complex scenes with multiple characters and specific movements, that is, it can understand and simulate the physical world in motion. OpenAI has invited a team of professional creatives to test it for feedback on its usefulness in a professional environment.

on February 16, Google announced the launch of a new Gemini 1.5 AI model, which adopts MOE architecture and can handle 128000 token standard context windows.

Sora scares the film and television industry?

in the early morning of February 16, Beijing time, without any notice, OpenAI, a global star AI start-up company, released the Wen sheng video model Sora. for the first time, AI generated a one-minute long multi-lens video. its high simulation degree picture, fine picture quality, multi-lens shooting and multi-angle mirror transportation for the real human world show that AI's understanding of the human world and the creative content generated by AI have reached a new level.

Guotai Junan Research News pointed out that Sora has three outstanding highlights. One is a 60-second video. Sora can maintain a high degree of fluency and stability between the main body and background of the video. The second is a single video multi-angle lens. Sora realizes multi-angle lens in one video, and the split mirror switching is logical and smooth. The third is the ability to understand the real world. Sora handles the details such as light and shadow reflection, movement mode and lens movement very well, which greatly improves the sense of reality.

compared with the current AI video track counterparts, Sora's video length of 60 seconds per prompt is much higher than Pika Labs's 3 seconds, Meta Emu Video's 4 seconds and Runway's Gen-2's 18 seconds.

on February 16, 360 founder Zhou hongyi posted a microblog mentioning his views on Sora. Zhou hongyi believed that the birth of Sora meant that the realization of AGI (general artificial intelligence) could be shortened from 10 years to one or two years.

Zhou Hongyi believes that the final competition in science and technology is to make talent density and deep accumulation, "Many people say that Sora's effect hangs Pika and Runway. This is normal, and the entrepreneurial team than, OpenAl this core technology of the company's strength is still very strong. Some people think that with AI after the start-up companies only need to do self-employed on the line, the actual today once again proved that this idea is very ridiculous."

Zhu Zhili, a young director from Hong Kong, China, told Blue Whale Finance, "Its (Sora) impact on the film industry is only a matter of time, because it has made the picture very real and detailed, including the picture of a woman on the streets of Tokyo, even the freckles on her face can be very real."

Zhou Hongyi believes that machines can produce a good video, but the theme, script, split-shot planning and line coordination of the video all require human creativity, at least human prompt words. A video or movie is made up of an infinite number of 60 seconds. Today Sora could be a huge disruption to the advertising, movie trailers, and short video industries, but it won't necessarily beat TikTok so quickly and is more likely to become a creative tool for TikTok.

AI infrastructure in demand

The market view is that 2022 is the year of the image, 2023 is the year of the sound wave, and 2024 is the year of the video. OpenAI says Sora is the foundation on which the world model is built and will continue to move towards AGI in the future.

CITIC Securities said that the breakthrough of multi-modal large model algorithm will bring revolutionary progress in autonomous driving, robotics and other technologies, continue to be optimistic about the long-term impact and changes of the current wave of generative AI on the technology industry, and continue to focus on leading manufacturers in computing power, algorithms, data, applications and other links.

Soochow Securities judged that multimodality is the starting point of AI business vision, which is expected to really reduce costs and increase efficiency for enterprises, and enterprises can use the saved costs to improve product, service quality or technological innovation, and promote further productivity improvement; at the same time, there may also be new user-generated content platforms with more space.

For the development of Sora, there is a strong demand for computing power. Guotai Junan pointed out that the Sora model promotes the rapid development of AI multimodal fields, AI creation and other related fields will usher in in-depth changes, AI enabling scope will be further expanded, and multimodal related training and reasoning applications will further enhance the relevant demand for computing power infrastructure.

coincidentally, Guosheng securities also holds the same view. it believes that Sora still conforms to the AI scale law (Scaling Law). OpenAI explained in the technical document that with the increase of training calculation amount, the sample quality is obviously improved, further proving that in the multimodal era, the demand for computing power will become one of the core bottlenecks.

multi-modal large model to drive the rapid growth of global computing power demand, domestic AI computing power ushered in the opportunity. According to the statistics of the Southern Wealth Network Trend Stock Selection System, there are currently 52 listed enterprises related to A- share domestic AI computing power. For example, the domestic AI computing power industry chain includes AI server components, server machines, computing power leasing, data centers and other links. AI server parts companies mainly include Haiguang Information, Cambrian, Godson Zhongke, Jingjiawei, etc. Server companies mainly include Hi-Tech Development, Digital China, Tuowei Information, Radio and Television Express, Fiberhome Communications, Tongfang Shares, etc. The computing power leasing companies mainly include Hengrun, Yunsai Zhilian, Hongbo, etc. Data center companies mainly include Aofei Data, Guanghuan New Network, Baoxin Software, Data Port, etc.

Tianfeng Securities released a research report saying that compared with overseas, it is optimistic that large domestic enterprises will use large models to empower their applications in depth, and it is also optimistic that the computing power will continue to grow at a high rate in the future. It is suggested to pay attention to the three main lines of AI multimodality, AI application and Huawei chain.

(1)AI multimodality: wanxing technology (300624.SZ), meitu company (01357)(jointly covered with overseas groups), easy point world (301171.SZ), focus technology (002315.SZ), danghong technology (688039.SH);

(2)AI application: jinshan office (688111.SH), hkust xunfei (002230.SZ), hang seng electronics (600570.SH), dingjie software (300378.SZ), fuxin software (688095.SH), ufida network (600588.SH), kingdee international (00268), pan-to-micro network (603039.89. <

(3) huawei chain : sea light information (688041.SH), cambrian (688256.SH), yuntianlifei (688343.SH), jingjiawei (300474.SZ) (electronic group joint coverage). (Integrated Investor Network, Blue Whale Finance, Brokerage Research Report)